build(ui): update dashboard version to 651732e

- [x] Update dashboard version in package files
- [x] Enhance dashboard description with analytics and metrics
- [x] Reorganize "Under the hood" section to 3-column layout
- [x] Update sqlite-vec description with hybrid storage details
- [x] Add Tree-sitter for AST-aware code chunking
- [x] Add CSR Graph for observation relationships
- [x] Expand features with graph-based search and AST-aware chunking
- [x] Add hybrid vector storage feature description
- [x] Add auto-tuning feature for dynamic optimization
- [x] Add vector storage and graph configuration options
- [x] Update FAQ with hybrid storage, AST chunking, and dashboard metrics

.github/workflows/autoupdate.yaml

@@ -17,4 +17,5 @@ jobs:
     with:
       go-version: ">=1.24"
       release-workflow: "release.yaml"
+      lfs: true
     secrets: inherit

.github/workflows/release.yaml

@@ -1,120 +1,121 @@
 name: Release
 
 on:
-  push:
-    branches:
-      - main
-    paths-ignore:
-      - "**.md"
-      - "docs/**"
-  workflow_dispatch:
+    push:
+        branches:
+            - main
+        paths-ignore:
+            - "**.md"
+            - "docs/**"
+    workflow_dispatch:
 
 permissions:
-  id-token: write
-  contents: write
-  packages: write
+    id-token: write
+    contents: write
+    packages: write
 
 jobs:
-  release:
-    uses: lukaszraczylo/shared-actions/.github/workflows/go-release-cgo.yaml@main
-    with:
-      go-version: ">=1.24"
-      node-enabled: true
-      node-version: "20"
-      node-build-script: "cd ui && npm ci && npm run build"
-      node-cache-dependency-path: "ui/package-lock.json"
-      node-output-path: "ui/dist"
-      node-embed-path: "internal/worker/static"
-    secrets: inherit
-
-  commit-marketplace:
-    needs: release
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
+    release:
+        uses: lukaszraczylo/shared-actions/.github/workflows/go-release-cgo.yaml@main
         with:
-          ref: main
+            go-version: ">=1.24"
+            node-enabled: true
+            node-version: "20"
+            node-build-script: "cd ui && npm ci && npm run build"
+            node-cache-dependency-path: "ui/package-lock.json"
+            node-output-path: "ui/dist"
+            node-embed-path: "internal/worker/static"
+            lfs: true
+        secrets: inherit
 
-      - name: Download checksums
-        run: |
-          VERSION="${{ needs.release.outputs.version_tag }}"
-          curl -sSL -o checksums.txt \
-            "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/${VERSION}/checksums.txt"
+    commit-marketplace:
+        needs: release
+        runs-on: ubuntu-latest
+        steps:
+            - uses: actions/checkout@v4
+              with:
+                  ref: main
 
-      - name: Generate marketplace.json with checksums
-        run: |
-          VERSION="${{ needs.release.outputs.version }}"
-          VERSION_TAG="${{ needs.release.outputs.version_tag }}"
-          TODAY=$(date -u +%Y-%m-%d)
+            - name: Download checksums
+              run: |
+                  VERSION="${{ needs.release.outputs.version_tag }}"
+                  curl -sSL -o checksums.txt \
+                    "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/${VERSION}/checksums.txt"
 
-          # Extract checksums from checksums.txt
-          SHA_DARWIN_ARM64=$(grep "darwin_arm64" checksums.txt | awk '{print $1}')
-          SHA_LINUX_AMD64=$(grep "linux_amd64" checksums.txt | awk '{print $1}')
-          SHA_WINDOWS_AMD64=$(grep "windows_amd64" checksums.txt | awk '{print $1}')
+            - name: Generate marketplace.json with checksums
+              run: |
+                  VERSION="${{ needs.release.outputs.version }}"
+                  VERSION_TAG="${{ needs.release.outputs.version_tag }}"
+                  TODAY=$(date -u +%Y-%m-%d)
 
-          # Generate marketplace.json
-          cat > marketplace.json << EOF
-          {
-            "\$schema": "https://anthropic.com/claude-code/marketplace.schema.json",
-            "name": "claude-mnemonic",
-            "version": "1.0.0",
-            "description": "Persistent memory system for Claude Code - stores observations, session summaries, and user prompts with semantic search",
-            "owner": {
-              "name": "lukaszraczylo",
-              "email": "[email protected]",
-              "url": "https://github.com/lukaszraczylo"
-            },
-            "repository": "https://github.com/lukaszraczylo/claude-mnemonic",
-            "plugins": [
-              {
-                "name": "claude-mnemonic",
-                "description": "Persistent memory system for Claude Code - Go implementation with SQLite and ChromaDB vector search",
-                "version": "${VERSION}",
-                "author": {
-                  "name": "lukaszraczylo",
-                  "url": "https://github.com/lukaszraczylo"
-                },
-                "category": "productivity",
-                "tags": ["memory", "persistence", "search", "context"],
-                "license": "MIT",
-                "homepage": "https://github.com/lukaszraczylo/claude-mnemonic",
-                "releases": {
-                  "latest": "${VERSION}",
-                  "versions": {
-                    "${VERSION}": {
-                      "releaseDate": "${TODAY}",
-                      "downloads": {
-                        "darwin-arm64": {
-                          "url": "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/${VERSION_TAG}/claude-mnemonic_${VERSION}_darwin_arm64.tar.gz",
-                          "format": "tar.gz",
-                          "sha256": "${SHA_DARWIN_ARM64}"
+                  # Extract checksums from checksums.txt
+                  SHA_DARWIN_ARM64=$(grep "darwin_arm64" checksums.txt | awk '{print $1}')
+                  SHA_LINUX_AMD64=$(grep "linux_amd64" checksums.txt | awk '{print $1}')
+                  SHA_WINDOWS_AMD64=$(grep "windows_amd64" checksums.txt | awk '{print $1}')
+
+                  # Generate marketplace.json
+                  cat > marketplace.json << EOF
+                  {
+                    "\$schema": "https://anthropic.com/claude-code/marketplace.schema.json",
+                    "name": "claude-mnemonic",
+                    "version": "1.0.0",
+                    "description": "Persistent memory system for Claude Code - stores observations, session summaries, and user prompts with semantic search",
+                    "owner": {
+                      "name": "lukaszraczylo",
+                      "email": "[email protected]",
+                      "url": "https://github.com/lukaszraczylo"
+                    },
+                    "repository": "https://github.com/lukaszraczylo/claude-mnemonic",
+                    "plugins": [
+                      {
+                        "name": "claude-mnemonic",
+                        "description": "Persistent memory system for Claude Code - Go implementation with SQLite and ChromaDB vector search",
+                        "version": "${VERSION}",
+                        "author": {
+                          "name": "lukaszraczylo",
+                          "url": "https://github.com/lukaszraczylo"
                         },
-                        "linux-amd64": {
-                          "url": "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/${VERSION_TAG}/claude-mnemonic_${VERSION}_linux_amd64.tar.gz",
-                          "format": "tar.gz",
-                          "sha256": "${SHA_LINUX_AMD64}"
-                        },
-                        "windows-amd64": {
-                          "url": "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/${VERSION_TAG}/claude-mnemonic_${VERSION}_windows_amd64.zip",
-                          "format": "zip",
-                          "sha256": "${SHA_WINDOWS_AMD64}"
+                        "category": "productivity",
+                        "tags": ["memory", "persistence", "search", "context"],
+                        "license": "MIT",
+                        "homepage": "https://github.com/lukaszraczylo/claude-mnemonic",
+                        "releases": {
+                          "latest": "${VERSION}",
+                          "versions": {
+                            "${VERSION}": {
+                              "releaseDate": "${TODAY}",
+                              "downloads": {
+                                "darwin-arm64": {
+                                  "url": "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/${VERSION_TAG}/claude-mnemonic_${VERSION}_darwin_arm64.tar.gz",
+                                  "format": "tar.gz",
+                                  "sha256": "${SHA_DARWIN_ARM64}"
+                                },
+                                "linux-amd64": {
+                                  "url": "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/${VERSION_TAG}/claude-mnemonic_${VERSION}_linux_amd64.tar.gz",
+                                  "format": "tar.gz",
+                                  "sha256": "${SHA_LINUX_AMD64}"
+                                },
+                                "windows-amd64": {
+                                  "url": "https://github.com/lukaszraczylo/claude-mnemonic/releases/download/${VERSION_TAG}/claude-mnemonic_${VERSION}_windows_amd64.zip",
+                                  "format": "zip",
+                                  "sha256": "${SHA_WINDOWS_AMD64}"
+                                }
+                              }
+                            }
+                          }
                         }
                       }
-                    }
+                    ]
                   }
-                }
-              }
-            ]
-          }
-          EOF
+                  EOF
 
-          echo "Generated marketplace.json:"
-          cat marketplace.json
+                  echo "Generated marketplace.json:"
+                  cat marketplace.json
 
-      - name: Commit and push
-        run: |
-          git config user.name "github-actions[bot]"
-          git config user.email "github-actions[bot]@users.noreply.github.com"
-          git add marketplace.json
-          git diff --staged --quiet || git commit -m "chore: update marketplace for ${{ needs.release.outputs.version_tag }}"
-          git push
+            - name: Commit and push
+              run: |
+                  git config user.name "github-actions[bot]"
+                  git config user.email "github-actions[bot]@users.noreply.github.com"
+                  git add marketplace.json
+                  git diff --staged --quiet || git commit -m "chore: update marketplace for ${{ needs.release.outputs.version_tag }}"
+                  git push

.golangci.yml

@@ -0,0 +1,35 @@
+linters-settings:
+  govet:
+    enable:
+      - fieldalignment
+  errcheck:
+    # Ignore error checks in test files for common test helpers
+    exclude-functions:
+      - (io.Closer).Close
+      - (*encoding/json.Encoder).Encode
+      - (io.Writer).Write
+
+linters:
+  enable:
+    - errcheck
+    - gosec
+    - govet
+    - gofmt
+    - staticcheck
+    - unused
+    - ineffassign
+    - typecheck
+
+issues:
+  exclude-dirs:
+    - vendor
+  # Exclude some linters from running on test files
+  exclude-rules:
+    - path: _test\.go
+      linters:
+        - errcheck
+        - gosec
+
+run:
+  timeout: 5m
+  tests: true

README.md

@@ -12,7 +12,21 @@ Claude Code forgets everything when your session ends. Claude Mnemonic fixes tha
 
 It captures what Claude learns during your coding sessions - bug fixes, architecture decisions, patterns that work - and brings that knowledge back in future conversations. No more re-explaining your codebase.
 
-## What's New in v0.6
+![Claude Mnemonic Dashboard](docs/public/claude-mnemonic.jpg)
+
+## What's New in v0.7
+
+- **Two-Stage Retrieval** - Cross-encoder reranking for dramatically improved search relevance
+- **Knowledge Graph** - Automatic relationship detection between observations with visual graph in dashboard
+  ![Knowledge Graph](docs/public/observation-relation-graph.jpg)
+- **Pattern Detection** - Identifies recurring patterns across sessions and projects
+- **Importance Scoring** - Time decay and voting system to surface the most valuable memories
+- **Query Expansion** - Reformulates searches to find semantically related content
+- **Conflict Detection** - Identifies and resolves contradictory observations
+- **Observation Lifecycle** - Memories can be superseded when better information arrives
+
+<details>
+<summary>Previous: v0.6</summary>
 
 - **Auto-Updates** - Automatically stays up-to-date with the latest version
 - **Slash Command: `/restart`** - Restart the worker directly from Claude Code
@@ -20,6 +34,7 @@ It captures what Claude learns during your coding sessions - bug fixes, architec
 - **Async Queue Processing** - Non-blocking observation capture for faster sessions
 - **Smarter Storage** - Filters out system/agent summaries to keep knowledge relevant
 - **Improved Reliability** - Better handling of connectivity issues and dead connections
+</details>
 
 ## Requirements
 
@@ -110,16 +125,39 @@ Config file: `~/.claude-mnemonic/settings.json`
 {
   "CLAUDE_MNEMONIC_WORKER_PORT": 37777,
   "CLAUDE_MNEMONIC_CONTEXT_OBSERVATIONS": 100,
-  "CLAUDE_MNEMONIC_CONTEXT_FULL_COUNT": 25
+  "CLAUDE_MNEMONIC_CONTEXT_FULL_COUNT": 25,
+  "CLAUDE_MNEMONIC_RERANKING_ENABLED": true
 }
 ```
 
+### Core Settings
+
 | Variable | Default | What it does |
 |----------|---------|--------------|
 | `WORKER_PORT` | `37777` | Dashboard & API port |
 | `CONTEXT_OBSERVATIONS` | `100` | Max memories per session |
-| `CONTEXT_FULL_COUNT` | `25` | Full detail memories (rest are condensed to title only) |
+| `CONTEXT_FULL_COUNT` | `25` | Full detail memories (rest are condensed) |
 | `CONTEXT_SESSION_COUNT` | `10` | Recent sessions to reference |
+| `CONTEXT_RELEVANCE_THRESHOLD` | `0.3` | Minimum similarity score (0.0-1.0) for inclusion |
+| `CONTEXT_MAX_PROMPT_RESULTS` | `10` | Max results per prompt search |
+
+### Reranking Settings (Two-Stage Retrieval)
+
+| Variable | Default | What it does |
+|----------|---------|--------------|
+| `RERANKING_ENABLED` | `true` | Enable cross-encoder reranking |
+| `RERANKING_CANDIDATES` | `100` | Candidates to retrieve before reranking |
+| `RERANKING_RESULTS` | `10` | Final results after reranking |
+| `RERANKING_ALPHA` | `0.7` | Score blend: alpha×rerank + (1-alpha)×original |
+| `RERANKING_PURE_MODE` | `false` | Use pure cross-encoder scores only |
+
+### Embedding Settings
+
+| Variable | Default | What it does |
+|----------|---------|--------------|
+| `EMBEDDING_MODEL` | `bge-v1.5` | Embedding model for semantic search |
+
+All variables are prefixed with `CLAUDE_MNEMONIC_` in the config file.
 
 ## Project vs Global scope
 
@@ -202,10 +240,11 @@ curl -sSL https://raw.githubusercontent.com/lukaszraczylo/claude-mnemonic/main/s
 
 - **SQLite + FTS5** - Full-text search for exact matches
 - **sqlite-vec** - Vector database embedded in SQLite
-- **all-MiniLM-L6-v2** - Local embedding model (384 dimensions) via ONNX Runtime
+- **Two-Stage Retrieval** - Bi-encoder (embedding) + cross-encoder (reranking) for high accuracy
+- **Local Models** - all-MiniLM-L6-v2 for embeddings, BGE reranker for relevance scoring
 - **Go** - Single binary, no external dependencies
 
-Everything runs locally. No Python. No external vector database. No API calls for embeddings.
+Everything runs locally. No Python. No external vector database. No API calls.
 
 ## Platform support
 

cmd/hooks/session-start/main.go

@@ -23,12 +23,12 @@ type Input struct {
 
 // Observation represents an observation from the API.
 type Observation struct {
-	ID        int64    `json:"id"`
 	Type      string   `json:"type"`
 	Title     string   `json:"title"`
 	Subtitle  string   `json:"subtitle"`
 	Narrative string   `json:"narrative"`
 	Facts     []string `json:"facts"`
+	ID        int64    `json:"id"`
 }
 
 func main() {
@@ -46,8 +46,8 @@ func main() {
 	}
 
 	var input Input
-	if err := json.Unmarshal(inputData, &input); err != nil {
-		hooks.WriteError("SessionStart", err)
+	if unmarshalErr := json.Unmarshal(inputData, &input); unmarshalErr != nil {
+		hooks.WriteError("SessionStart", unmarshalErr)
 		os.Exit(1)
 	}
 

cmd/hooks/statusline/main.go

@@ -44,21 +44,21 @@ type StatusInput struct {
 
 // WorkerStats is the response from the worker's /api/stats endpoint.
 type WorkerStats struct {
-	Uptime              string `json:"uptime"`
-	ActiveSessions      int    `json:"activeSessions"`
-	QueueDepth          int    `json:"queueDepth"`
-	IsProcessing        bool   `json:"isProcessing"`
-	ConnectedClients    int    `json:"connectedClients"`
-	SessionsToday       int    `json:"sessionsToday"`
-	Ready               bool   `json:"ready"`
-	Project             string `json:"project,omitempty"`
-	ProjectObservations int    `json:"projectObservations,omitempty"`
-	Retrieval           struct {
+	Uptime    string `json:"uptime"`
+	Project   string `json:"project,omitempty"`
+	Retrieval struct {
 		TotalRequests      int64 `json:"TotalRequests"`
 		ObservationsServed int64 `json:"ObservationsServed"`
 		SearchRequests     int64 `json:"SearchRequests"`
 		ContextInjections  int64 `json:"ContextInjections"`
 	} `json:"retrieval"`
+	ActiveSessions      int  `json:"activeSessions"`
+	QueueDepth          int  `json:"queueDepth"`
+	ConnectedClients    int  `json:"connectedClients"`
+	SessionsToday       int  `json:"sessionsToday"`
+	ProjectObservations int  `json:"projectObservations,omitempty"`
+	IsProcessing        bool `json:"isProcessing"`
+	Ready               bool `json:"ready"`
 }
 
 // ANSI color codes

cmd/hooks/stop/main.go

@@ -14,17 +14,17 @@ import (
 // Input is the hook input from Claude Code.
 type Input struct {
 	hooks.BaseInput
-	StopHookActive bool   `json:"stop_hook_active"`
 	TranscriptPath string `json:"transcript_path"`
+	StopHookActive bool   `json:"stop_hook_active"`
 }
 
 // TranscriptMessage represents a message in the transcript JSONL file.
 type TranscriptMessage struct {
-	Type    string `json:"type"`
 	Message struct {
+		Content any    `json:"content"`
 		Role    string `json:"role"`
-		Content any    `json:"content"` // Can be string or array
 	} `json:"message"`
+	Type string `json:"type"` // Can be string or array
 }
 
 // extractTextContent extracts text content from message content (handles both string and array formats).

cmd/hooks/user-prompt/main.go

@@ -35,8 +35,8 @@ func main() {
 	}
 
 	var input Input
-	if err := json.Unmarshal(inputData, &input); err != nil {
-		hooks.WriteError("UserPromptSubmit", err)
+	if unmarshalErr := json.Unmarshal(inputData, &input); unmarshalErr != nil {
+		hooks.WriteError("UserPromptSubmit", unmarshalErr)
 		os.Exit(1)
 	}
 
@@ -61,12 +61,8 @@ func main() {
 
 	searchResult, _ := hooks.GET(port, searchURL)
 	if observations, ok := searchResult["observations"].([]interface{}); ok && len(observations) > 0 {
-		// Limit to top 5 most relevant observations
-		maxObs := 5
-		if len(observations) < maxObs {
-			maxObs = len(observations)
-		}
-		observations = observations[:maxObs]
+		// Results are already filtered by relevance threshold and capped by max_results
+		// from the server-side config (ContextRelevanceThreshold, ContextMaxPromptResults)
 		observationCount = len(observations)
 
 		// Build context from search results

docs/src/App.vue

@@ -29,6 +29,21 @@
       subtitle="Stop re-explaining your codebase. Claude Mnemonic captures bug fixes, architecture decisions, and coding patterns - then brings them back exactly when you need them."
     />
 
+    <!-- Dashboard Preview -->
+    <section class="py-12 lg:py-16 px-4 sm:px-6 relative">
+      <div class="max-w-6xl mx-auto">
+        <div class="relative rounded-xl overflow-hidden border border-slate-700/50 shadow-2xl shadow-amber-500/5">
+          <div class="absolute inset-0 bg-gradient-to-t from-slate-950 via-transparent to-transparent pointer-events-none z-10"></div>
+          <img
+            src="/claude-mnemonic.jpg"
+            alt="Claude Mnemonic Dashboard"
+            class="w-full h-auto"
+          />
+        </div>
+        <p class="text-center text-slate-500 text-sm mt-4">The dashboard at localhost:37777 - browse, search, and manage your memories. View graph stats, vector metrics, storage savings, and performance analytics.</p>
+      </div>
+    </section>
+
     <!-- Problem Section -->
     <section class="py-20 lg:py-28 px-4 sm:px-6 relative">
       <div class="max-w-6xl mx-auto grid lg:grid-cols-2 gap-8 lg:gap-16 items-center">
@@ -85,6 +100,21 @@
       </div>
     </section>
 
+    <!-- Knowledge Graph Preview -->
+    <section class="py-16 lg:py-20 px-4 sm:px-6 relative">
+      <div class="max-w-5xl mx-auto">
+        <SectionHeader title="See how knowledge connects" subtitle="The knowledge graph reveals relationships between your memories automatically" />
+        <div class="relative rounded-xl overflow-hidden border border-slate-700/50 shadow-2xl shadow-purple-500/5">
+          <img
+            src="/observation-relation-graph.jpg"
+            alt="Knowledge Graph Visualization"
+            class="w-full h-auto"
+          />
+        </div>
+        <p class="text-center text-slate-500 text-sm mt-4">Click any observation to explore its relationships - see what causes what, what fixes what, and how concepts evolve</p>
+      </div>
+    </section>
+
     <!-- Before/After Section -->
     <section class="py-20 lg:py-28 px-4 sm:px-6">
       <div class="max-w-6xl mx-auto">
@@ -274,7 +304,7 @@
     <section class="py-20 lg:py-28 px-4 sm:px-6">
       <div class="max-w-6xl mx-auto">
         <SectionHeader title="Under the hood" subtitle="Built with simplicity and performance in mind" />
-        <div class="grid sm:grid-cols-2 lg:grid-cols-4 gap-4 sm:gap-6 text-center">
+        <div class="grid sm:grid-cols-2 lg:grid-cols-3 gap-4 sm:gap-6 text-center">
           <div class="glass rounded-2xl p-6 sm:p-8 hover:border-amber-500/30 transition-colors">
             <div class="text-3xl sm:text-4xl font-bold text-amber-500 mb-2">Go</div>
             <p class="text-slate-400 text-xs sm:text-sm">Single binary. Fast startup, low memory. Zero runtime dependencies.</p>
@@ -285,11 +315,19 @@
           </div>
           <div class="glass rounded-2xl p-6 sm:p-8 hover:border-amber-500/30 transition-colors">
             <div class="text-3xl sm:text-4xl font-bold text-amber-500 mb-2">sqlite-vec</div>
-            <p class="text-slate-400 text-xs sm:text-sm">Embedded vector database. No external services required.</p>
+            <p class="text-slate-400 text-xs sm:text-sm">Hybrid vector storage with LEANN-inspired selective embeddings. 60-80% storage reduction.</p>
           </div>
           <div class="glass rounded-2xl p-6 sm:p-8 hover:border-amber-500/30 transition-colors">
-            <div class="text-3xl sm:text-4xl font-bold text-amber-500 mb-2">MiniLM</div>
-            <p class="text-slate-400 text-xs sm:text-sm">Local embeddings via ONNX. "Fix auth" finds "JWT issue" automatically.</p>
+            <div class="text-3xl sm:text-4xl font-bold text-amber-500 mb-2">BGE</div>
+            <p class="text-slate-400 text-xs sm:text-sm">Two-stage retrieval: bi-encoder embeddings + cross-encoder reranking for high accuracy.</p>
+          </div>
+          <div class="glass rounded-2xl p-6 sm:p-8 hover:border-amber-500/30 transition-colors">
+            <div class="text-3xl sm:text-4xl font-bold text-amber-500 mb-2">Tree-sitter</div>
+            <p class="text-slate-400 text-xs sm:text-sm">AST-aware code chunking respects function boundaries for Go, Python, and TypeScript.</p>
+          </div>
+          <div class="glass rounded-2xl p-6 sm:p-8 hover:border-amber-500/30 transition-colors">
+            <div class="text-3xl sm:text-4xl font-bold text-amber-500 mb-2">CSR Graph</div>
+            <p class="text-slate-400 text-xs sm:text-sm">Memory-efficient observation relationship graph with edge detection and hub identification.</p>
           </div>
         </div>
       </div>
@@ -386,10 +424,13 @@ const activeTab = ref('macos')
 
 const features = [
   { icon: 'fas fa-brain', title: 'Learns as you work', description: 'Every bug fix, every architecture decision, every "aha moment" - captured automatically without breaking your flow.' },
+  { icon: 'fas fa-search', title: 'Two-stage retrieval', description: 'Cross-encoder reranking delivers highly relevant results. Finds what you need even with vague queries like "that auth thing".' },
+  { icon: 'fas fa-project-diagram', title: 'Graph-based search', description: 'LEANN Phase 2: Graph relationships between observations (file overlap, semantic similarity, temporal proximity) for smarter context retrieval.' },
+  { icon: 'fas fa-microchip', title: 'AST-aware chunking', description: 'Intelligent code splitting respects function boundaries. Go, Python, and TypeScript code is chunked at semantic boundaries, not arbitrary line counts.' },
+  { icon: 'fas fa-database', title: 'Hybrid vector storage', description: 'LEANN-inspired selective storage: frequently-accessed "hub" observations store embeddings, others recompute on-demand. 60-80% storage savings with <50ms latency.' },
   { icon: 'fas fa-folder-tree', title: 'Project-aware context', description: 'Your React knowledge stays in React projects. Your Go patterns stay in Go projects. No context pollution.' },
-  { icon: 'fas fa-globe', title: 'Shared best practices', description: 'Security patterns, performance tips, and universal learnings automatically available across all your projects.' },
-  { icon: 'fas fa-search', title: 'Finds what matters', description: 'Semantic search finds relevant memories even when you don\'t remember the exact words. "That auth thing" just works.' },
-  { icon: 'fas fa-chart-line', title: 'Live statusline', description: 'Real-time metrics right in Claude Code: memories served, searches performed, and project memory count at a glance.' },
+  { icon: 'fas fa-chart-line', title: 'Smart scoring', description: 'Importance decay, pattern detection, and conflict resolution ensure the most valuable memories surface first.' },
+  { icon: 'fas fa-gauge-high', title: 'Auto-tuning', description: 'Dynamic hub threshold adjustment based on query performance. Automatically balances storage efficiency with search latency for your workload.' },
   { icon: 'fas fa-lock', title: '100% private', description: 'Your code context never leaves your machine. No telemetry. No cloud sync. Your memories are yours.' },
 ]
 
@@ -415,8 +456,12 @@ const installCommands = {
 const configOptions = [
   { name: 'CLAUDE_MNEMONIC_WORKER_PORT', description: 'HTTP port for the worker service (default: 37777)', icon: 'fas fa-network-wired' },
   { name: 'CLAUDE_MNEMONIC_CONTEXT_OBSERVATIONS', description: 'Maximum observations injected per session (default: 100)', icon: 'fas fa-layer-group' },
-  { name: 'CLAUDE_MNEMONIC_CONTEXT_FULL_COUNT', description: 'Observations with full narrative detail, rest are condensed (default: 25)', icon: 'fas fa-expand' },
-  { name: 'CLAUDE_MNEMONIC_MODEL', description: 'Model for processing observations (default: haiku)', icon: 'fas fa-microchip' },
+  { name: 'CLAUDE_MNEMONIC_RERANKING_ENABLED', description: 'Enable cross-encoder reranking for improved search relevance (default: true)', icon: 'fas fa-sort-amount-down' },
+  { name: 'CLAUDE_MNEMONIC_CONTEXT_RELEVANCE_THRESHOLD', description: 'Minimum similarity score for inclusion, 0.0-1.0 (default: 0.3)', icon: 'fas fa-filter' },
+  { name: 'CLAUDE_MNEMONIC_VECTOR_STORAGE_STRATEGY', description: 'Storage strategy: "hub" (default), "always", or "on_demand"', icon: 'fas fa-database' },
+  { name: 'CLAUDE_MNEMONIC_GRAPH_ENABLED', description: 'Enable graph-based search with observation relationships (default: true)', icon: 'fas fa-project-diagram' },
+  { name: 'CLAUDE_MNEMONIC_GRAPH_MAX_HOPS', description: 'Maximum graph traversal depth for search expansion (default: 2)', icon: 'fas fa-route' },
+  { name: 'CLAUDE_MNEMONIC_GRAPH_REBUILD_INTERVAL_MIN', description: 'How often to rebuild the observation graph in minutes (default: 60)', icon: 'fas fa-clock' },
 ]
 
 const requiredDeps = [
@@ -427,9 +472,11 @@ const requiredDeps = [
 const faqs = [
   { question: 'Will it confuse Claude with wrong context?', answer: 'No. Mnemonic uses project isolation and semantic relevance scoring. Only memories from the current project (or global best practices) are injected, and only when they\'re actually relevant to your prompt.' },
   { question: 'What exactly gets saved?', answer: 'Bug fixes with context ("Fixed race condition by adding mutex"), architecture decisions ("Using repository pattern for data access"), conventions ("All API routes prefixed with /api/v1"), and learnings you want to preserve.' },
-  { question: 'Can I delete or edit memories?', answer: 'Yes. The web dashboard at localhost:37777 lets you browse, search, edit, and delete any memory. You\'re always in control.' },
+  { question: 'How does hybrid vector storage work?', answer: 'LEANN-inspired selective storage: frequently-accessed "hub" observations (identified by access patterns and graph centrality) store embeddings. Infrequently-accessed observations recompute embeddings on-demand during search. This reduces storage by 60-80% with minimal latency impact (<50ms).' },
+  { question: 'Can I delete or edit memories?', answer: 'Yes. The web dashboard at localhost:37777 lets you browse, search, edit, and delete any memory. You can also view graph relationships, storage metrics, and performance analytics. You\'re always in control.' },
   { question: 'Does it work with my existing Claude Code setup?', answer: 'Yes. Mnemonic installs as a Claude Code plugin with hooks. Your existing workflows, settings, and shortcuts remain unchanged.' },
   { question: 'What if I switch between projects frequently?', answer: 'That\'s the point. Each project has isolated memories. Switch from your Python ML project to your TypeScript app - context switches automatically.' },
-  { question: 'Is there a performance impact?', answer: 'Minimal. The Go worker is lightweight (typically under 30MB RAM). Context injection at session start takes milliseconds for most projects.' },
+  { question: 'Is there a performance impact?', answer: 'Minimal. The Go worker is lightweight (typically under 30MB RAM). Hybrid storage and auto-tuning optimize for your workload. Context injection at session start takes milliseconds for most projects.' },
+  { question: 'What is AST-aware chunking?', answer: 'When processing code observations, Mnemonic uses Tree-sitter parsers to respect function and class boundaries instead of arbitrary line limits. Go, Python, and TypeScript code is chunked at semantic boundaries for better search accuracy.' },
 ]
 </script>

go.mod

@@ -11,6 +11,7 @@ require (
 	github.com/goccy/go-json v0.10.5
 	github.com/mattn/go-sqlite3 v1.14.32
 	github.com/rs/zerolog v1.34.0
+	github.com/smacker/go-tree-sitter v0.0.0-20240827094217-dd81d9e9be82
 	github.com/stretchr/testify v1.11.1
 	github.com/sugarme/tokenizer v0.3.0
 	github.com/yalue/onnxruntime_go v1.25.0
@@ -26,6 +27,7 @@ require (
 	github.com/pmezard/go-difflib v1.0.0 // indirect
 	github.com/rivo/uniseg v0.4.7 // indirect
 	github.com/schollz/progressbar/v2 v2.15.0 // indirect
+	github.com/stretchr/objx v0.5.2 // indirect
 	github.com/sugarme/regexpset v0.0.0-20200920021344-4d4ec8eaf93c // indirect
 	golang.org/x/sys v0.39.0 // indirect
 	golang.org/x/text v0.32.0 // indirect

go.sum

@@ -38,7 +38,11 @@ github.com/rs/zerolog v1.34.0 h1:k43nTLIwcTVQAncfCw4KZ2VY6ukYoZaBPNOE8txlOeY=
 github.com/rs/zerolog v1.34.0/go.mod h1:bJsvje4Z08ROH4Nhs5iH600c3IkWhwp44iRc54W6wYQ=
 github.com/schollz/progressbar/v2 v2.15.0 h1:dVzHQ8fHRmtPjD3K10jT3Qgn/+H+92jhPrhmxIJfDz8=
 github.com/schollz/progressbar/v2 v2.15.0/go.mod h1:UdPq3prGkfQ7MOzZKlDRpYKcFqEMczbD7YmbPgpzKMI=
+github.com/smacker/go-tree-sitter v0.0.0-20240827094217-dd81d9e9be82 h1:6C8qej6f1bStuePVkLSFxoU22XBS165D3klxlzRg8F4=
+github.com/smacker/go-tree-sitter v0.0.0-20240827094217-dd81d9e9be82/go.mod h1:xe4pgH49k4SsmkQq5OT8abwhWmnzkhpgnXeekbx2efw=
 github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME=
+github.com/stretchr/objx v0.5.2 h1:xuMeJ0Sdp5ZMRXx/aWO6RZxdr3beISkG5/G/aIRr3pY=
+github.com/stretchr/objx v0.5.2/go.mod h1:FRsXN1f5AsAjCGJKqEizvkpNtU+EGNCLh3NxZ/8L+MA=
 github.com/stretchr/testify v1.3.0/go.mod h1:M5WIy9Dh21IEIfnGCwXGc5bZfKNJtfHm1UVUgZn+9EI=
 github.com/stretchr/testify v1.11.1 h1:7s2iGBzp5EwR7/aIZr8ao5+dra3wiQyKjjFuvgVKu7U=
 github.com/stretchr/testify v1.11.1/go.mod h1:wZwfW3scLgRK+23gO65QZefKpKQRnfz6sD981Nm4B6U=

internal/chunking/golang/chunker.go

@@ -0,0 +1,285 @@
+// Package golang provides AST-aware chunking for Go source files.
+package golang
+
+import (
+	"context"
+	"fmt"
+	"go/ast"
+	"go/parser"
+	"go/token"
+	"os"
+	"strings"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/chunking"
+)
+
+// Chunker implements AST-aware chunking for Go files.
+type Chunker struct {
+	options chunking.ChunkOptions
+}
+
+// NewChunker creates a new Go chunker.
+func NewChunker(options chunking.ChunkOptions) *Chunker {
+	return &Chunker{options: options}
+}
+
+// Language returns the language this chunker supports.
+func (c *Chunker) Language() chunking.Language {
+	return chunking.LanguageGo
+}
+
+// SupportedExtensions returns the file extensions this chunker handles.
+func (c *Chunker) SupportedExtensions() []string {
+	return []string{".go"}
+}
+
+// Chunk parses a Go source file and returns semantic code chunks.
+func (c *Chunker) Chunk(ctx context.Context, filePath string) ([]chunking.Chunk, error) {
+	// Read file content
+	content, err := os.ReadFile(filePath)
+	if err != nil {
+		return nil, fmt.Errorf("read file: %w", err)
+	}
+
+	// Parse the Go file
+	fset := token.NewFileSet()
+	file, err := parser.ParseFile(fset, filePath, content, parser.ParseComments)
+	if err != nil {
+		return nil, fmt.Errorf("parse Go file: %w", err)
+	}
+
+	chunks := make([]chunking.Chunk, 0)
+	sourceLines := strings.Split(string(content), "\n")
+
+	// Extract chunks from declarations
+	for _, decl := range file.Decls {
+		switch d := decl.(type) {
+		case *ast.FuncDecl:
+			chunk := c.extractFunction(fset, d, sourceLines, filePath)
+			if chunk != nil {
+				chunks = append(chunks, *chunk)
+			}
+		case *ast.GenDecl:
+			extracted := c.extractGenDecl(fset, d, sourceLines, filePath)
+			chunks = append(chunks, extracted...)
+		}
+	}
+
+	return chunks, nil
+}
+
+// extractFunction extracts a function or method declaration as a chunk.
+func (c *Chunker) extractFunction(fset *token.FileSet, fn *ast.FuncDecl, sourceLines []string, filePath string) *chunking.Chunk {
+	// Skip unexported if configured
+	if !c.options.IncludePrivate && !fn.Name.IsExported() {
+		return nil
+	}
+
+	startPos := fset.Position(fn.Pos())
+	endPos := fset.Position(fn.End())
+
+	chunk := &chunking.Chunk{
+		FilePath:  filePath,
+		Language:  chunking.LanguageGo,
+		Name:      fn.Name.Name,
+		StartLine: startPos.Line,
+		EndLine:   endPos.Line,
+	}
+
+	// Determine if this is a method or a function
+	if fn.Recv != nil && len(fn.Recv.List) > 0 {
+		chunk.Type = chunking.ChunkTypeMethod
+		chunk.ParentName = c.extractReceiverType(fn.Recv)
+	} else {
+		chunk.Type = chunking.ChunkTypeFunction
+	}
+
+	// Extract content
+	chunk.Content = c.extractLines(sourceLines, startPos.Line, endPos.Line)
+
+	// Extract signature (function declaration without body)
+	chunk.Signature = c.extractFunctionSignature(fn, fset, sourceLines)
+
+	// Extract doc comment
+	if c.options.IncludeDocComments && fn.Doc != nil {
+		chunk.DocComment = strings.TrimSpace(fn.Doc.Text())
+	}
+
+	return chunk
+}
+
+// extractGenDecl extracts general declarations (type, const, var).
+func (c *Chunker) extractGenDecl(fset *token.FileSet, gd *ast.GenDecl, sourceLines []string, filePath string) []chunking.Chunk {
+	var chunks []chunking.Chunk
+
+	for _, spec := range gd.Specs {
+		switch s := spec.(type) {
+		case *ast.TypeSpec:
+			chunk := c.extractTypeSpec(fset, gd, s, sourceLines, filePath)
+			if chunk != nil {
+				chunks = append(chunks, *chunk)
+			}
+		case *ast.ValueSpec:
+			// Handle const and var declarations
+			chunk := c.extractValueSpec(fset, gd, s, sourceLines, filePath)
+			if chunk != nil {
+				chunks = append(chunks, *chunk)
+			}
+		}
+	}
+
+	return chunks
+}
+
+// extractTypeSpec extracts a type declaration (struct, interface, type alias).
+func (c *Chunker) extractTypeSpec(fset *token.FileSet, gd *ast.GenDecl, ts *ast.TypeSpec, sourceLines []string, filePath string) *chunking.Chunk {
+	// Skip unexported if configured
+	if !c.options.IncludePrivate && !ts.Name.IsExported() {
+		return nil
+	}
+
+	startPos := fset.Position(gd.Pos())
+	endPos := fset.Position(gd.End())
+
+	chunk := &chunking.Chunk{
+		FilePath:  filePath,
+		Language:  chunking.LanguageGo,
+		Name:      ts.Name.Name,
+		StartLine: startPos.Line,
+		EndLine:   endPos.Line,
+		Content:   c.extractLines(sourceLines, startPos.Line, endPos.Line),
+	}
+
+	// Determine chunk type based on type expression
+	switch ts.Type.(type) {
+	case *ast.StructType:
+		chunk.Type = chunking.ChunkTypeClass // Treat struct as class
+	case *ast.InterfaceType:
+		chunk.Type = chunking.ChunkTypeInterface
+	default:
+		chunk.Type = chunking.ChunkTypeType
+	}
+
+	// Extract doc comment
+	if c.options.IncludeDocComments && gd.Doc != nil {
+		chunk.DocComment = strings.TrimSpace(gd.Doc.Text())
+	}
+
+	return chunk
+}
+
+// extractValueSpec extracts const or var declarations.
+func (c *Chunker) extractValueSpec(fset *token.FileSet, gd *ast.GenDecl, vs *ast.ValueSpec, sourceLines []string, filePath string) *chunking.Chunk {
+	// Skip if all names are unexported and we're excluding private
+	if !c.options.IncludePrivate {
+		allUnexported := true
+		for _, name := range vs.Names {
+			if name.IsExported() {
+				allUnexported = false
+				break
+			}
+		}
+		if allUnexported {
+			return nil
+		}
+	}
+
+	startPos := fset.Position(gd.Pos())
+	endPos := fset.Position(gd.End())
+
+	// Use first name as the chunk name, join multiple if present
+	names := make([]string, len(vs.Names))
+	for i, name := range vs.Names {
+		names[i] = name.Name
+	}
+
+	chunk := &chunking.Chunk{
+		FilePath:  filePath,
+		Language:  chunking.LanguageGo,
+		Name:      strings.Join(names, ", "),
+		StartLine: startPos.Line,
+		EndLine:   endPos.Line,
+		Content:   c.extractLines(sourceLines, startPos.Line, endPos.Line),
+	}
+
+	// Set type based on token
+	if gd.Tok == token.CONST {
+		chunk.Type = chunking.ChunkTypeConst
+	} else {
+		chunk.Type = chunking.ChunkTypeVar
+	}
+
+	// Extract doc comment
+	if c.options.IncludeDocComments && gd.Doc != nil {
+		chunk.DocComment = strings.TrimSpace(gd.Doc.Text())
+	}
+
+	return chunk
+}
+
+// extractReceiverType extracts the receiver type name from a method.
+func (c *Chunker) extractReceiverType(recv *ast.FieldList) string {
+	if len(recv.List) == 0 {
+		return ""
+	}
+
+	field := recv.List[0]
+	switch t := field.Type.(type) {
+	case *ast.Ident:
+		return t.Name
+	case *ast.StarExpr:
+		if ident, ok := t.X.(*ast.Ident); ok {
+			return ident.Name
+		}
+	}
+
+	return ""
+}
+
+// extractFunctionSignature extracts the function signature without the body.
+func (c *Chunker) extractFunctionSignature(fn *ast.FuncDecl, fset *token.FileSet, sourceLines []string) string {
+	if fn.Body == nil {
+		// No body, return entire declaration
+		startPos := fset.Position(fn.Pos())
+		endPos := fset.Position(fn.End())
+		return c.extractLines(sourceLines, startPos.Line, endPos.Line)
+	}
+
+	// Extract from start of function to just before body
+	startPos := fset.Position(fn.Pos())
+	bodyPos := fset.Position(fn.Body.Pos())
+
+	// If body is on the same line, extract just that line up to the opening brace
+	if startPos.Line == bodyPos.Line {
+		line := sourceLines[startPos.Line-1]
+		// Find the opening brace position
+		if idx := strings.Index(line[startPos.Column-1:], "{"); idx >= 0 {
+			return strings.TrimSpace(line[startPos.Column-1 : startPos.Column-1+idx])
+		}
+		return strings.TrimSpace(line[startPos.Column-1:])
+	}
+
+	// Get lines from start to the line containing the opening brace
+	sig := c.extractLines(sourceLines, startPos.Line, bodyPos.Line)
+	// Remove the opening brace and anything after it
+	if idx := strings.Index(sig, "{"); idx >= 0 {
+		sig = sig[:idx]
+	}
+	return strings.TrimSpace(sig)
+}
+
+// extractLines extracts a range of lines from source (1-indexed, inclusive).
+func (c *Chunker) extractLines(lines []string, start, end int) string {
+	if start < 1 || end < start || start > len(lines) {
+		return ""
+	}
+
+	// Adjust for 0-indexed array (start and end are 1-indexed)
+	startIdx := start - 1
+	endIdx := end
+	if endIdx > len(lines) {
+		endIdx = len(lines)
+	}
+
+	return strings.Join(lines[startIdx:endIdx], "\n")
+}

internal/chunking/golang/chunker_test.go

@@ -0,0 +1,214 @@
+package golang
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/chunking"
+)
+
+func TestGoChunker_BasicFunctions(t *testing.T) {
+	// Create temp test file
+	tmpDir := t.TempDir()
+	testFile := filepath.Join(tmpDir, "test.go")
+
+	testCode := `package main
+
+import "fmt"
+
+// Greet prints a greeting message
+func Greet(name string) {
+	fmt.Printf("Hello, %s!\n", name)
+}
+
+// Add adds two numbers
+func Add(a, b int) int {
+	return a + b
+}
+
+// unexported function should be included by default
+func helper() string {
+	return "helper"
+}
+`
+
+	if err := os.WriteFile(testFile, []byte(testCode), 0600); err != nil {
+		t.Fatalf("Failed to create test file: %v", err)
+	}
+
+	// Create chunker with default options
+	chunker := NewChunker(chunking.DefaultChunkOptions())
+
+	// Chunk the file
+	chunks, err := chunker.Chunk(context.Background(), testFile)
+	if err != nil {
+		t.Fatalf("Chunk() failed: %v", err)
+	}
+
+	// Verify we got all functions
+	if len(chunks) != 3 {
+		t.Errorf("Expected 3 chunks (Greet, Add, helper), got %d", len(chunks))
+	}
+
+	// Verify chunk details
+	expectedNames := map[string]bool{
+		"Greet":  false,
+		"Add":    false,
+		"helper": false,
+	}
+
+	for _, chunk := range chunks {
+		if chunk.Type != chunking.ChunkTypeFunction {
+			t.Errorf("Expected chunk type 'function', got '%s'", chunk.Type)
+		}
+
+		if chunk.Language != chunking.LanguageGo {
+			t.Errorf("Expected language 'go', got '%s'", chunk.Language)
+		}
+
+		if _, ok := expectedNames[chunk.Name]; !ok {
+			t.Errorf("Unexpected function name: %s", chunk.Name)
+		} else {
+			expectedNames[chunk.Name] = true
+		}
+
+		// Verify content is non-empty
+		if chunk.Content == "" {
+			t.Errorf("Chunk %s has empty content", chunk.Name)
+		}
+
+		// Verify signature is present for functions
+		if chunk.Signature == "" {
+			t.Errorf("Chunk %s has empty signature", chunk.Name)
+		}
+	}
+
+	// Verify all expected functions were found
+	for name, found := range expectedNames {
+		if !found {
+			t.Errorf("Expected function %s not found", name)
+		}
+	}
+}
+
+func TestGoChunker_StructsAndMethods(t *testing.T) {
+	tmpDir := t.TempDir()
+	testFile := filepath.Join(tmpDir, "test.go")
+
+	testCode := `package main
+
+// User represents a user
+type User struct {
+	ID   int
+	Name string
+}
+
+// GetName returns the user's name
+func (u *User) GetName() string {
+	return u.Name
+}
+
+// SetName sets the user's name
+func (u *User) SetName(name string) {
+	u.Name = name
+}
+`
+
+	if err := os.WriteFile(testFile, []byte(testCode), 0600); err != nil {
+		t.Fatalf("Failed to create test file: %v", err)
+	}
+
+	chunker := NewChunker(chunking.DefaultChunkOptions())
+	chunks, err := chunker.Chunk(context.Background(), testFile)
+	if err != nil {
+		t.Fatalf("Chunk() failed: %v", err)
+	}
+
+	// Should have 1 struct + 2 methods = 3 chunks
+	if len(chunks) != 3 {
+		t.Errorf("Expected 3 chunks (User struct, GetName, SetName), got %d", len(chunks))
+	}
+
+	// Find the struct and methods
+	var structChunk, getNameChunk, setNameChunk *chunking.Chunk
+	for i := range chunks {
+		switch chunks[i].Name {
+		case "User":
+			structChunk = &chunks[i]
+		case "GetName":
+			getNameChunk = &chunks[i]
+		case "SetName":
+			setNameChunk = &chunks[i]
+		}
+	}
+
+	// Verify struct
+	if structChunk == nil {
+		t.Fatal("User struct not found")
+	}
+	if structChunk.Type != chunking.ChunkTypeClass {
+		t.Errorf("Expected User to be ChunkTypeClass, got %s", structChunk.Type)
+	}
+
+	// Verify methods
+	if getNameChunk == nil {
+		t.Fatal("GetName method not found")
+	}
+	if getNameChunk.Type != chunking.ChunkTypeMethod {
+		t.Errorf("Expected GetName to be ChunkTypeMethod, got %s", getNameChunk.Type)
+	}
+	if getNameChunk.ParentName != "User" {
+		t.Errorf("Expected GetName parent to be 'User', got '%s'", getNameChunk.ParentName)
+	}
+
+	if setNameChunk == nil {
+		t.Fatal("SetName method not found")
+	}
+	if setNameChunk.Type != chunking.ChunkTypeMethod {
+		t.Errorf("Expected SetName to be ChunkTypeMethod, got %s", setNameChunk.Type)
+	}
+	if setNameChunk.ParentName != "User" {
+		t.Errorf("Expected SetName parent to be 'User', got '%s'", setNameChunk.ParentName)
+	}
+}
+
+func TestGoChunker_DocComments(t *testing.T) {
+	tmpDir := t.TempDir()
+	testFile := filepath.Join(tmpDir, "test.go")
+
+	testCode := `package main
+
+// Calculate performs a calculation.
+// It takes two integers and returns their sum.
+func Calculate(a, b int) int {
+	return a + b
+}
+`
+
+	if err := os.WriteFile(testFile, []byte(testCode), 0600); err != nil {
+		t.Fatalf("Failed to create test file: %v", err)
+	}
+
+	chunker := NewChunker(chunking.DefaultChunkOptions())
+	chunks, err := chunker.Chunk(context.Background(), testFile)
+	if err != nil {
+		t.Fatalf("Chunk() failed: %v", err)
+	}
+
+	if len(chunks) != 1 {
+		t.Fatalf("Expected 1 chunk, got %d", len(chunks))
+	}
+
+	chunk := chunks[0]
+	if chunk.DocComment == "" {
+		t.Error("Expected doc comment to be present")
+	}
+
+	// Doc comment should contain the comment text
+	expectedComment := "Calculate performs a calculation.\nIt takes two integers and returns their sum."
+	if chunk.DocComment != expectedComment {
+		t.Errorf("Expected doc comment '%s', got '%s'", expectedComment, chunk.DocComment)
+	}
+}

internal/chunking/manager.go

@@ -0,0 +1,106 @@
+package chunking
+
+import (
+	"context"
+	"fmt"
+	"path/filepath"
+	"strings"
+)
+
+// Manager dispatches files to appropriate language-specific chunkers.
+type Manager struct {
+	chunkers map[string]Chunker // extension -> chunker
+	options  ChunkOptions
+}
+
+// NewManager creates a new chunking manager with the given chunkers.
+func NewManager(chunkers []Chunker, options ChunkOptions) *Manager {
+	m := &Manager{
+		chunkers: make(map[string]Chunker),
+		options:  options,
+	}
+
+	// Register chunkers by their supported extensions
+	for _, chunker := range chunkers {
+		for _, ext := range chunker.SupportedExtensions() {
+			m.chunkers[ext] = chunker
+		}
+	}
+
+	return m
+}
+
+// ChunkFile chunks a single file using the appropriate language chunker.
+// Returns an error if no chunker is found for the file extension.
+func (m *Manager) ChunkFile(ctx context.Context, filePath string) ([]Chunk, error) {
+	ext := strings.ToLower(filepath.Ext(filePath))
+	chunker, ok := m.chunkers[ext]
+	if !ok {
+		return nil, fmt.Errorf("no chunker for extension %s", ext)
+	}
+
+	chunks, err := chunker.Chunk(ctx, filePath)
+	if err != nil {
+		return nil, fmt.Errorf("chunk %s: %w", filePath, err)
+	}
+
+	// Apply options-based filtering
+	filtered := make([]Chunk, 0, len(chunks))
+	for _, chunk := range chunks {
+		// Filter by minimum lines
+		if m.options.MinLines > 0 {
+			lineCount := chunk.EndLine - chunk.StartLine + 1
+			if lineCount < m.options.MinLines {
+				continue
+			}
+		}
+
+		// Filter by maximum chunk size
+		if m.options.MaxChunkSize > 0 && len(chunk.Content) > m.options.MaxChunkSize {
+			// TODO: Consider splitting large chunks intelligently
+			// For now, skip chunks that are too large
+			continue
+		}
+
+		filtered = append(filtered, chunk)
+	}
+
+	return filtered, nil
+}
+
+// ChunkFiles chunks multiple files in parallel.
+// Returns a map of file path to chunks, and any errors encountered.
+// Errors for individual files do not stop processing of other files.
+func (m *Manager) ChunkFiles(ctx context.Context, filePaths []string) (map[string][]Chunk, []error) {
+	results := make(map[string][]Chunk)
+	var errors []error
+
+	for _, filePath := range filePaths {
+		chunks, err := m.ChunkFile(ctx, filePath)
+		if err != nil {
+			errors = append(errors, fmt.Errorf("%s: %w", filePath, err))
+			continue
+		}
+		if len(chunks) > 0 {
+			results[filePath] = chunks
+		}
+	}
+
+	return results, errors
+}
+
+// SupportsFile checks if the manager can chunk the given file based on extension.
+func (m *Manager) SupportsFile(filePath string) bool {
+	ext := strings.ToLower(filepath.Ext(filePath))
+	_, ok := m.chunkers[ext]
+	return ok
+}
+
+// SupportedExtensions returns all file extensions supported by registered chunkers.
+func (m *Manager) SupportedExtensions() []string {
+	exts := make([]string, 0, len(m.chunkers))
+	for ext := range m.chunkers {
+		exts = append(exts, ext)
+	}
+	return exts
+}

internal/chunking/manager_test.go

@@ -0,0 +1,162 @@
+package chunking
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+// mockChunker is a test chunker that returns dummy chunks
+type mockChunker struct{}
+
+func (m *mockChunker) Chunk(ctx context.Context, filePath string) ([]Chunk, error) {
+	// Just return an empty chunk for testing
+	return []Chunk{
+		{
+			FilePath:  filePath,
+			Language:  LanguageGo,
+			Type:      ChunkTypeFunction,
+			Name:      "TestFunc",
+			StartLine: 1,
+			EndLine:   1,
+			Content:   "test",
+		},
+	}, nil
+}
+
+func (m *mockChunker) Language() Language {
+	return LanguageGo
+}
+
+func (m *mockChunker) SupportedExtensions() []string {
+	return []string{".go", ".py", ".ts"}
+}
+
+func TestManager_ChunkMultipleFiles(t *testing.T) {
+	tmpDir := t.TempDir()
+
+	// Create a Go file
+	goFile := filepath.Join(tmpDir, "test.go")
+	goCode := `package main
+
+func Hello() string {
+	return "hello"
+}
+`
+	if err := os.WriteFile(goFile, []byte(goCode), 0600); err != nil {
+		t.Fatalf("Failed to create Go file: %v", err)
+	}
+
+	// Create a Python file
+	pyFile := filepath.Join(tmpDir, "test.py")
+	pyCode := `def greet(name):
+    return f"Hello, {name}!"
+
+class User:
+    def __init__(self, name):
+        self.name = name
+`
+	if err := os.WriteFile(pyFile, []byte(pyCode), 0600); err != nil {
+		t.Fatalf("Failed to create Python file: %v", err)
+	}
+
+	// Create a TypeScript file
+	tsFile := filepath.Join(tmpDir, "test.ts")
+	tsCode := `function add(a: number, b: number): number {
+    return a + b;
+}
+
+class Calculator {
+    multiply(a: number, b: number): number {
+        return a * b;
+    }
+}
+`
+	if err := os.WriteFile(tsFile, []byte(tsCode), 0600); err != nil {
+		t.Fatalf("Failed to create TypeScript file: %v", err)
+	}
+
+	// Create manager
+	manager := NewManager([]Chunker{&mockChunker{}}, DefaultChunkOptions())
+
+	// Test SupportsFile
+	if !manager.SupportsFile(goFile) {
+		t.Error("Manager should support .go files")
+	}
+	if !manager.SupportsFile(pyFile) {
+		t.Error("Manager should support .py files")
+	}
+	if !manager.SupportsFile(tsFile) {
+		t.Error("Manager should support .ts files")
+	}
+
+	unsupportedFile := filepath.Join(tmpDir, "test.txt")
+	if manager.SupportsFile(unsupportedFile) {
+		t.Error("Manager should not support .txt files")
+	}
+
+	// Test ChunkFiles
+	results, errs := manager.ChunkFiles(context.Background(), []string{goFile, pyFile, tsFile})
+	if len(errs) > 0 {
+		t.Errorf("ChunkFiles returned errors: %v", errs)
+	}
+
+	if len(results) != 3 {
+		t.Errorf("Expected results for 3 files, got %d", len(results))
+	}
+
+	// Verify each file has chunks
+	for _, file := range []string{goFile, pyFile, tsFile} {
+		if chunks, ok := results[file]; !ok || len(chunks) == 0 {
+			t.Errorf("No chunks found for file %s", file)
+		}
+	}
+}
+
+// mockChunkerWithExts is a test chunker with configurable extensions
+type mockChunkerWithExts struct {
+	exts []string
+}
+
+func (m *mockChunkerWithExts) Chunk(ctx context.Context, filePath string) ([]Chunk, error) {
+	return nil, nil
+}
+
+func (m *mockChunkerWithExts) Language() Language {
+	return LanguageGo
+}
+
+func (m *mockChunkerWithExts) SupportedExtensions() []string {
+	return m.exts
+}
+
+func TestManager_SupportedExtensions(t *testing.T) {
+
+	// Create manager with mock chunkers
+	manager := NewManager([]Chunker{
+		&mockChunkerWithExts{exts: []string{".go"}},
+		&mockChunkerWithExts{exts: []string{".py", ".pyw"}},
+	}, DefaultChunkOptions())
+
+	exts := manager.SupportedExtensions()
+	expectedExts := map[string]bool{
+		".go":  false,
+		".py":  false,
+		".pyw": false,
+	}
+
+	for _, ext := range exts {
+		if _, ok := expectedExts[ext]; ok {
+			expectedExts[ext] = true
+		} else {
+			t.Errorf("Unexpected extension: %s", ext)
+		}
+	}
+
+	for ext, found := range expectedExts {
+		if !found {
+			t.Errorf("Expected extension %s not found", ext)
+		}
+	}
+}

internal/chunking/python/chunker.go

@@ -0,0 +1,291 @@
+// Package python provides AST-aware chunking for Python source files using tree-sitter.
+package python
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"strings"
+
+	sitter "github.com/smacker/go-tree-sitter"
+	"github.com/smacker/go-tree-sitter/python"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/chunking"
+)
+
+// Chunker implements AST-aware chunking for Python files.
+type Chunker struct {
+	parser  *sitter.Parser
+	options chunking.ChunkOptions
+}
+
+// NewChunker creates a new Python chunker.
+func NewChunker(options chunking.ChunkOptions) *Chunker {
+	parser := sitter.NewParser()
+	parser.SetLanguage(python.GetLanguage())
+
+	return &Chunker{
+		options: options,
+		parser:  parser,
+	}
+}
+
+// Language returns the language this chunker supports.
+func (c *Chunker) Language() chunking.Language {
+	return chunking.LanguagePython
+}
+
+// SupportedExtensions returns the file extensions this chunker handles.
+func (c *Chunker) SupportedExtensions() []string {
+	return []string{".py"}
+}
+
+// Chunk parses a Python source file and returns semantic code chunks.
+func (c *Chunker) Chunk(ctx context.Context, filePath string) ([]chunking.Chunk, error) {
+	// Read file content
+	content, err := os.ReadFile(filePath)
+	if err != nil {
+		return nil, fmt.Errorf("read file: %w", err)
+	}
+
+	// Parse the Python file
+	tree, err := c.parser.ParseCtx(ctx, nil, content)
+	if err != nil {
+		return nil, fmt.Errorf("parse Python file: %w", err)
+	}
+	defer tree.Close()
+
+	sourceLines := strings.Split(string(content), "\n")
+	chunks := make([]chunking.Chunk, 0)
+
+	// Walk the AST and extract chunks
+	c.walkNode(tree.RootNode(), content, sourceLines, filePath, "", &chunks)
+
+	return chunks, nil
+}
+
+// walkNode recursively walks the tree-sitter AST and extracts chunks.
+func (c *Chunker) walkNode(node *sitter.Node, source []byte, sourceLines []string, filePath string, parentName string, chunks *[]chunking.Chunk) {
+	nodeType := node.Type()
+
+	switch nodeType {
+	case "function_definition":
+		chunk := c.extractFunction(node, source, sourceLines, filePath, parentName)
+		if chunk != nil {
+			*chunks = append(*chunks, *chunk)
+		}
+
+	case "class_definition":
+		chunk := c.extractClass(node, source, sourceLines, filePath)
+		if chunk != nil {
+			*chunks = append(*chunks, *chunk)
+
+			// Walk class body to find methods
+			for i := 0; i < int(node.ChildCount()); i++ {
+				child := node.Child(i)
+				if child.Type() == "block" {
+					c.walkNode(child, source, sourceLines, filePath, chunk.Name, chunks)
+				}
+			}
+		}
+		return // Don't walk children again
+
+	case "block":
+		// Walk statements in block
+		for i := 0; i < int(node.ChildCount()); i++ {
+			c.walkNode(node.Child(i), source, sourceLines, filePath, parentName, chunks)
+		}
+		return
+	}
+
+	// Walk all children
+	for i := 0; i < int(node.ChildCount()); i++ {
+		c.walkNode(node.Child(i), source, sourceLines, filePath, parentName, chunks)
+	}
+}
+
+// extractFunction extracts a function definition chunk.
+func (c *Chunker) extractFunction(node *sitter.Node, source []byte, sourceLines []string, filePath string, parentName string) *chunking.Chunk {
+	// Find function name
+	var nameNode *sitter.Node
+	for i := 0; i < int(node.ChildCount()); i++ {
+		child := node.Child(i)
+		if child.Type() == "identifier" {
+			nameNode = child
+			break
+		}
+	}
+
+	if nameNode == nil {
+		return nil
+	}
+
+	name := nameNode.Content(source)
+
+	// Skip private functions if configured
+	if !c.options.IncludePrivate && strings.HasPrefix(name, "_") && !strings.HasPrefix(name, "__") {
+		return nil
+	}
+
+	startLine := int(node.StartPoint().Row) + 1
+	endLine := int(node.EndPoint().Row) + 1
+
+	chunk := &chunking.Chunk{
+		FilePath:   filePath,
+		Language:   chunking.LanguagePython,
+		Name:       name,
+		ParentName: parentName,
+		StartLine:  startLine,
+		EndLine:    endLine,
+		Content:    c.extractLines(sourceLines, startLine, endLine),
+	}
+
+	// Determine if this is a method or function
+	if parentName != "" {
+		chunk.Type = chunking.ChunkTypeMethod
+	} else {
+		chunk.Type = chunking.ChunkTypeFunction
+	}
+
+	// Extract signature (def line)
+	chunk.Signature = c.extractFunctionSignature(node, source, sourceLines)
+
+	// Extract docstring as doc comment
+	if c.options.IncludeDocComments {
+		chunk.DocComment = c.extractDocstring(node, source)
+	}
+
+	return chunk
+}
+
+// extractClass extracts a class definition chunk.
+func (c *Chunker) extractClass(node *sitter.Node, source []byte, sourceLines []string, filePath string) *chunking.Chunk {
+	// Find class name
+	var nameNode *sitter.Node
+	for i := 0; i < int(node.ChildCount()); i++ {
+		child := node.Child(i)
+		if child.Type() == "identifier" {
+			nameNode = child
+			break
+		}
+	}
+
+	if nameNode == nil {
+		return nil
+	}
+
+	name := nameNode.Content(source)
+
+	// Skip private classes if configured
+	if !c.options.IncludePrivate && strings.HasPrefix(name, "_") && !strings.HasPrefix(name, "__") {
+		return nil
+	}
+
+	startLine := int(node.StartPoint().Row) + 1
+	endLine := int(node.EndPoint().Row) + 1
+
+	chunk := &chunking.Chunk{
+		FilePath:  filePath,
+		Language:  chunking.LanguagePython,
+		Type:      chunking.ChunkTypeClass,
+		Name:      name,
+		StartLine: startLine,
+		EndLine:   endLine,
+		Content:   c.extractLines(sourceLines, startLine, endLine),
+	}
+
+	// Extract class signature (class line)
+	chunk.Signature = c.extractClassSignature(node, source, sourceLines)
+
+	// Extract docstring as doc comment
+	if c.options.IncludeDocComments {
+		chunk.DocComment = c.extractDocstring(node, source)
+	}
+
+	return chunk
+}
+
+// extractFunctionSignature extracts the function definition line.
+func (c *Chunker) extractFunctionSignature(node *sitter.Node, source []byte, sourceLines []string) string {
+	startLine := int(node.StartPoint().Row) + 1
+
+	// Find the colon that ends the signature
+	for i := 0; i < int(node.ChildCount()); i++ {
+		child := node.Child(i)
+		if child.Type() == ":" {
+			endLine := int(child.EndPoint().Row) + 1
+			return strings.TrimSpace(c.extractLines(sourceLines, startLine, endLine))
+		}
+	}
+
+	// Fallback: just return first line
+	return strings.TrimSpace(c.extractLines(sourceLines, startLine, startLine))
+}
+
+// extractClassSignature extracts the class definition line.
+func (c *Chunker) extractClassSignature(node *sitter.Node, source []byte, sourceLines []string) string {
+	startLine := int(node.StartPoint().Row) + 1
+
+	// Find the colon that ends the signature
+	for i := 0; i < int(node.ChildCount()); i++ {
+		child := node.Child(i)
+		if child.Type() == ":" {
+			endLine := int(child.EndPoint().Row) + 1
+			return strings.TrimSpace(c.extractLines(sourceLines, startLine, endLine))
+		}
+	}
+
+	// Fallback: just return first line
+	return strings.TrimSpace(c.extractLines(sourceLines, startLine, startLine))
+}
+
+// extractDocstring extracts the docstring from a function or class.
+func (c *Chunker) extractDocstring(node *sitter.Node, source []byte) string {
+	// Find the block
+	var blockNode *sitter.Node
+	for i := 0; i < int(node.ChildCount()); i++ {
+		child := node.Child(i)
+		if child.Type() == "block" {
+			blockNode = child
+			break
+		}
+	}
+
+	if blockNode == nil {
+		return ""
+	}
+
+	// Check if first statement in block is a string (docstring)
+	for i := 0; i < int(blockNode.ChildCount()); i++ {
+		child := blockNode.Child(i)
+		if child.Type() == "expression_statement" {
+			// Check if it contains a string
+			for j := 0; j < int(child.ChildCount()); j++ {
+				grandchild := child.Child(j)
+				if grandchild.Type() == "string" {
+					docstring := grandchild.Content(source)
+					// Remove quotes
+					docstring = strings.Trim(docstring, `"'`)
+					return strings.TrimSpace(docstring)
+				}
+			}
+		}
+	}
+
+	return ""
+}
+
+// extractLines extracts a range of lines from source (1-indexed, inclusive).
+func (c *Chunker) extractLines(lines []string, start, end int) string {
+	if start < 1 || end < start || start > len(lines) {
+		return ""
+	}
+
+	startIdx := start - 1
+	endIdx := end
+	if endIdx > len(lines) {
+		endIdx = len(lines)
+	}
+
+	return strings.Join(lines[startIdx:endIdx], "\n")
+}

internal/chunking/types.go

@@ -0,0 +1,140 @@
+// Package chunking provides AST-aware code chunking for semantic code search.
+// Chunks code files into logical units (functions, classes, methods) that preserve
+// semantic boundaries for better vector embedding and retrieval.
+package chunking
+
+import (
+	"context"
+	"fmt"
+	"strings"
+)
+
+// ChunkType represents the type of code chunk.
+type ChunkType string
+
+const (
+	// ChunkTypeFunction represents a standalone function.
+	ChunkTypeFunction ChunkType = "function"
+	// ChunkTypeMethod represents a method on a class/struct/type.
+	ChunkTypeMethod ChunkType = "method"
+	// ChunkTypeClass represents a class or struct definition.
+	ChunkTypeClass ChunkType = "class"
+	// ChunkTypeInterface represents an interface definition.
+	ChunkTypeInterface ChunkType = "interface"
+	// ChunkTypeType represents a type alias or type definition.
+	ChunkTypeType ChunkType = "type"
+	// ChunkTypeConst represents constant declarations.
+	ChunkTypeConst ChunkType = "const"
+	// ChunkTypeVar represents variable declarations.
+	ChunkTypeVar ChunkType = "var"
+)
+
+// Language represents a programming language.
+type Language string
+
+const (
+	// LanguageGo represents the Go programming language.
+	LanguageGo Language = "go"
+	// LanguagePython represents the Python programming language.
+	LanguagePython Language = "python"
+	// LanguageTypeScript represents the TypeScript programming language.
+	LanguageTypeScript Language = "typescript"
+	// LanguageJavaScript represents the JavaScript programming language.
+	LanguageJavaScript Language = "javascript"
+)
+
+// Chunk represents a semantic code chunk with AST-derived boundaries.
+type Chunk struct {
+	Metadata   map[string]interface{}
+	FilePath   string
+	Language   Language
+	Type       ChunkType
+	Name       string
+	ParentName string
+	Content    string
+	Signature  string
+	DocComment string
+	StartLine  int
+	EndLine    int
+}
+
+// Identifier returns a human-readable identifier for this chunk.
+// Format: "ParentName.Name" for methods, "Name" for top-level.
+func (c *Chunk) Identifier() string {
+	if c.ParentName != "" {
+		return fmt.Sprintf("%s.%s", c.ParentName, c.Name)
+	}
+	return c.Name
+}
+
+// LineRange returns a human-readable line range.
+// Format: "L123-L456"
+func (c *Chunk) LineRange() string {
+	return fmt.Sprintf("L%d-L%d", c.StartLine, c.EndLine)
+}
+
+// SearchableContent returns content optimized for semantic search.
+// Combines signature, doc comment, and content in a structured format.
+func (c *Chunk) SearchableContent() string {
+	var parts []string
+
+	// Include signature for functions/methods
+	if c.Signature != "" {
+		parts = append(parts, c.Signature)
+	}
+
+	// Include doc comment
+	if c.DocComment != "" {
+		parts = append(parts, c.DocComment)
+	}
+
+	// Include actual content
+	if c.Content != "" {
+		parts = append(parts, c.Content)
+	}
+
+	return strings.Join(parts, "\n\n")
+}
+
+// Chunker is the interface for language-specific code chunkers.
+type Chunker interface {
+	// Chunk parses a source file and returns semantic code chunks.
+	// Returns an error if the file cannot be parsed or read.
+	Chunk(ctx context.Context, filePath string) ([]Chunk, error)
+
+	// Language returns the language this chunker supports.
+	Language() Language
+
+	// SupportedExtensions returns file extensions this chunker handles.
+	// Example: []string{".go"} for Go chunker
+	SupportedExtensions() []string
+}
+
+// ChunkOptions provides options for chunking behavior.
+type ChunkOptions struct {
+	// MaxChunkSize is the maximum size of a chunk in bytes.
+	// Chunks larger than this will be split (respecting boundaries where possible).
+	// 0 means no limit.
+	MaxChunkSize int
+
+	// IncludeDocComments controls whether to include documentation comments.
+	IncludeDocComments bool
+
+	// IncludePrivate controls whether to include private/unexported symbols.
+	IncludePrivate bool
+
+	// MinLines is the minimum number of lines for a chunk to be included.
+	// Chunks smaller than this will be skipped.
+	// 0 means no minimum.
+	MinLines int
+}
+
+// DefaultChunkOptions returns sensible default options.
+func DefaultChunkOptions() ChunkOptions {
+	return ChunkOptions{
+		MaxChunkSize:       8192, // ~8KB per chunk (well under token limit)
+		IncludeDocComments: true,
+		IncludePrivate:     true, // Include all symbols for comprehensive search
+		MinLines:           0,    // No minimum - include even single-line functions
+	}
+}

internal/chunking/typescript/chunker.go

@@ -0,0 +1,403 @@
+// Package typescript provides AST-aware chunking for TypeScript and JavaScript source files using tree-sitter.
+package typescript
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"strings"
+
+	sitter "github.com/smacker/go-tree-sitter"
+	"github.com/smacker/go-tree-sitter/typescript/typescript"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/chunking"
+)
+
+// Chunker implements AST-aware chunking for TypeScript/JavaScript files.
+type Chunker struct {
+	parser  *sitter.Parser
+	options chunking.ChunkOptions
+}
+
+// NewChunker creates a new TypeScript chunker.
+func NewChunker(options chunking.ChunkOptions) *Chunker {
+	parser := sitter.NewParser()
+	parser.SetLanguage(typescript.GetLanguage())
+
+	return &Chunker{
+		options: options,
+		parser:  parser,
+	}
+}
+
+// Language returns the language this chunker supports.
+func (c *Chunker) Language() chunking.Language {
+	return chunking.LanguageTypeScript
+}
+
+// SupportedExtensions returns the file extensions this chunker handles.
+func (c *Chunker) SupportedExtensions() []string {
+	return []string{".ts", ".tsx", ".js", ".jsx"}
+}
+
+// Chunk parses a TypeScript/JavaScript source file and returns semantic code chunks.
+func (c *Chunker) Chunk(ctx context.Context, filePath string) ([]chunking.Chunk, error) {
+	// Read file content
+	content, err := os.ReadFile(filePath)
+	if err != nil {
+		return nil, fmt.Errorf("read file: %w", err)
+	}
+
+	// Parse the file
+	tree, err := c.parser.ParseCtx(ctx, nil, content)
+	if err != nil {
+		return nil, fmt.Errorf("parse TypeScript file: %w", err)
+	}
+	defer tree.Close()
+
+	sourceLines := strings.Split(string(content), "\n")
+	chunks := make([]chunking.Chunk, 0)
+
+	// Walk the AST and extract chunks
+	c.walkNode(tree.RootNode(), content, sourceLines, filePath, "", &chunks)
+
+	return chunks, nil
+}
+
+// walkNode recursively walks the tree-sitter AST and extracts chunks.
+func (c *Chunker) walkNode(node *sitter.Node, source []byte, sourceLines []string, filePath string, parentName string, chunks *[]chunking.Chunk) {
+	nodeType := node.Type()
+
+	switch nodeType {
+	case "function_declaration":
+		chunk := c.extractFunction(node, source, sourceLines, filePath, parentName)
+		if chunk != nil {
+			*chunks = append(*chunks, *chunk)
+		}
+
+	case "method_definition":
+		chunk := c.extractMethod(node, source, sourceLines, filePath, parentName)
+		if chunk != nil {
+			*chunks = append(*chunks, *chunk)
+		}
+
+	case "arrow_function", "function_expression":
+		// Handle arrow functions and function expressions assigned to variables
+		chunk := c.extractFunctionExpression(node, source, sourceLines, filePath, parentName)
+		if chunk != nil {
+			*chunks = append(*chunks, *chunk)
+		}
+
+	case "class_declaration":
+		chunk := c.extractClass(node, source, sourceLines, filePath)
+		if chunk != nil {
+			*chunks = append(*chunks, *chunk)
+
+			// Walk class body to find methods
+			for i := 0; i < int(node.ChildCount()); i++ {
+				child := node.Child(i)
+				if child.Type() == "class_body" {
+					c.walkNode(child, source, sourceLines, filePath, chunk.Name, chunks)
+				}
+			}
+		}
+		return // Don't walk children again
+
+	case "interface_declaration":
+		chunk := c.extractInterface(node, source, sourceLines, filePath)
+		if chunk != nil {
+			*chunks = append(*chunks, *chunk)
+		}
+
+	case "type_alias_declaration":
+		chunk := c.extractTypeAlias(node, source, sourceLines, filePath)
+		if chunk != nil {
+			*chunks = append(*chunks, *chunk)
+		}
+	}
+
+	// Walk all children
+	for i := 0; i < int(node.ChildCount()); i++ {
+		c.walkNode(node.Child(i), source, sourceLines, filePath, parentName, chunks)
+	}
+}
+
+// extractFunction extracts a function declaration.
+func (c *Chunker) extractFunction(node *sitter.Node, source []byte, sourceLines []string, filePath string, parentName string) *chunking.Chunk {
+	name := c.findChildContent(node, "identifier", source)
+	if name == "" {
+		return nil
+	}
+
+	startLine := int(node.StartPoint().Row) + 1
+	endLine := int(node.EndPoint().Row) + 1
+
+	chunk := &chunking.Chunk{
+		FilePath:   filePath,
+		Language:   chunking.LanguageTypeScript,
+		Type:       chunking.ChunkTypeFunction,
+		Name:       name,
+		ParentName: parentName,
+		StartLine:  startLine,
+		EndLine:    endLine,
+		Content:    c.extractLines(sourceLines, startLine, endLine),
+		Signature:  c.extractFunctionSignature(node, source, sourceLines),
+	}
+
+	// Extract JSDoc comment
+	if c.options.IncludeDocComments {
+		chunk.DocComment = c.extractComment(node, source)
+	}
+
+	return chunk
+}
+
+// extractMethod extracts a method definition from a class.
+func (c *Chunker) extractMethod(node *sitter.Node, source []byte, sourceLines []string, filePath string, parentName string) *chunking.Chunk {
+	name := c.findChildContent(node, "property_identifier", source)
+	if name == "" {
+		return nil
+	}
+
+	// Skip private methods if configured
+	if !c.options.IncludePrivate && strings.HasPrefix(name, "_") {
+		return nil
+	}
+
+	startLine := int(node.StartPoint().Row) + 1
+	endLine := int(node.EndPoint().Row) + 1
+
+	chunk := &chunking.Chunk{
+		FilePath:   filePath,
+		Language:   chunking.LanguageTypeScript,
+		Type:       chunking.ChunkTypeMethod,
+		Name:       name,
+		ParentName: parentName,
+		StartLine:  startLine,
+		EndLine:    endLine,
+		Content:    c.extractLines(sourceLines, startLine, endLine),
+		Signature:  c.extractMethodSignature(node, source, sourceLines),
+	}
+
+	// Extract JSDoc comment
+	if c.options.IncludeDocComments {
+		chunk.DocComment = c.extractComment(node, source)
+	}
+
+	return chunk
+}
+
+// extractFunctionExpression extracts arrow functions and function expressions.
+func (c *Chunker) extractFunctionExpression(node *sitter.Node, source []byte, sourceLines []string, filePath string, parentName string) *chunking.Chunk {
+	// Try to find the variable name from parent
+	parent := node.Parent()
+	if parent == nil {
+		return nil
+	}
+
+	var name string
+	if parent.Type() == "variable_declarator" {
+		name = c.findChildContent(parent, "identifier", source)
+	} else if parent.Type() == "assignment_expression" {
+		// Handle const foo = () => {}
+		for i := 0; i < int(parent.ChildCount()); i++ {
+			child := parent.Child(i)
+			if child.Type() == "identifier" || child.Type() == "member_expression" {
+				name = child.Content(source)
+				break
+			}
+		}
+	}
+
+	if name == "" {
+		return nil // Anonymous function, skip
+	}
+
+	startLine := int(node.StartPoint().Row) + 1
+	endLine := int(node.EndPoint().Row) + 1
+
+	chunk := &chunking.Chunk{
+		FilePath:   filePath,
+		Language:   chunking.LanguageTypeScript,
+		Type:       chunking.ChunkTypeFunction,
+		Name:       name,
+		ParentName: parentName,
+		StartLine:  startLine,
+		EndLine:    endLine,
+		Content:    c.extractLines(sourceLines, startLine, endLine),
+	}
+
+	return chunk
+}
+
+// extractClass extracts a class declaration.
+func (c *Chunker) extractClass(node *sitter.Node, source []byte, sourceLines []string, filePath string) *chunking.Chunk {
+	name := c.findChildContent(node, "type_identifier", source)
+	if name == "" {
+		return nil
+	}
+
+	startLine := int(node.StartPoint().Row) + 1
+	endLine := int(node.EndPoint().Row) + 1
+
+	chunk := &chunking.Chunk{
+		FilePath:  filePath,
+		Language:  chunking.LanguageTypeScript,
+		Type:      chunking.ChunkTypeClass,
+		Name:      name,
+		StartLine: startLine,
+		EndLine:   endLine,
+		Content:   c.extractLines(sourceLines, startLine, endLine),
+		Signature: c.extractClassSignature(node, source, sourceLines),
+	}
+
+	// Extract JSDoc comment
+	if c.options.IncludeDocComments {
+		chunk.DocComment = c.extractComment(node, source)
+	}
+
+	return chunk
+}
+
+// extractInterface extracts an interface declaration.
+func (c *Chunker) extractInterface(node *sitter.Node, source []byte, sourceLines []string, filePath string) *chunking.Chunk {
+	name := c.findChildContent(node, "type_identifier", source)
+	if name == "" {
+		return nil
+	}
+
+	startLine := int(node.StartPoint().Row) + 1
+	endLine := int(node.EndPoint().Row) + 1
+
+	chunk := &chunking.Chunk{
+		FilePath:  filePath,
+		Language:  chunking.LanguageTypeScript,
+		Type:      chunking.ChunkTypeInterface,
+		Name:      name,
+		StartLine: startLine,
+		EndLine:   endLine,
+		Content:   c.extractLines(sourceLines, startLine, endLine),
+	}
+
+	// Extract JSDoc comment
+	if c.options.IncludeDocComments {
+		chunk.DocComment = c.extractComment(node, source)
+	}
+
+	return chunk
+}
+
+// extractTypeAlias extracts a type alias declaration.
+func (c *Chunker) extractTypeAlias(node *sitter.Node, source []byte, sourceLines []string, filePath string) *chunking.Chunk {
+	name := c.findChildContent(node, "type_identifier", source)
+	if name == "" {
+		return nil
+	}
+
+	startLine := int(node.StartPoint().Row) + 1
+	endLine := int(node.EndPoint().Row) + 1
+
+	chunk := &chunking.Chunk{
+		FilePath:  filePath,
+		Language:  chunking.LanguageTypeScript,
+		Type:      chunking.ChunkTypeType,
+		Name:      name,
+		StartLine: startLine,
+		EndLine:   endLine,
+		Content:   c.extractLines(sourceLines, startLine, endLine),
+	}
+
+	return chunk
+}
+
+// findChildContent finds the first child of the given type and returns its content.
+func (c *Chunker) findChildContent(node *sitter.Node, childType string, source []byte) string {
+	for i := 0; i < int(node.ChildCount()); i++ {
+		child := node.Child(i)
+		if child.Type() == childType {
+			return child.Content(source)
+		}
+	}
+	return ""
+}
+
+// extractFunctionSignature extracts the function signature.
+func (c *Chunker) extractFunctionSignature(node *sitter.Node, source []byte, sourceLines []string) string {
+	startLine := int(node.StartPoint().Row) + 1
+
+	// Find the opening brace of the body
+	for i := 0; i < int(node.ChildCount()); i++ {
+		child := node.Child(i)
+		if child.Type() == "statement_block" {
+			endLine := int(child.StartPoint().Row) + 1
+			return strings.TrimSpace(c.extractLines(sourceLines, startLine, endLine-1))
+		}
+	}
+
+	// Fallback: just return first line
+	return strings.TrimSpace(c.extractLines(sourceLines, startLine, startLine))
+}
+
+// extractMethodSignature extracts the method signature.
+func (c *Chunker) extractMethodSignature(node *sitter.Node, source []byte, sourceLines []string) string {
+	startLine := int(node.StartPoint().Row) + 1
+
+	// Find the opening brace of the body
+	for i := 0; i < int(node.ChildCount()); i++ {
+		child := node.Child(i)
+		if child.Type() == "statement_block" {
+			endLine := int(child.StartPoint().Row) + 1
+			return strings.TrimSpace(c.extractLines(sourceLines, startLine, endLine-1))
+		}
+	}
+
+	return strings.TrimSpace(c.extractLines(sourceLines, startLine, startLine))
+}
+
+// extractClassSignature extracts the class declaration line.
+func (c *Chunker) extractClassSignature(node *sitter.Node, source []byte, sourceLines []string) string {
+	startLine := int(node.StartPoint().Row) + 1
+
+	// Find the opening brace of the class body
+	for i := 0; i < int(node.ChildCount()); i++ {
+		child := node.Child(i)
+		if child.Type() == "class_body" {
+			endLine := int(child.StartPoint().Row) + 1
+			return strings.TrimSpace(c.extractLines(sourceLines, startLine, endLine-1))
+		}
+	}
+
+	return strings.TrimSpace(c.extractLines(sourceLines, startLine, startLine))
+}
+
+// extractComment extracts JSDoc or other comments from a node.
+func (c *Chunker) extractComment(node *sitter.Node, source []byte) string {
+	// Check previous sibling for comment
+	prevSibling := node.PrevSibling()
+	if prevSibling != nil && prevSibling.Type() == "comment" {
+		comment := prevSibling.Content(source)
+		// Remove comment markers
+		comment = strings.TrimPrefix(comment, "/**")
+		comment = strings.TrimPrefix(comment, "/*")
+		comment = strings.TrimSuffix(comment, "*/")
+		comment = strings.TrimPrefix(comment, "//")
+		return strings.TrimSpace(comment)
+	}
+
+	return ""
+}
+
+// extractLines extracts a range of lines from source (1-indexed, inclusive).
+func (c *Chunker) extractLines(lines []string, start, end int) string {
+	if start < 1 || end < start || start > len(lines) {
+		return ""
+	}
+
+	startIdx := start - 1
+	endIdx := end
+	if endIdx > len(lines) {
+		endIdx = len(lines)
+	}
+
+	return strings.Join(lines[startIdx:endIdx], "\n")
+}

internal/config/config.go

@@ -37,30 +37,34 @@ var CriticalConcepts = []string{
 
 // Config holds the application configuration.
 type Config struct {
-	// Worker settings
-	WorkerPort int `json:"worker_port"`
-
-	// Database settings
-	DBPath   string `json:"db_path"`
-	MaxConns int    `json:"max_conns"`
-
-	// SDK Agent settings
-	Model          string `json:"model"`
-	ClaudeCodePath string `json:"claude_code_path"`
-
-	// Embedding settings
-	EmbeddingModel string `json:"embedding_model"` // e.g., "bge-v1.5"
-
-	// Context injection settings
-	ContextObservations    int      `json:"context_observations"`
-	ContextFullCount       int      `json:"context_full_count"`
-	ContextSessionCount    int      `json:"context_session_count"`
-	ContextShowReadTokens  bool     `json:"context_show_read_tokens"`
-	ContextShowWorkTokens  bool     `json:"context_show_work_tokens"`
-	ContextFullField       string   `json:"context_full_field"`
-	ContextShowLastSummary bool     `json:"context_show_last_summary"`
-	ContextObsTypes        []string `json:"context_obs_types"`
-	ContextObsConcepts     []string `json:"context_obs_concepts"`
+	ContextFullField          string   `json:"context_full_field"`
+	DBPath                    string   `json:"db_path"`
+	Model                     string   `json:"model"`
+	ClaudeCodePath            string   `json:"claude_code_path"`
+	EmbeddingModel            string   `json:"embedding_model"`
+	ContextObsConcepts        []string `json:"context_obs_concepts"`
+	ContextObsTypes           []string `json:"context_obs_types"`
+	RerankingMinImprovement   float64  `json:"reranking_min_improvement"`
+	RerankingCandidates       int      `json:"reranking_candidates"`
+	RerankingAlpha            float64  `json:"reranking_alpha"`
+	GraphEdgeWeight           float64  `json:"graph_edge_weight"`
+	WorkerPort                int      `json:"worker_port"`
+	ContextMaxPromptResults   int      `json:"context_max_prompt_results"`
+	ContextObservations       int      `json:"context_observations"`
+	ContextFullCount          int      `json:"context_full_count"`
+	ContextSessionCount       int      `json:"context_session_count"`
+	ContextRelevanceThreshold float64  `json:"context_relevance_threshold"`
+	MaxConns                  int      `json:"max_conns"`
+	RerankingResults          int      `json:"reranking_results"`
+	GraphMaxHops              int      `json:"graph_max_hops"`
+	GraphBranchFactor         int      `json:"graph_branch_factor"`
+	GraphRebuildIntervalMin   int      `json:"graph_rebuild_interval_min"`
+	ContextShowLastSummary    bool     `json:"context_show_last_summary"`
+	RerankingEnabled          bool     `json:"reranking_enabled"`
+	ContextShowWorkTokens     bool     `json:"context_show_work_tokens"`
+	ContextShowReadTokens     bool     `json:"context_show_read_tokens"`
+	RerankingPureMode         bool     `json:"reranking_pure_mode"`
+	GraphEnabled              bool     `json:"graph_enabled"`
 }
 
 var (
@@ -128,20 +132,32 @@ const DefaultEmbeddingModel = "bge-v1.5"
 // Default returns a Config with default values.
 func Default() *Config {
 	return &Config{
-		WorkerPort:             DefaultWorkerPort,
-		DBPath:                 DBPath(),
-		MaxConns:               4,
-		Model:                  DefaultModel,
-		EmbeddingModel:         DefaultEmbeddingModel,
-		ContextObservations:    100,
-		ContextFullCount:       25,
-		ContextSessionCount:    10,
-		ContextShowReadTokens:  true,
-		ContextShowWorkTokens:  true,
-		ContextFullField:       "narrative",
-		ContextShowLastSummary: true,
-		ContextObsTypes:        DefaultObservationTypes,
-		ContextObsConcepts:     DefaultObservationConcepts,
+		WorkerPort:                DefaultWorkerPort,
+		DBPath:                    DBPath(),
+		MaxConns:                  4,
+		Model:                     DefaultModel,
+		EmbeddingModel:            DefaultEmbeddingModel,
+		RerankingEnabled:          true, // Enable by default for improved relevance
+		RerankingCandidates:       100,  // Retrieve top 100 candidates
+		RerankingResults:          10,   // Return top 10 after reranking
+		RerankingAlpha:            0.7,  // Favor cross-encoder score
+		RerankingMinImprovement:   0,    // Always apply reranking
+		GraphEnabled:              true, // Enable graph-aware search by default
+		GraphMaxHops:              2,    // Two-hop traversal
+		GraphBranchFactor:         5,    // Expand top 5 neighbors per node
+		GraphEdgeWeight:           0.3,  // Minimum edge weight to follow
+		GraphRebuildIntervalMin:   60,   // Rebuild graph every 60 minutes
+		ContextObservations:       100,
+		ContextFullCount:          25,
+		ContextSessionCount:       10,
+		ContextShowReadTokens:     true,
+		ContextShowWorkTokens:     true,
+		ContextFullField:          "narrative",
+		ContextShowLastSummary:    true,
+		ContextObsTypes:           DefaultObservationTypes,
+		ContextObsConcepts:        DefaultObservationConcepts,
+		ContextRelevanceThreshold: 0.3, // Minimum 30% similarity to include
+		ContextMaxPromptResults:   10,  // Cap at 10 results max (0 = no cap, threshold only)
 	}
 }
 
@@ -176,6 +192,25 @@ func Load() (*Config, error) {
 	if v, ok := settings["CLAUDE_MNEMONIC_EMBEDDING_MODEL"].(string); ok && v != "" {
 		cfg.EmbeddingModel = v
 	}
+	// Reranking settings
+	if v, ok := settings["CLAUDE_MNEMONIC_RERANKING_ENABLED"].(bool); ok {
+		cfg.RerankingEnabled = v
+	}
+	if v, ok := settings["CLAUDE_MNEMONIC_RERANKING_CANDIDATES"].(float64); ok && v > 0 {
+		cfg.RerankingCandidates = int(v)
+	}
+	if v, ok := settings["CLAUDE_MNEMONIC_RERANKING_RESULTS"].(float64); ok && v > 0 {
+		cfg.RerankingResults = int(v)
+	}
+	if v, ok := settings["CLAUDE_MNEMONIC_RERANKING_ALPHA"].(float64); ok && v >= 0 && v <= 1 {
+		cfg.RerankingAlpha = v
+	}
+	if v, ok := settings["CLAUDE_MNEMONIC_RERANKING_MIN_IMPROVEMENT"].(float64); ok && v >= 0 {
+		cfg.RerankingMinImprovement = v
+	}
+	if v, ok := settings["CLAUDE_MNEMONIC_RERANKING_PURE_MODE"].(bool); ok {
+		cfg.RerankingPureMode = v
+	}
 	if v, ok := settings["CLAUDE_MNEMONIC_CONTEXT_OBSERVATIONS"].(float64); ok {
 		cfg.ContextObservations = int(v)
 	}
@@ -191,6 +226,28 @@ func Load() (*Config, error) {
 	if v, ok := settings["CLAUDE_MNEMONIC_CONTEXT_OBS_CONCEPTS"].(string); ok && v != "" {
 		cfg.ContextObsConcepts = splitTrim(v)
 	}
+	if v, ok := settings["CLAUDE_MNEMONIC_CONTEXT_RELEVANCE_THRESHOLD"].(float64); ok && v >= 0 && v <= 1 {
+		cfg.ContextRelevanceThreshold = v
+	}
+	if v, ok := settings["CLAUDE_MNEMONIC_CONTEXT_MAX_PROMPT_RESULTS"].(float64); ok && v >= 0 {
+		cfg.ContextMaxPromptResults = int(v)
+	}
+	// Graph settings
+	if v, ok := settings["CLAUDE_MNEMONIC_GRAPH_ENABLED"].(bool); ok {
+		cfg.GraphEnabled = v
+	}
+	if v, ok := settings["CLAUDE_MNEMONIC_GRAPH_MAX_HOPS"].(float64); ok && v > 0 {
+		cfg.GraphMaxHops = int(v)
+	}
+	if v, ok := settings["CLAUDE_MNEMONIC_GRAPH_BRANCH_FACTOR"].(float64); ok && v > 0 {
+		cfg.GraphBranchFactor = int(v)
+	}
+	if v, ok := settings["CLAUDE_MNEMONIC_GRAPH_EDGE_WEIGHT"].(float64); ok && v >= 0 && v <= 1 {
+		cfg.GraphEdgeWeight = v
+	}
+	if v, ok := settings["CLAUDE_MNEMONIC_GRAPH_REBUILD_INTERVAL_MIN"].(float64); ok && v > 0 {
+		cfg.GraphRebuildIntervalMin = int(v)
+	}
 
 	return cfg, nil
 }

internal/config/config_test.go

@@ -121,8 +121,8 @@ func (s *ConfigSuite) TestLoad_TableDriven() {
 	tests := []struct {
 		name           string
 		settingsJSON   string
-		expectedPort   int
 		expectedModel  string
+		expectedPort   int
 		expectedObsObs int
 	}{
 		{
@@ -183,12 +183,12 @@ func (s *ConfigSuite) TestLoad_TableDriven() {
 			s.Require().NoError(err)
 
 			if tt.settingsJSON != "" {
-				err := os.WriteFile(
+				writeErr := os.WriteFile(
 					filepath.Join(tempDir, ".claude-mnemonic", "settings.json"),
 					[]byte(tt.settingsJSON),
 					0600,
 				)
-				s.Require().NoError(err)
+				s.Require().NoError(writeErr)
 			}
 
 			cfg, err := Load()

internal/db/sqlite/conflict.go

@@ -0,0 +1,276 @@
+// Package sqlite provides SQLite database operations for claude-mnemonic.
+package sqlite
+
+import (
+	"context"
+	"time"
+
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+)
+
+// SupersededRetentionDays is the number of days to keep superseded observations before deletion.
+const SupersededRetentionDays = 3
+
+// ConflictStore provides conflict-related database operations.
+type ConflictStore struct {
+	store *Store
+}
+
+// NewConflictStore creates a new conflict store.
+func NewConflictStore(store *Store) *ConflictStore {
+	return &ConflictStore{store: store}
+}
+
+// StoreConflict stores a new observation conflict.
+func (s *ConflictStore) StoreConflict(ctx context.Context, conflict *models.ObservationConflict) (int64, error) {
+	const query = `
+		INSERT INTO observation_conflicts
+		(newer_obs_id, older_obs_id, conflict_type, resolution, reason, detected_at, detected_at_epoch, resolved, resolved_at)
+		VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
+	`
+
+	result, err := s.store.ExecContext(ctx, query,
+		conflict.NewerObsID, conflict.OlderObsID,
+		string(conflict.ConflictType), string(conflict.Resolution),
+		conflict.Reason, conflict.DetectedAt, conflict.DetectedAtEpoch,
+		conflict.Resolved, conflict.ResolvedAt,
+	)
+	if err != nil {
+		return 0, err
+	}
+
+	return result.LastInsertId()
+}
+
+// MarkObservationSuperseded marks an observation as superseded.
+func (s *ConflictStore) MarkObservationSuperseded(ctx context.Context, obsID int64) error {
+	const query = `UPDATE observations SET is_superseded = 1 WHERE id = ?`
+	_, err := s.store.ExecContext(ctx, query, obsID)
+	return err
+}
+
+// MarkObservationsSuperseded marks multiple observations as superseded.
+func (s *ConflictStore) MarkObservationsSuperseded(ctx context.Context, obsIDs []int64) error {
+	if len(obsIDs) == 0 {
+		return nil
+	}
+
+	query := `UPDATE observations SET is_superseded = 1 WHERE id IN (?` + repeatPlaceholders(len(obsIDs)-1) + `)` // #nosec G202 -- uses parameterized placeholders
+	args := int64SliceToInterface(obsIDs)
+	_, err := s.store.db.ExecContext(ctx, query, args...)
+	return err
+}
+
+// GetConflictsByObservationID retrieves all conflicts involving an observation.
+func (s *ConflictStore) GetConflictsByObservationID(ctx context.Context, obsID int64) ([]*models.ObservationConflict, error) {
+	const query = `
+		SELECT id, newer_obs_id, older_obs_id, conflict_type, resolution, reason,
+		       detected_at, detected_at_epoch, resolved, resolved_at
+		FROM observation_conflicts
+		WHERE newer_obs_id = ? OR older_obs_id = ?
+		ORDER BY detected_at_epoch DESC
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, obsID, obsID)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return s.scanConflictRows(rows)
+}
+
+// GetUnresolvedConflicts retrieves all unresolved conflicts.
+func (s *ConflictStore) GetUnresolvedConflicts(ctx context.Context, limit int) ([]*models.ObservationConflict, error) {
+	const query = `
+		SELECT id, newer_obs_id, older_obs_id, conflict_type, resolution, reason,
+		       detected_at, detected_at_epoch, resolved, resolved_at
+		FROM observation_conflicts
+		WHERE resolved = 0
+		ORDER BY detected_at_epoch DESC
+		LIMIT ?
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, limit)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return s.scanConflictRows(rows)
+}
+
+// GetSupersededObservationIDs returns IDs of all observations that have been superseded.
+func (s *ConflictStore) GetSupersededObservationIDs(ctx context.Context, project string) ([]int64, error) {
+	const query = `
+		SELECT DISTINCT older_obs_id
+		FROM observation_conflicts oc
+		JOIN observations o ON o.id = oc.older_obs_id
+		WHERE oc.resolution = 'prefer_newer'
+		  AND (o.project = ? OR o.scope = 'global')
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, project)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	var ids []int64
+	for rows.Next() {
+		var id int64
+		if err := rows.Scan(&id); err != nil {
+			return nil, err
+		}
+		ids = append(ids, id)
+	}
+	return ids, rows.Err()
+}
+
+// ResolveConflict marks a conflict as resolved.
+func (s *ConflictStore) ResolveConflict(ctx context.Context, conflictID int64, resolution models.ConflictResolution) error {
+	now := time.Now().Format(time.RFC3339)
+	const query = `
+		UPDATE observation_conflicts
+		SET resolved = 1, resolved_at = ?, resolution = ?
+		WHERE id = ?
+	`
+	_, err := s.store.ExecContext(ctx, query, now, string(resolution), conflictID)
+	return err
+}
+
+// DeleteConflictsByObservationID deletes all conflicts involving an observation.
+// Called when an observation is deleted.
+func (s *ConflictStore) DeleteConflictsByObservationID(ctx context.Context, obsID int64) error {
+	const query = `DELETE FROM observation_conflicts WHERE newer_obs_id = ? OR older_obs_id = ?`
+	_, err := s.store.ExecContext(ctx, query, obsID, obsID)
+	return err
+}
+
+// ConflictWithDetails contains a conflict with its observation details.
+type ConflictWithDetails struct {
+	Conflict      *models.ObservationConflict
+	NewerObsTitle string
+	OlderObsTitle string
+}
+
+// CleanupSupersededObservations deletes observations that have been superseded for longer than
+// SupersededRetentionDays. Returns the IDs of deleted observations for downstream cleanup (e.g., vector DB).
+func (s *ConflictStore) CleanupSupersededObservations(ctx context.Context, project string) ([]int64, error) {
+	// Calculate cutoff time (3 days ago in milliseconds)
+	cutoffEpoch := time.Now().AddDate(0, 0, -SupersededRetentionDays).UnixMilli()
+
+	// First, find the IDs that will be deleted
+	// We delete observations that:
+	// 1. Are marked as superseded
+	// 2. Have a conflict record where they are the older observation
+	// 3. The conflict was detected more than 3 days ago
+	const selectQuery = `
+		SELECT DISTINCT o.id FROM observations o
+		JOIN observation_conflicts oc ON o.id = oc.older_obs_id
+		WHERE o.is_superseded = 1
+		  AND o.project = ?
+		  AND oc.detected_at_epoch < ?
+	`
+
+	rows, err := s.store.QueryContext(ctx, selectQuery, project, cutoffEpoch)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	var toDelete []int64
+	for rows.Next() {
+		var id int64
+		if scanErr := rows.Scan(&id); scanErr != nil {
+			return nil, scanErr
+		}
+		toDelete = append(toDelete, id)
+	}
+	if rowsErr := rows.Err(); rowsErr != nil {
+		return nil, rowsErr
+	}
+
+	if len(toDelete) == 0 {
+		return nil, nil
+	}
+
+	// Delete the conflict records first (due to foreign key constraints)
+	for _, obsID := range toDelete {
+		if delErr := s.DeleteConflictsByObservationID(ctx, obsID); delErr != nil {
+			return nil, delErr
+		}
+	}
+
+	// Delete the observations
+	deleteQuery := `DELETE FROM observations WHERE id IN (?` + repeatPlaceholders(len(toDelete)-1) + `)` // #nosec G202 -- uses parameterized placeholders
+	args := int64SliceToInterface(toDelete)
+	_, err = s.store.db.ExecContext(ctx, deleteQuery, args...)
+	if err != nil {
+		return nil, err
+	}
+
+	return toDelete, nil
+}
+
+// GetConflictsWithDetails retrieves all conflicts with observation titles for display.
+func (s *ConflictStore) GetConflictsWithDetails(ctx context.Context, project string, limit int) ([]*ConflictWithDetails, error) {
+	const query = `
+		SELECT oc.id, oc.newer_obs_id, oc.older_obs_id, oc.conflict_type, oc.resolution, oc.reason,
+		       oc.detected_at, oc.detected_at_epoch, oc.resolved, oc.resolved_at,
+		       COALESCE(newer.title, '') as newer_title,
+		       COALESCE(older.title, '') as older_title
+		FROM observation_conflicts oc
+		JOIN observations newer ON newer.id = oc.newer_obs_id
+		JOIN observations older ON older.id = oc.older_obs_id
+		WHERE newer.project = ? OR older.project = ?
+		ORDER BY oc.detected_at_epoch DESC
+		LIMIT ?
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, project, project, limit)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	var results []*ConflictWithDetails
+	for rows.Next() {
+		var c models.ObservationConflict
+		var cwd ConflictWithDetails
+		if err := rows.Scan(
+			&c.ID, &c.NewerObsID, &c.OlderObsID,
+			&c.ConflictType, &c.Resolution, &c.Reason,
+			&c.DetectedAt, &c.DetectedAtEpoch,
+			&c.Resolved, &c.ResolvedAt,
+			&cwd.NewerObsTitle, &cwd.OlderObsTitle,
+		); err != nil {
+			return nil, err
+		}
+		cwd.Conflict = &c
+		results = append(results, &cwd)
+	}
+	return results, rows.Err()
+}
+
+// scanConflictRows scans multiple conflicts from rows.
+func (s *ConflictStore) scanConflictRows(rows interface {
+	Next() bool
+	Scan(...interface{}) error
+	Err() error
+}) ([]*models.ObservationConflict, error) {
+	var conflicts []*models.ObservationConflict
+	for rows.Next() {
+		var c models.ObservationConflict
+		if err := rows.Scan(
+			&c.ID, &c.NewerObsID, &c.OlderObsID,
+			&c.ConflictType, &c.Resolution, &c.Reason,
+			&c.DetectedAt, &c.DetectedAtEpoch,
+			&c.Resolved, &c.ResolvedAt,
+		); err != nil {
+			return nil, err
+		}
+		conflicts = append(conflicts, &c)
+	}
+	return conflicts, rows.Err()
+}

internal/db/sqlite/helpers_test.go

@@ -54,14 +54,14 @@ func TestNullInt(t *testing.T) {
 func TestRepeatPlaceholders(t *testing.T) {
 	tests := []struct {
 		name     string
-		n        int
 		expected string
+		n        int
 	}{
-		{"zero", 0, ""},
-		{"negative", -1, ""},
-		{"one", 1, ", ?"},
-		{"two", 2, ", ?, ?"},
-		{"three", 3, ", ?, ?, ?"},
+		{name: "zero", n: 0, expected: ""},
+		{name: "negative", n: -1, expected: ""},
+		{name: "one", n: 1, expected: ", ?"},
+		{name: "two", n: 2, expected: ", ?, ?"},
+		{name: "three", n: 3, expected: ", ?, ?, ?"},
 	}
 
 	for _, tt := range tests {

internal/db/sqlite/migrations.go

@@ -9,9 +9,9 @@ import (
 
 // Migration represents a database schema migration.
 type Migration struct {
-	Version int
 	Name    string
 	SQL     string
+	Version int
 }
 
 // Migrations is the list of all database migrations in order.
@@ -304,6 +304,192 @@ var Migrations = []Migration{
 			);
 		`,
 	},
+	{
+		Version: 20,
+		Name:    "importance_scoring",
+		SQL: `
+			-- Importance scoring system for observations
+			-- Implements multi-factor scoring: type weight, recency decay, user feedback, concept weights, retrieval boost
+
+			-- Cached importance score (recalculated periodically)
+			ALTER TABLE observations ADD COLUMN importance_score REAL DEFAULT 1.0;
+
+			-- User feedback: -1 = thumbs down, 0 = neutral, 1 = thumbs up
+			ALTER TABLE observations ADD COLUMN user_feedback INTEGER DEFAULT 0;
+
+			-- Retrieval tracking: how many times this observation was returned in searches
+			ALTER TABLE observations ADD COLUMN retrieval_count INTEGER DEFAULT 0;
+
+			-- Last time this observation was retrieved (for analytics)
+			ALTER TABLE observations ADD COLUMN last_retrieved_at_epoch INTEGER;
+
+			-- Timestamp of last score recalculation
+			ALTER TABLE observations ADD COLUMN score_updated_at_epoch INTEGER;
+
+			-- Index for importance-based sorting (primary ordering strategy)
+			CREATE INDEX IF NOT EXISTS idx_observations_importance
+			ON observations(importance_score DESC, created_at_epoch DESC);
+
+			-- Index for finding observations needing score recalculation
+			CREATE INDEX IF NOT EXISTS idx_observations_score_updated
+			ON observations(score_updated_at_epoch);
+
+			-- Configurable concept weights table
+			-- Allows runtime tuning of how much each concept contributes to importance
+			CREATE TABLE IF NOT EXISTS concept_weights (
+				concept TEXT PRIMARY KEY,
+				weight REAL NOT NULL DEFAULT 0.1,
+				updated_at TEXT NOT NULL
+			);
+
+			-- Seed default concept weights (security highest, tooling lowest)
+			INSERT OR IGNORE INTO concept_weights (concept, weight, updated_at) VALUES
+				('security', 0.30, datetime('now')),
+				('gotcha', 0.25, datetime('now')),
+				('best-practice', 0.20, datetime('now')),
+				('anti-pattern', 0.20, datetime('now')),
+				('architecture', 0.15, datetime('now')),
+				('performance', 0.15, datetime('now')),
+				('error-handling', 0.15, datetime('now')),
+				('pattern', 0.10, datetime('now')),
+				('testing', 0.10, datetime('now')),
+				('debugging', 0.10, datetime('now')),
+				('workflow', 0.05, datetime('now')),
+				('tooling', 0.05, datetime('now'));
+		`,
+	},
+	{
+		Version: 21,
+		Name:    "observation_conflicts",
+		SQL: `
+			-- Observation conflicts table for tracking contradictions and superseded observations
+			-- Implements Issue #5: Contradiction & Obsolescence Detection
+			CREATE TABLE IF NOT EXISTS observation_conflicts (
+				id INTEGER PRIMARY KEY AUTOINCREMENT,
+				newer_obs_id INTEGER NOT NULL,
+				older_obs_id INTEGER NOT NULL,
+				conflict_type TEXT NOT NULL CHECK(conflict_type IN ('superseded', 'contradicts', 'outdated_pattern')),
+				resolution TEXT NOT NULL CHECK(resolution IN ('prefer_newer', 'prefer_older', 'manual')),
+				reason TEXT,
+				detected_at TEXT NOT NULL,
+				detected_at_epoch INTEGER NOT NULL,
+				resolved INTEGER DEFAULT 0,
+				resolved_at TEXT,
+				FOREIGN KEY(newer_obs_id) REFERENCES observations(id) ON DELETE CASCADE,
+				FOREIGN KEY(older_obs_id) REFERENCES observations(id) ON DELETE CASCADE
+			);
+
+			-- Index for looking up conflicts by observation ID
+			CREATE INDEX IF NOT EXISTS idx_conflicts_newer ON observation_conflicts(newer_obs_id);
+			CREATE INDEX IF NOT EXISTS idx_conflicts_older ON observation_conflicts(older_obs_id);
+			CREATE INDEX IF NOT EXISTS idx_conflicts_unresolved ON observation_conflicts(resolved, detected_at_epoch DESC);
+
+			-- Add is_superseded column to observations for quick filtering
+			-- Set to 1 when this observation has been superseded by a newer one
+			ALTER TABLE observations ADD COLUMN is_superseded INTEGER DEFAULT 0;
+
+			-- Index for filtering out superseded observations in queries
+			CREATE INDEX IF NOT EXISTS idx_observations_superseded ON observations(is_superseded, importance_score DESC);
+		`,
+	},
+	{
+		Version: 22,
+		Name:    "patterns_table",
+		SQL: `
+			-- Pattern Recognition Engine (Issue #7)
+			-- Tracks recurring patterns detected across observations
+			-- Enables Claude to reference historical insights: "I've encountered this pattern 12 times."
+			CREATE TABLE IF NOT EXISTS patterns (
+				id INTEGER PRIMARY KEY AUTOINCREMENT,
+				name TEXT NOT NULL,
+				type TEXT NOT NULL CHECK(type IN ('bug', 'refactor', 'architecture', 'anti-pattern', 'best-practice')),
+				description TEXT,
+				signature TEXT,  -- JSON array of keywords/concepts for detection
+				recommendation TEXT,  -- What works for this pattern
+				frequency INTEGER DEFAULT 1,  -- How many times encountered
+				projects TEXT,  -- JSON array of projects where seen
+				observation_ids TEXT,  -- JSON array of source observation IDs
+				status TEXT DEFAULT 'active' CHECK(status IN ('active', 'deprecated', 'merged')),
+				merged_into_id INTEGER,  -- If status is 'merged', which pattern it merged into
+				confidence REAL DEFAULT 0.5,  -- Detection confidence (0.0-1.0)
+				last_seen_at TEXT NOT NULL,
+				last_seen_at_epoch INTEGER NOT NULL,
+				created_at TEXT NOT NULL,
+				created_at_epoch INTEGER NOT NULL,
+				FOREIGN KEY(merged_into_id) REFERENCES patterns(id) ON DELETE SET NULL
+			);
+
+			-- Indexes for efficient pattern queries
+			CREATE INDEX IF NOT EXISTS idx_patterns_type ON patterns(type);
+			CREATE INDEX IF NOT EXISTS idx_patterns_status ON patterns(status);
+			CREATE INDEX IF NOT EXISTS idx_patterns_frequency ON patterns(frequency DESC);
+			CREATE INDEX IF NOT EXISTS idx_patterns_confidence ON patterns(confidence DESC);
+			CREATE INDEX IF NOT EXISTS idx_patterns_last_seen ON patterns(last_seen_at_epoch DESC);
+
+			-- FTS5 virtual table for pattern search
+			CREATE VIRTUAL TABLE IF NOT EXISTS patterns_fts USING fts5(
+				name, description, recommendation,
+				content='patterns',
+				content_rowid='id'
+			);
+
+			-- Triggers for FTS5 sync
+			CREATE TRIGGER IF NOT EXISTS patterns_ai AFTER INSERT ON patterns BEGIN
+				INSERT INTO patterns_fts(rowid, name, description, recommendation)
+				VALUES (new.id, new.name, new.description, new.recommendation);
+			END;
+
+			CREATE TRIGGER IF NOT EXISTS patterns_ad AFTER DELETE ON patterns BEGIN
+				INSERT INTO patterns_fts(patterns_fts, rowid, name, description, recommendation)
+				VALUES('delete', old.id, old.name, old.description, old.recommendation);
+			END;
+
+			CREATE TRIGGER IF NOT EXISTS patterns_au AFTER UPDATE ON patterns BEGIN
+				INSERT INTO patterns_fts(patterns_fts, rowid, name, description, recommendation)
+				VALUES('delete', old.id, old.name, old.description, old.recommendation);
+				INSERT INTO patterns_fts(rowid, name, description, recommendation)
+				VALUES (new.id, new.name, new.description, new.recommendation);
+			END;
+		`,
+	},
+	{
+		Version: 23,
+		Name:    "observation_relations",
+		SQL: `
+			-- Knowledge Graph: Observation Relations (Issue #4)
+			-- Tracks explicit relationships between observations for knowledge graph navigation.
+			-- Enables queries like "What caused this bug?" or "What depends on this decision?"
+			CREATE TABLE IF NOT EXISTS observation_relations (
+				id INTEGER PRIMARY KEY AUTOINCREMENT,
+				source_id INTEGER NOT NULL,
+				target_id INTEGER NOT NULL,
+				relation_type TEXT NOT NULL CHECK(relation_type IN ('causes', 'fixes', 'supersedes', 'depends_on', 'relates_to', 'evolves_from')),
+				confidence REAL NOT NULL DEFAULT 0.5,
+				detection_source TEXT NOT NULL CHECK(detection_source IN ('file_overlap', 'embedding_similarity', 'temporal_proximity', 'narrative_mention', 'concept_overlap', 'type_progression')),
+				reason TEXT,
+				created_at TEXT NOT NULL,
+				created_at_epoch INTEGER NOT NULL,
+				FOREIGN KEY(source_id) REFERENCES observations(id) ON DELETE CASCADE,
+				FOREIGN KEY(target_id) REFERENCES observations(id) ON DELETE CASCADE,
+				UNIQUE(source_id, target_id, relation_type)
+			);
+
+			-- Index for finding relations by source observation
+			CREATE INDEX IF NOT EXISTS idx_relations_source ON observation_relations(source_id);
+
+			-- Index for finding relations by target observation
+			CREATE INDEX IF NOT EXISTS idx_relations_target ON observation_relations(target_id);
+
+			-- Index for relation type queries
+			CREATE INDEX IF NOT EXISTS idx_relations_type ON observation_relations(relation_type);
+
+			-- Index for confidence-based filtering
+			CREATE INDEX IF NOT EXISTS idx_relations_confidence ON observation_relations(confidence DESC);
+
+			-- Index for finding all relations involving an observation (either direction)
+			CREATE INDEX IF NOT EXISTS idx_relations_both ON observation_relations(source_id, target_id);
+		`,
+	},
 }
 
 // MigrationManager handles database schema migrations.
@@ -353,11 +539,11 @@ func (m *MigrationManager) ApplyMigration(migration Migration) error {
 	if err != nil {
 		return fmt.Errorf("begin transaction: %w", err)
 	}
-	defer tx.Rollback()
+	defer func() { _ = tx.Rollback() }()
 
 	// Execute migration SQL
-	if _, err := tx.Exec(migration.SQL); err != nil {
-		return fmt.Errorf("execute migration %d (%s): %w", migration.Version, migration.Name, err)
+	if _, execErr := tx.Exec(migration.SQL); execErr != nil {
+		return fmt.Errorf("execute migration %d (%s): %w", migration.Version, migration.Name, execErr)
 	}
 
 	// Record migration

internal/db/sqlite/observation.go

@@ -11,14 +11,27 @@ import (
 	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
 )
 
+// observationColumns is the standard list of columns to select for observations.
+// This ensures consistency across all observation queries and includes importance scoring fields.
+const observationColumns = `id, sdk_session_id, project, COALESCE(scope, 'project') as scope, type,
+       title, subtitle, facts, narrative, concepts, files_read, files_modified, file_mtimes,
+       prompt_number, discovery_tokens, created_at, created_at_epoch,
+       COALESCE(importance_score, 1.0) as importance_score,
+       COALESCE(user_feedback, 0) as user_feedback,
+       COALESCE(retrieval_count, 0) as retrieval_count,
+       last_retrieved_at_epoch, score_updated_at_epoch,
+       COALESCE(is_superseded, 0) as is_superseded`
+
 // CleanupFunc is a callback for when observations are cleaned up.
 // Receives the IDs of deleted observations for downstream cleanup (e.g., vector DB).
 type CleanupFunc func(ctx context.Context, deletedIDs []int64)
 
 // ObservationStore provides observation-related database operations.
 type ObservationStore struct {
-	store       *Store
-	cleanupFunc CleanupFunc
+	store         *Store
+	cleanupFunc   CleanupFunc
+	conflictStore *ConflictStore
+	relationStore *RelationStore
 }
 
 // NewObservationStore creates a new observation store.
@@ -31,6 +44,16 @@ func (s *ObservationStore) SetCleanupFunc(fn CleanupFunc) {
 	s.cleanupFunc = fn
 }
 
+// SetConflictStore sets the conflict store for conflict detection.
+func (s *ObservationStore) SetConflictStore(conflictStore *ConflictStore) {
+	s.conflictStore = conflictStore
+}
+
+// SetRelationStore sets the relation store for relationship detection.
+func (s *ObservationStore) SetRelationStore(relationStore *RelationStore) {
+	s.relationStore = relationStore
+}
+
 // StoreObservation stores a new observation.
 func (s *ObservationStore) StoreObservation(ctx context.Context, sdkSessionID, project string, obs *models.ParsedObservation, promptNumber int, discoveryTokens int64) (int64, int64, error) {
 	now := time.Now()
@@ -86,9 +109,112 @@ func (s *ObservationStore) StoreObservation(ctx context.Context, sdkSessionID, p
 		}(project)
 	}
 
+	// Detect conflicts with existing observations (async to not block handler)
+	if s.conflictStore != nil && project != "" {
+		go func(newObsID int64, proj string, parsedObs *models.ParsedObservation) {
+			conflictCtx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+			defer cancel()
+			s.detectAndStoreConflicts(conflictCtx, newObsID, proj, parsedObs)
+		}(id, project, obs)
+	}
+
+	// Detect relationships with existing observations (async to not block handler)
+	if s.relationStore != nil && project != "" {
+		go func(newObsID int64, proj string) {
+			relationCtx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+			defer cancel()
+			s.detectAndStoreRelations(relationCtx, newObsID, proj)
+		}(id, project)
+	}
+
 	return id, nowEpoch, nil
 }
 
+// detectAndStoreConflicts detects conflicts between a new observation and existing ones.
+func (s *ObservationStore) detectAndStoreConflicts(ctx context.Context, newObsID int64, project string, parsedObs *models.ParsedObservation) {
+	// Fetch the newly stored observation
+	newObs, err := s.GetObservationByID(ctx, newObsID)
+	if err != nil || newObs == nil {
+		return
+	}
+
+	// Fetch recent observations from the same project to check for conflicts
+	existing, err := s.GetRecentObservations(ctx, project, 50)
+	if err != nil {
+		return
+	}
+
+	// Detect conflicts
+	conflicts := models.DetectConflictsWithExisting(newObs, existing)
+
+	// Store conflicts and mark superseded observations
+	var supersededIDs []int64
+	for _, result := range conflicts {
+		for _, olderID := range result.OlderObsIDs {
+			conflict := models.NewObservationConflict(
+				newObsID, olderID,
+				result.Type, result.Resolution, result.Reason,
+			)
+			if _, err := s.conflictStore.StoreConflict(ctx, conflict); err != nil {
+				continue
+			}
+
+			// If resolution is to prefer newer, mark older as superseded
+			if result.Resolution == models.ResolutionPreferNewer {
+				supersededIDs = append(supersededIDs, olderID)
+			}
+		}
+	}
+
+	// Mark superseded observations
+	if len(supersededIDs) > 0 {
+		_ = s.conflictStore.MarkObservationsSuperseded(ctx, supersededIDs)
+	}
+
+	// Cleanup old superseded observations (older than 3 days)
+	deletedIDs, _ := s.conflictStore.CleanupSupersededObservations(ctx, project)
+	if len(deletedIDs) > 0 && s.cleanupFunc != nil {
+		s.cleanupFunc(ctx, deletedIDs)
+	}
+}
+
+// MinRelationConfidence is the minimum confidence threshold for storing relations.
+const MinRelationConfidence = 0.4
+
+// detectAndStoreRelations detects relationships between a new observation and existing ones.
+func (s *ObservationStore) detectAndStoreRelations(ctx context.Context, newObsID int64, project string) {
+	// Fetch the newly stored observation
+	newObs, err := s.GetObservationByID(ctx, newObsID)
+	if err != nil || newObs == nil {
+		return
+	}
+
+	// Fetch recent observations from the same project to check for relations
+	existing, err := s.GetRecentObservations(ctx, project, 50)
+	if err != nil {
+		return
+	}
+
+	// Detect relationships using the models package detection logic
+	results := models.DetectRelationsWithExisting(newObs, existing, MinRelationConfidence)
+	if len(results) == 0 {
+		return
+	}
+
+	// Convert detection results to relation objects
+	relations := make([]*models.ObservationRelation, len(results))
+	for i, r := range results {
+		relations[i] = models.NewObservationRelation(
+			r.SourceID, r.TargetID,
+			r.RelationType, r.Confidence,
+			r.DetectionSource, r.Reason,
+		)
+	}
+
+	// Store all relations
+	_ = s.relationStore.StoreRelations(ctx, relations)
+}
+
 // ensureSessionExists creates a session if it doesn't exist.
 func (s *ObservationStore) ensureSessionExists(ctx context.Context, sdkSessionID, project string) error {
 	return EnsureSessionExists(ctx, s.store, sdkSessionID, project)
@@ -96,13 +222,7 @@ func (s *ObservationStore) ensureSessionExists(ctx context.Context, sdkSessionID
 
 // GetObservationByID retrieves an observation by ID.
 func (s *ObservationStore) GetObservationByID(ctx context.Context, id int64) (*models.Observation, error) {
-	const query = `
-		SELECT id, sdk_session_id, project, COALESCE(scope, 'project') as scope, type, title, subtitle, facts, narrative,
-		       concepts, files_read, files_modified, file_mtimes, prompt_number, discovery_tokens,
-		       created_at, created_at_epoch
-		FROM observations
-		WHERE id = ?
-	`
+	query := `SELECT ` + observationColumns + ` FROM observations WHERE id = ?`
 
 	obs, err := scanObservation(s.store.QueryRowContext(ctx, query, id))
 	if err == sql.ErrNoRows {
@@ -112,6 +232,7 @@ func (s *ObservationStore) GetObservationByID(ctx context.Context, id int64) (*m
 }
 
 // GetObservationsByIDs retrieves observations by a list of IDs.
+// Results are ordered by importance_score DESC by default, with created_at_epoch as secondary sort.
 func (s *ObservationStore) GetObservationsByIDs(ctx context.Context, ids []int64, orderBy string, limit int) ([]*models.Observation, error) {
 	if len(ids) == 0 {
 		return nil, nil
@@ -119,18 +240,22 @@ func (s *ObservationStore) GetObservationsByIDs(ctx context.Context, ids []int64
 
 	// Build query with placeholders
 	// #nosec G202 -- query uses parameterized placeholders, not user input
-	query := `
-		SELECT id, sdk_session_id, project, COALESCE(scope, 'project') as scope, type, title, subtitle, facts, narrative,
-		       concepts, files_read, files_modified, file_mtimes, prompt_number, discovery_tokens,
-		       created_at, created_at_epoch
+	query := `SELECT ` + observationColumns + `
 		FROM observations
 		WHERE id IN (?` + repeatPlaceholders(len(ids)-1) + `)
-		ORDER BY created_at_epoch `
+		ORDER BY `
 
-	if orderBy == "date_asc" {
-		query += "ASC"
-	} else {
-		query += "DESC"
+	// Default to importance-based ordering
+	switch orderBy {
+	case "date_asc":
+		query += "created_at_epoch ASC"
+	case "date_desc":
+		query += "created_at_epoch DESC"
+	case "importance":
+		query += "importance_score DESC, created_at_epoch DESC"
+	default:
+		// Default: importance first, then recency
+		query += "COALESCE(importance_score, 1.0) DESC, created_at_epoch DESC"
 	}
 
 	if limit > 0 {
@@ -154,14 +279,56 @@ func (s *ObservationStore) GetObservationsByIDs(ctx context.Context, ids []int64
 
 // GetRecentObservations retrieves recent observations for a project.
 // This includes project-scoped observations for the specified project AND global observations.
+// Results are ordered by importance_score DESC, then created_at_epoch DESC.
 func (s *ObservationStore) GetRecentObservations(ctx context.Context, project string, limit int) ([]*models.Observation, error) {
-	const query = `
-		SELECT id, sdk_session_id, project, COALESCE(scope, 'project') as scope, type, title, subtitle, facts, narrative,
-		       concepts, files_read, files_modified, file_mtimes, prompt_number, discovery_tokens,
-		       created_at, created_at_epoch
+	query := `SELECT ` + observationColumns + `
 		FROM observations
 		WHERE (project = ? AND (scope IS NULL OR scope = 'project'))
 		   OR scope = 'global'
+		ORDER BY COALESCE(importance_score, 1.0) DESC, created_at_epoch DESC
+		LIMIT ?
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, project, limit)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return scanObservationRows(rows)
+}
+
+// GetActiveObservations retrieves recent non-superseded observations for a project.
+// This excludes observations that have been marked as superseded by newer ones.
+// Use this for context injection where you want to avoid outdated/contradicted advice.
+// Results are ordered by importance_score DESC, then created_at_epoch DESC.
+func (s *ObservationStore) GetActiveObservations(ctx context.Context, project string, limit int) ([]*models.Observation, error) {
+	query := `SELECT ` + observationColumns + `
+		FROM observations
+		WHERE ((project = ? AND (scope IS NULL OR scope = 'project'))
+		   OR scope = 'global')
+		  AND COALESCE(is_superseded, 0) = 0
+		ORDER BY COALESCE(importance_score, 1.0) DESC, created_at_epoch DESC
+		LIMIT ?
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, project, limit)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return scanObservationRows(rows)
+}
+
+// GetSupersededObservations retrieves observations that have been superseded by newer ones.
+// Use this for verification/debugging to see which observations were marked as outdated.
+// Results are ordered by created_at_epoch DESC.
+func (s *ObservationStore) GetSupersededObservations(ctx context.Context, project string, limit int) ([]*models.Observation, error) {
+	query := `SELECT ` + observationColumns + `
+		FROM observations
+		WHERE project = ?
+		  AND COALESCE(is_superseded, 0) = 1
 		ORDER BY created_at_epoch DESC
 		LIMIT ?
 	`
@@ -178,14 +345,12 @@ func (s *ObservationStore) GetRecentObservations(ctx context.Context, project st
 // GetObservationsByProjectStrict retrieves observations strictly for a specific project.
 // Unlike GetRecentObservations, this does NOT include global observations from other projects.
 // Use this for dashboard filtering where the user expects to see only that project's data.
+// Results are ordered by importance_score DESC, then created_at_epoch DESC.
 func (s *ObservationStore) GetObservationsByProjectStrict(ctx context.Context, project string, limit int) ([]*models.Observation, error) {
-	const query = `
-		SELECT id, sdk_session_id, project, COALESCE(scope, 'project') as scope, type, title, subtitle, facts, narrative,
-		       concepts, files_read, files_modified, file_mtimes, prompt_number, discovery_tokens,
-		       created_at, created_at_epoch
+	query := `SELECT ` + observationColumns + `
 		FROM observations
 		WHERE project = ?
-		ORDER BY created_at_epoch DESC
+		ORDER BY COALESCE(importance_score, 1.0) DESC, created_at_epoch DESC
 		LIMIT ?
 	`
 
@@ -210,13 +375,11 @@ func (s *ObservationStore) GetObservationCount(ctx context.Context, project stri
 }
 
 // GetAllRecentObservations retrieves recent observations across all projects.
+// Results are ordered by importance_score DESC, then created_at_epoch DESC.
 func (s *ObservationStore) GetAllRecentObservations(ctx context.Context, limit int) ([]*models.Observation, error) {
-	const query = `
-		SELECT id, sdk_session_id, project, COALESCE(scope, 'project') as scope, type, title, subtitle, facts, narrative,
-		       concepts, files_read, files_modified, file_mtimes, prompt_number, discovery_tokens,
-		       created_at, created_at_epoch
+	query := `SELECT ` + observationColumns + `
 		FROM observations
-		ORDER BY created_at_epoch DESC
+		ORDER BY COALESCE(importance_score, 1.0) DESC, created_at_epoch DESC
 		LIMIT ?
 	`
 
@@ -231,10 +394,7 @@ func (s *ObservationStore) GetAllRecentObservations(ctx context.Context, limit i
 
 // GetAllObservations retrieves all observations (for vector rebuild).
 func (s *ObservationStore) GetAllObservations(ctx context.Context) ([]*models.Observation, error) {
-	const query = `
-		SELECT id, sdk_session_id, project, COALESCE(scope, 'project') as scope, type, title, subtitle, facts, narrative,
-		       concepts, files_read, files_modified, file_mtimes, prompt_number, discovery_tokens,
-		       created_at, created_at_epoch
+	query := `SELECT ` + observationColumns + `
 		FROM observations
 		ORDER BY id
 	`
@@ -249,6 +409,7 @@ func (s *ObservationStore) GetAllObservations(ctx context.Context) ([]*models.Ob
 }
 
 // SearchObservationsFTS performs full-text search on observations.
+// Results are ordered by FTS rank (relevance), then by importance_score.
 func (s *ObservationStore) SearchObservationsFTS(ctx context.Context, query, project string, limit int) ([]*models.Observation, error) {
 	if limit <= 0 {
 		limit = 10
@@ -264,15 +425,21 @@ func (s *ObservationStore) SearchObservationsFTS(ctx context.Context, query, pro
 	ftsTerms := strings.Join(keywords, " OR ")
 
 	// Use FTS5 to search title, subtitle, and narrative
-	const ftsQuery = `
+	// Include importance scoring columns and order by rank then importance
+	ftsQuery := `
 		SELECT o.id, o.sdk_session_id, o.project, COALESCE(o.scope, 'project') as scope, o.type,
 		       o.title, o.subtitle, o.facts, o.narrative, o.concepts, o.files_read, o.files_modified,
-		       o.file_mtimes, o.prompt_number, o.discovery_tokens, o.created_at, o.created_at_epoch
+		       o.file_mtimes, o.prompt_number, o.discovery_tokens, o.created_at, o.created_at_epoch,
+		       COALESCE(o.importance_score, 1.0) as importance_score,
+		       COALESCE(o.user_feedback, 0) as user_feedback,
+		       COALESCE(o.retrieval_count, 0) as retrieval_count,
+		       o.last_retrieved_at_epoch, o.score_updated_at_epoch,
+		       COALESCE(o.is_superseded, 0) as is_superseded
 		FROM observations o
 		JOIN observations_fts fts ON o.id = fts.rowid
 		WHERE observations_fts MATCH ?
 		  AND (o.project = ? OR o.scope = 'global')
-		ORDER BY rank
+		ORDER BY rank, COALESCE(o.importance_score, 1.0) DESC
 		LIMIT ?
 	`
 
@@ -297,6 +464,7 @@ func (s *ObservationStore) SearchObservationsFTS(ctx context.Context, query, pro
 }
 
 // searchObservationsLike performs fallback LIKE search on observations.
+// Results are ordered by importance_score DESC, then created_at_epoch DESC.
 func (s *ObservationStore) searchObservationsLike(ctx context.Context, keywords []string, project string, limit int) ([]*models.Observation, error) {
 	if len(keywords) == 0 {
 		return nil, nil
@@ -313,14 +481,11 @@ func (s *ObservationStore) searchObservationsLike(ctx context.Context, keywords
 	}
 
 	// #nosec G202 -- query uses parameterized placeholders, not user input
-	query := `
-		SELECT id, sdk_session_id, project, COALESCE(scope, 'project') as scope, type,
-		       title, subtitle, facts, narrative, concepts, files_read, files_modified,
-		       file_mtimes, prompt_number, discovery_tokens, created_at, created_at_epoch
+	query := `SELECT ` + observationColumns + `
 		FROM observations
 		WHERE (` + strings.Join(conditions, " OR ") + `)
 		  AND (project = ? OR scope = 'global')
-		ORDER BY created_at_epoch DESC
+		ORDER BY COALESCE(importance_score, 1.0) DESC, created_at_epoch DESC
 		LIMIT ?
 	`
 	args = append(args, project, limit)
@@ -420,13 +585,13 @@ func (s *ObservationStore) CleanupOldObservations(ctx context.Context, project s
 	var toDelete []int64
 	for rows.Next() {
 		var id int64
-		if err := rows.Scan(&id); err != nil {
-			return nil, err
+		if scanErr := rows.Scan(&id); scanErr != nil {
+			return nil, scanErr
 		}
 		toDelete = append(toDelete, id)
 	}
-	if err := rows.Err(); err != nil {
-		return nil, err
+	if rowsErr := rows.Err(); rowsErr != nil {
+		return nil, rowsErr
 	}
 
 	if len(toDelete) == 0 {
@@ -464,6 +629,11 @@ func scanObservation(scanner interface{ Scan(...interface{}) error }) (*models.O
 		&obs.Concepts, &obs.FilesRead, &obs.FilesModified, &obs.FileMtimes,
 		&obs.PromptNumber, &obs.DiscoveryTokens,
 		&obs.CreatedAt, &obs.CreatedAtEpoch,
+		// Importance scoring fields
+		&obs.ImportanceScore, &obs.UserFeedback, &obs.RetrievalCount,
+		&obs.LastRetrievedAt, &obs.ScoreUpdatedAt,
+		// Conflict detection fields
+		&obs.IsSuperseded,
 	); err != nil {
 		return nil, err
 	}

internal/db/sqlite/observation_test.go

@@ -65,10 +65,10 @@ func (s *ObservationStoreSuite) TestStoreObservation_TableDriven() {
 	ctx := context.Background()
 
 	tests := []struct {
+		obs          *models.ParsedObservation
 		name         string
 		sdkSessionID string
 		project      string
-		obs          *models.ParsedObservation
 		promptNum    int
 		tokens       int64
 		wantErr      bool
@@ -308,8 +308,8 @@ func (s *ObservationStoreSuite) TestGetObservationsByIDs() {
 
 	tests := []struct {
 		name      string
-		queryIDs  []int64
 		orderBy   string
+		queryIDs  []int64
 		limit     int
 		wantCount int
 	}{

internal/db/sqlite/pattern.go

@@ -0,0 +1,370 @@
+// Package sqlite provides SQLite database operations for claude-mnemonic.
+package sqlite
+
+import (
+	"context"
+	"database/sql"
+	"encoding/json"
+	"time"
+
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+)
+
+// patternColumns is the standard list of columns to select for patterns.
+const patternColumns = `id, name, type, description, signature, recommendation,
+       frequency, projects, observation_ids, status, merged_into_id, confidence,
+       last_seen_at, last_seen_at_epoch, created_at, created_at_epoch`
+
+// PatternCleanupFunc is a callback for when patterns are deleted.
+type PatternCleanupFunc func(ctx context.Context, deletedIDs []int64)
+
+// PatternStore provides pattern-related database operations.
+type PatternStore struct {
+	store       *Store
+	cleanupFunc PatternCleanupFunc
+}
+
+// NewPatternStore creates a new pattern store.
+func NewPatternStore(store *Store) *PatternStore {
+	return &PatternStore{store: store}
+}
+
+// SetCleanupFunc sets the callback for when patterns are deleted.
+func (s *PatternStore) SetCleanupFunc(fn PatternCleanupFunc) {
+	s.cleanupFunc = fn
+}
+
+// StorePattern stores a new pattern.
+func (s *PatternStore) StorePattern(ctx context.Context, pattern *models.Pattern) (int64, error) {
+	signatureJSON, _ := json.Marshal(pattern.Signature)
+	projectsJSON, _ := json.Marshal(pattern.Projects)
+	obsIDsJSON, _ := json.Marshal(pattern.ObservationIDs)
+
+	const query = `
+		INSERT INTO patterns
+		(name, type, description, signature, recommendation, frequency, projects,
+		 observation_ids, status, merged_into_id, confidence,
+		 last_seen_at, last_seen_at_epoch, created_at, created_at_epoch)
+		VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+	`
+
+	result, err := s.store.ExecContext(ctx, query,
+		pattern.Name, string(pattern.Type),
+		nullString(pattern.Description.String), string(signatureJSON),
+		nullString(pattern.Recommendation.String),
+		pattern.Frequency, string(projectsJSON), string(obsIDsJSON),
+		string(pattern.Status), nullInt64(pattern.MergedIntoID),
+		pattern.Confidence, pattern.LastSeenAt, pattern.LastSeenEpoch,
+		pattern.CreatedAt, pattern.CreatedAtEpoch,
+	)
+	if err != nil {
+		return 0, err
+	}
+
+	return result.LastInsertId()
+}
+
+// UpdatePattern updates an existing pattern.
+func (s *PatternStore) UpdatePattern(ctx context.Context, pattern *models.Pattern) error {
+	signatureJSON, _ := json.Marshal(pattern.Signature)
+	projectsJSON, _ := json.Marshal(pattern.Projects)
+	obsIDsJSON, _ := json.Marshal(pattern.ObservationIDs)
+
+	const query = `
+		UPDATE patterns SET
+			name = ?, type = ?, description = ?, signature = ?, recommendation = ?,
+			frequency = ?, projects = ?, observation_ids = ?, status = ?,
+			merged_into_id = ?, confidence = ?, last_seen_at = ?, last_seen_at_epoch = ?
+		WHERE id = ?
+	`
+
+	_, err := s.store.ExecContext(ctx, query,
+		pattern.Name, string(pattern.Type),
+		nullString(pattern.Description.String), string(signatureJSON),
+		nullString(pattern.Recommendation.String),
+		pattern.Frequency, string(projectsJSON), string(obsIDsJSON),
+		string(pattern.Status), nullInt64(pattern.MergedIntoID),
+		pattern.Confidence, pattern.LastSeenAt, pattern.LastSeenEpoch,
+		pattern.ID,
+	)
+	return err
+}
+
+// GetPatternByID retrieves a pattern by ID.
+func (s *PatternStore) GetPatternByID(ctx context.Context, id int64) (*models.Pattern, error) {
+	query := `SELECT ` + patternColumns + ` FROM patterns WHERE id = ?`
+
+	row := s.store.QueryRowContext(ctx, query, id)
+	return scanPattern(row)
+}
+
+// GetPatternByName retrieves a pattern by name.
+func (s *PatternStore) GetPatternByName(ctx context.Context, name string) (*models.Pattern, error) {
+	query := `SELECT ` + patternColumns + ` FROM patterns WHERE name = ? AND status = 'active'`
+
+	row := s.store.QueryRowContext(ctx, query, name)
+	pattern, err := scanPattern(row)
+	if err == sql.ErrNoRows {
+		return nil, nil
+	}
+	return pattern, err
+}
+
+// GetActivePatterns retrieves all active patterns.
+func (s *PatternStore) GetActivePatterns(ctx context.Context, limit int) ([]*models.Pattern, error) {
+	query := `SELECT ` + patternColumns + `
+		FROM patterns
+		WHERE status = 'active'
+		ORDER BY frequency DESC, confidence DESC
+		LIMIT ?`
+
+	rows, err := s.store.QueryContext(ctx, query, limit)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return scanPatternRows(rows)
+}
+
+// GetPatternsByType retrieves patterns of a specific type.
+func (s *PatternStore) GetPatternsByType(ctx context.Context, patternType models.PatternType, limit int) ([]*models.Pattern, error) {
+	query := `SELECT ` + patternColumns + `
+		FROM patterns
+		WHERE type = ? AND status = 'active'
+		ORDER BY frequency DESC, confidence DESC
+		LIMIT ?`
+
+	rows, err := s.store.QueryContext(ctx, query, string(patternType), limit)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return scanPatternRows(rows)
+}
+
+// GetPatternsByProject retrieves patterns that have been observed in a specific project.
+func (s *PatternStore) GetPatternsByProject(ctx context.Context, project string, limit int) ([]*models.Pattern, error) {
+	// Use JSON path to search within the projects array
+	query := `SELECT ` + patternColumns + `
+		FROM patterns
+		WHERE status = 'active'
+		AND EXISTS (
+			SELECT 1 FROM json_each(projects)
+			WHERE json_each.value = ?
+		)
+		ORDER BY frequency DESC, confidence DESC
+		LIMIT ?`
+
+	rows, err := s.store.QueryContext(ctx, query, project, limit)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return scanPatternRows(rows)
+}
+
+// FindMatchingPatterns searches for patterns that match a given signature.
+func (s *PatternStore) FindMatchingPatterns(ctx context.Context, signature []string, minScore float64) ([]*models.Pattern, error) {
+	// Get all active patterns and filter by signature match in Go
+	// This is simpler than complex SQL for JSON array matching
+	patterns, err := s.GetActivePatterns(ctx, 100)
+	if err != nil {
+		return nil, err
+	}
+
+	var matches []*models.Pattern
+	for _, pattern := range patterns {
+		score := models.CalculateMatchScore(signature, pattern.Signature)
+		if score >= minScore {
+			matches = append(matches, pattern)
+		}
+	}
+	return matches, nil
+}
+
+// MarkPatternDeprecated marks a pattern as deprecated.
+func (s *PatternStore) MarkPatternDeprecated(ctx context.Context, id int64) error {
+	const query = `UPDATE patterns SET status = 'deprecated' WHERE id = ?`
+	_, err := s.store.ExecContext(ctx, query, id)
+	return err
+}
+
+// MergePatterns merges a source pattern into a target pattern.
+func (s *PatternStore) MergePatterns(ctx context.Context, sourceID, targetID int64) error {
+	// Get both patterns
+	source, err := s.GetPatternByID(ctx, sourceID)
+	if err != nil {
+		return err
+	}
+	target, err := s.GetPatternByID(ctx, targetID)
+	if err != nil {
+		return err
+	}
+
+	// Merge source into target
+	target.Frequency += source.Frequency
+	for _, proj := range source.Projects {
+		found := false
+		for _, existing := range target.Projects {
+			if existing == proj {
+				found = true
+				break
+			}
+		}
+		if !found {
+			target.Projects = append(target.Projects, proj)
+		}
+	}
+	for _, obsID := range source.ObservationIDs {
+		found := false
+		for _, existing := range target.ObservationIDs {
+			if existing == obsID {
+				found = true
+				break
+			}
+		}
+		if !found {
+			target.ObservationIDs = append(target.ObservationIDs, obsID)
+		}
+	}
+
+	// Update target
+	if err := s.UpdatePattern(ctx, target); err != nil {
+		return err
+	}
+
+	// Mark source as merged
+	source.Status = models.PatternStatusMerged
+	source.MergedIntoID = sql.NullInt64{Int64: targetID, Valid: true}
+	return s.UpdatePattern(ctx, source)
+}
+
+// DeletePattern deletes a pattern by ID.
+func (s *PatternStore) DeletePattern(ctx context.Context, id int64) error {
+	const query = `DELETE FROM patterns WHERE id = ?`
+	_, err := s.store.ExecContext(ctx, query, id)
+	if err == nil && s.cleanupFunc != nil {
+		s.cleanupFunc(ctx, []int64{id})
+	}
+	return err
+}
+
+// SearchPatternsFTS performs full-text search on patterns.
+func (s *PatternStore) SearchPatternsFTS(ctx context.Context, searchQuery string, limit int) ([]*models.Pattern, error) {
+	query := `SELECT p.` + patternColumns + `
+		FROM patterns p
+		JOIN patterns_fts fts ON p.id = fts.rowid
+		WHERE patterns_fts MATCH ?
+		AND p.status = 'active'
+		ORDER BY rank
+		LIMIT ?`
+
+	rows, err := s.store.QueryContext(ctx, query, searchQuery, limit)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return scanPatternRows(rows)
+}
+
+// GetPatternStats returns statistics about patterns.
+func (s *PatternStore) GetPatternStats(ctx context.Context) (*PatternStats, error) {
+	const query = `
+		SELECT
+			COUNT(*) as total,
+			COUNT(CASE WHEN status = 'active' THEN 1 END) as active,
+			COUNT(CASE WHEN status = 'deprecated' THEN 1 END) as deprecated,
+			COUNT(CASE WHEN status = 'merged' THEN 1 END) as merged,
+			COALESCE(SUM(frequency), 0) as total_occurrences,
+			COALESCE(AVG(confidence), 0) as avg_confidence,
+			COUNT(CASE WHEN type = 'bug' THEN 1 END) as bugs,
+			COUNT(CASE WHEN type = 'refactor' THEN 1 END) as refactors,
+			COUNT(CASE WHEN type = 'architecture' THEN 1 END) as architectures,
+			COUNT(CASE WHEN type = 'anti-pattern' THEN 1 END) as anti_patterns,
+			COUNT(CASE WHEN type = 'best-practice' THEN 1 END) as best_practices
+		FROM patterns
+	`
+
+	var stats PatternStats
+	err := s.store.QueryRowContext(ctx, query).Scan(
+		&stats.Total, &stats.Active, &stats.Deprecated, &stats.Merged,
+		&stats.TotalOccurrences, &stats.AvgConfidence,
+		&stats.Bugs, &stats.Refactors, &stats.Architectures,
+		&stats.AntiPatterns, &stats.BestPractices,
+	)
+	return &stats, err
+}
+
+// PatternStats contains aggregate statistics about patterns.
+type PatternStats struct {
+	Total            int     `json:"total"`
+	Active           int     `json:"active"`
+	Deprecated       int     `json:"deprecated"`
+	Merged           int     `json:"merged"`
+	TotalOccurrences int     `json:"total_occurrences"`
+	AvgConfidence    float64 `json:"avg_confidence"`
+	Bugs             int     `json:"bugs"`
+	Refactors        int     `json:"refactors"`
+	Architectures    int     `json:"architectures"`
+	AntiPatterns     int     `json:"anti_patterns"`
+	BestPractices    int     `json:"best_practices"`
+}
+
+// scanPattern scans a single pattern from a row scanner.
+func scanPattern(scanner interface{ Scan(...interface{}) error }) (*models.Pattern, error) {
+	var pattern models.Pattern
+	if err := scanner.Scan(
+		&pattern.ID, &pattern.Name, &pattern.Type,
+		&pattern.Description, &pattern.Signature, &pattern.Recommendation,
+		&pattern.Frequency, &pattern.Projects, &pattern.ObservationIDs,
+		&pattern.Status, &pattern.MergedIntoID, &pattern.Confidence,
+		&pattern.LastSeenAt, &pattern.LastSeenEpoch,
+		&pattern.CreatedAt, &pattern.CreatedAtEpoch,
+	); err != nil {
+		return nil, err
+	}
+	return &pattern, nil
+}
+
+// scanPatternRows scans multiple patterns from rows.
+func scanPatternRows(rows *sql.Rows) ([]*models.Pattern, error) {
+	var patterns []*models.Pattern
+	for rows.Next() {
+		pattern, err := scanPattern(rows)
+		if err != nil {
+			return nil, err
+		}
+		patterns = append(patterns, pattern)
+	}
+	return patterns, rows.Err()
+}
+
+// nullInt64 converts sql.NullInt64 to the value needed for database insertion.
+func nullInt64(n sql.NullInt64) interface{} {
+	if n.Valid {
+		return n.Int64
+	}
+	return nil
+}
+
+// IncrementPatternFrequency atomically increments a pattern's frequency and updates last_seen.
+func (s *PatternStore) IncrementPatternFrequency(ctx context.Context, id int64, project string, observationID int64) error {
+	now := time.Now()
+
+	// Get current pattern
+	pattern, err := s.GetPatternByID(ctx, id)
+	if err != nil {
+		return err
+	}
+
+	// Add occurrence
+	pattern.AddOccurrence(project, observationID)
+	pattern.LastSeenAt = now.Format(time.RFC3339)
+	pattern.LastSeenEpoch = now.UnixMilli()
+
+	return s.UpdatePattern(ctx, pattern)
+}

internal/db/sqlite/pattern_test.go

@@ -0,0 +1,507 @@
+package sqlite
+
+import (
+	"context"
+	"database/sql"
+	"testing"
+	"time"
+
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+)
+
+// setupPatternTestStore creates a test store with patterns table.
+func setupPatternTestStore(t *testing.T) *Store {
+	t.Helper()
+	db, _, cleanup := testDB(t)
+	t.Cleanup(cleanup)
+	createBaseTables(t, db)
+	return newStoreFromDB(db)
+}
+
+func TestPatternStore_StoreAndGet(t *testing.T) {
+	store := setupPatternTestStore(t)
+
+	patternStore := NewPatternStore(store)
+	ctx := context.Background()
+
+	// Create test pattern
+	pattern := &models.Pattern{
+		Name:           "Test Pattern",
+		Type:           models.PatternTypeBug,
+		Description:    sql.NullString{String: "A test pattern", Valid: true},
+		Signature:      []string{"nil", "error"},
+		Recommendation: sql.NullString{String: "Always check for nil", Valid: true},
+		Frequency:      1,
+		Projects:       []string{"project1"},
+		ObservationIDs: []int64{1, 2},
+		Status:         models.PatternStatusActive,
+		Confidence:     0.5,
+		LastSeenAt:     time.Now().Format(time.RFC3339),
+		LastSeenEpoch:  time.Now().UnixMilli(),
+		CreatedAt:      time.Now().Format(time.RFC3339),
+		CreatedAtEpoch: time.Now().UnixMilli(),
+	}
+
+	// Store pattern
+	id, err := patternStore.StorePattern(ctx, pattern)
+	if err != nil {
+		t.Fatalf("StorePattern() error = %v", err)
+	}
+	if id <= 0 {
+		t.Errorf("Expected positive ID, got %d", id)
+	}
+
+	// Get pattern by ID
+	retrieved, err := patternStore.GetPatternByID(ctx, id)
+	if err != nil {
+		t.Fatalf("GetPatternByID() error = %v", err)
+	}
+
+	if retrieved.Name != pattern.Name {
+		t.Errorf("Expected name %s, got %s", pattern.Name, retrieved.Name)
+	}
+	if retrieved.Type != pattern.Type {
+		t.Errorf("Expected type %s, got %s", pattern.Type, retrieved.Type)
+	}
+	if len(retrieved.Signature) != len(pattern.Signature) {
+		t.Errorf("Expected %d signature elements, got %d",
+			len(pattern.Signature), len(retrieved.Signature))
+	}
+}
+
+func TestPatternStore_GetByName(t *testing.T) {
+	store := setupPatternTestStore(t)
+
+	patternStore := NewPatternStore(store)
+	ctx := context.Background()
+
+	pattern := createTestPattern("Unique Name Pattern")
+	_, err := patternStore.StorePattern(ctx, pattern)
+	if err != nil {
+		t.Fatalf("StorePattern() error = %v", err)
+	}
+
+	// Get by name
+	retrieved, err := patternStore.GetPatternByName(ctx, "Unique Name Pattern")
+	if err != nil {
+		t.Fatalf("GetPatternByName() error = %v", err)
+	}
+	if retrieved == nil {
+		t.Fatal("Expected pattern, got nil")
+	}
+	if retrieved.Name != "Unique Name Pattern" {
+		t.Errorf("Expected name 'Unique Name Pattern', got '%s'", retrieved.Name)
+	}
+
+	// Get non-existent pattern
+	nonExistent, err := patternStore.GetPatternByName(ctx, "Non Existent")
+	if err != nil {
+		t.Fatalf("GetPatternByName() error = %v", err)
+	}
+	if nonExistent != nil {
+		t.Errorf("Expected nil for non-existent pattern, got %v", nonExistent)
+	}
+}
+
+func TestPatternStore_GetActivePatterns(t *testing.T) {
+	store := setupPatternTestStore(t)
+
+	patternStore := NewPatternStore(store)
+	ctx := context.Background()
+
+	// Create multiple patterns with different statuses
+	active1 := createTestPattern("Active 1")
+	active1.Frequency = 5
+	active2 := createTestPattern("Active 2")
+	active2.Frequency = 3
+	deprecated := createTestPattern("Deprecated")
+	deprecated.Status = models.PatternStatusDeprecated
+
+	patternStore.StorePattern(ctx, active1)
+	patternStore.StorePattern(ctx, active2)
+	patternStore.StorePattern(ctx, deprecated)
+
+	// Get active patterns
+	patterns, err := patternStore.GetActivePatterns(ctx, 10)
+	if err != nil {
+		t.Fatalf("GetActivePatterns() error = %v", err)
+	}
+
+	if len(patterns) != 2 {
+		t.Errorf("Expected 2 active patterns, got %d", len(patterns))
+	}
+
+	// Check order (should be by frequency descending)
+	if len(patterns) >= 2 {
+		if patterns[0].Frequency < patterns[1].Frequency {
+			t.Errorf("Patterns not ordered by frequency descending")
+		}
+	}
+}
+
+func TestPatternStore_GetPatternsByType(t *testing.T) {
+	store := setupPatternTestStore(t)
+
+	patternStore := NewPatternStore(store)
+	ctx := context.Background()
+
+	// Create patterns of different types
+	bugPattern := createTestPattern("Bug Pattern")
+	bugPattern.Type = models.PatternTypeBug
+
+	refactorPattern := createTestPattern("Refactor Pattern")
+	refactorPattern.Type = models.PatternTypeRefactor
+
+	patternStore.StorePattern(ctx, bugPattern)
+	patternStore.StorePattern(ctx, refactorPattern)
+
+	// Get by type
+	bugs, err := patternStore.GetPatternsByType(ctx, models.PatternTypeBug, 10)
+	if err != nil {
+		t.Fatalf("GetPatternsByType() error = %v", err)
+	}
+	if len(bugs) != 1 {
+		t.Errorf("Expected 1 bug pattern, got %d", len(bugs))
+	}
+
+	refactors, err := patternStore.GetPatternsByType(ctx, models.PatternTypeRefactor, 10)
+	if err != nil {
+		t.Fatalf("GetPatternsByType() error = %v", err)
+	}
+	if len(refactors) != 1 {
+		t.Errorf("Expected 1 refactor pattern, got %d", len(refactors))
+	}
+}
+
+func TestPatternStore_GetPatternsByProject(t *testing.T) {
+	store := setupPatternTestStore(t)
+
+	patternStore := NewPatternStore(store)
+	ctx := context.Background()
+
+	// Create patterns with different projects
+	pattern1 := createTestPattern("Pattern 1")
+	pattern1.Projects = []string{"project-a", "project-b"}
+
+	pattern2 := createTestPattern("Pattern 2")
+	pattern2.Projects = []string{"project-b", "project-c"}
+
+	patternStore.StorePattern(ctx, pattern1)
+	patternStore.StorePattern(ctx, pattern2)
+
+	// Get by project
+	projectA, err := patternStore.GetPatternsByProject(ctx, "project-a", 10)
+	if err != nil {
+		t.Fatalf("GetPatternsByProject() error = %v", err)
+	}
+	if len(projectA) != 1 {
+		t.Errorf("Expected 1 pattern for project-a, got %d", len(projectA))
+	}
+
+	projectB, err := patternStore.GetPatternsByProject(ctx, "project-b", 10)
+	if err != nil {
+		t.Fatalf("GetPatternsByProject() error = %v", err)
+	}
+	if len(projectB) != 2 {
+		t.Errorf("Expected 2 patterns for project-b, got %d", len(projectB))
+	}
+}
+
+func TestPatternStore_UpdatePattern(t *testing.T) {
+	store := setupPatternTestStore(t)
+
+	patternStore := NewPatternStore(store)
+	ctx := context.Background()
+
+	// Create and store pattern
+	pattern := createTestPattern("Original Name")
+	id, _ := patternStore.StorePattern(ctx, pattern)
+
+	// Update pattern
+	pattern.ID = id
+	pattern.Name = "Updated Name"
+	pattern.Frequency = 10
+	pattern.Confidence = 0.9
+
+	err := patternStore.UpdatePattern(ctx, pattern)
+	if err != nil {
+		t.Fatalf("UpdatePattern() error = %v", err)
+	}
+
+	// Verify update
+	updated, _ := patternStore.GetPatternByID(ctx, id)
+	if updated.Name != "Updated Name" {
+		t.Errorf("Expected name 'Updated Name', got '%s'", updated.Name)
+	}
+	if updated.Frequency != 10 {
+		t.Errorf("Expected frequency 10, got %d", updated.Frequency)
+	}
+	if updated.Confidence != 0.9 {
+		t.Errorf("Expected confidence 0.9, got %f", updated.Confidence)
+	}
+}
+
+func TestPatternStore_DeletePattern(t *testing.T) {
+	store := setupPatternTestStore(t)
+
+	patternStore := NewPatternStore(store)
+	ctx := context.Background()
+
+	// Create and store pattern
+	pattern := createTestPattern("To Delete")
+	id, _ := patternStore.StorePattern(ctx, pattern)
+
+	// Delete pattern
+	err := patternStore.DeletePattern(ctx, id)
+	if err != nil {
+		t.Fatalf("DeletePattern() error = %v", err)
+	}
+
+	// Verify deletion
+	deleted, err := patternStore.GetPatternByID(ctx, id)
+	if err != sql.ErrNoRows {
+		t.Errorf("Expected ErrNoRows, got %v", err)
+	}
+	if deleted != nil {
+		t.Errorf("Expected nil for deleted pattern")
+	}
+}
+
+func TestPatternStore_MarkPatternDeprecated(t *testing.T) {
+	store := setupPatternTestStore(t)
+
+	patternStore := NewPatternStore(store)
+	ctx := context.Background()
+
+	// Create and store pattern
+	pattern := createTestPattern("To Deprecate")
+	id, _ := patternStore.StorePattern(ctx, pattern)
+
+	// Mark as deprecated
+	err := patternStore.MarkPatternDeprecated(ctx, id)
+	if err != nil {
+		t.Fatalf("MarkPatternDeprecated() error = %v", err)
+	}
+
+	// Verify status
+	deprecated, _ := patternStore.GetPatternByID(ctx, id)
+	if deprecated.Status != models.PatternStatusDeprecated {
+		t.Errorf("Expected status 'deprecated', got '%s'", deprecated.Status)
+	}
+}
+
+func TestPatternStore_MergePatterns(t *testing.T) {
+	store := setupPatternTestStore(t)
+
+	patternStore := NewPatternStore(store)
+	ctx := context.Background()
+
+	// Create source and target patterns
+	source := createTestPattern("Source Pattern")
+	source.Frequency = 3
+	source.Projects = []string{"proj1", "proj2"}
+	source.ObservationIDs = []int64{1, 2, 3}
+
+	target := createTestPattern("Target Pattern")
+	target.Frequency = 2
+	target.Projects = []string{"proj2", "proj3"}
+	target.ObservationIDs = []int64{4, 5}
+
+	sourceID, _ := patternStore.StorePattern(ctx, source)
+	targetID, _ := patternStore.StorePattern(ctx, target)
+
+	// Merge
+	err := patternStore.MergePatterns(ctx, sourceID, targetID)
+	if err != nil {
+		t.Fatalf("MergePatterns() error = %v", err)
+	}
+
+	// Verify source is marked as merged
+	mergedSource, _ := patternStore.GetPatternByID(ctx, sourceID)
+	if mergedSource.Status != models.PatternStatusMerged {
+		t.Errorf("Expected source status 'merged', got '%s'", mergedSource.Status)
+	}
+	if !mergedSource.MergedIntoID.Valid || mergedSource.MergedIntoID.Int64 != targetID {
+		t.Errorf("Expected source merged_into_id to be %d", targetID)
+	}
+
+	// Verify target has combined data
+	mergedTarget, _ := patternStore.GetPatternByID(ctx, targetID)
+	expectedFrequency := 5 // 3 + 2
+	if mergedTarget.Frequency != expectedFrequency {
+		t.Errorf("Expected merged frequency %d, got %d", expectedFrequency, mergedTarget.Frequency)
+	}
+	// Should have 3 unique projects: proj1, proj2, proj3
+	if len(mergedTarget.Projects) != 3 {
+		t.Errorf("Expected 3 projects after merge, got %d", len(mergedTarget.Projects))
+	}
+	// Should have 5 observation IDs
+	if len(mergedTarget.ObservationIDs) != 5 {
+		t.Errorf("Expected 5 observation IDs after merge, got %d", len(mergedTarget.ObservationIDs))
+	}
+}
+
+func TestPatternStore_FindMatchingPatterns(t *testing.T) {
+	store := setupPatternTestStore(t)
+
+	patternStore := NewPatternStore(store)
+	ctx := context.Background()
+
+	// Create patterns with known signatures
+	pattern1 := createTestPattern("Pattern 1")
+	pattern1.Signature = []string{"nil", "error", "handling"}
+
+	pattern2 := createTestPattern("Pattern 2")
+	pattern2.Signature = []string{"nil", "pointer", "check"}
+
+	pattern3 := createTestPattern("Pattern 3")
+	pattern3.Signature = []string{"refactor", "extract", "method"}
+
+	patternStore.StorePattern(ctx, pattern1)
+	patternStore.StorePattern(ctx, pattern2)
+	patternStore.StorePattern(ctx, pattern3)
+
+	// Find patterns matching "nil" related signature
+	matches, err := patternStore.FindMatchingPatterns(ctx, []string{"nil", "error"}, 0.3)
+	if err != nil {
+		t.Fatalf("FindMatchingPatterns() error = %v", err)
+	}
+
+	if len(matches) < 1 {
+		t.Errorf("Expected at least 1 match, got %d", len(matches))
+	}
+
+	// Verify no match for unrelated signature
+	noMatches, err := patternStore.FindMatchingPatterns(ctx, []string{"completely", "different"}, 0.5)
+	if err != nil {
+		t.Fatalf("FindMatchingPatterns() error = %v", err)
+	}
+	if len(noMatches) != 0 {
+		t.Errorf("Expected 0 matches for unrelated signature, got %d", len(noMatches))
+	}
+}
+
+func TestPatternStore_IncrementPatternFrequency(t *testing.T) {
+	store := setupPatternTestStore(t)
+
+	patternStore := NewPatternStore(store)
+	ctx := context.Background()
+
+	// Create pattern
+	pattern := createTestPattern("Frequency Test")
+	pattern.Frequency = 1
+	pattern.Projects = []string{"proj1"}
+	pattern.ObservationIDs = []int64{1}
+
+	id, _ := patternStore.StorePattern(ctx, pattern)
+
+	// Increment frequency
+	err := patternStore.IncrementPatternFrequency(ctx, id, "proj2", 2)
+	if err != nil {
+		t.Fatalf("IncrementPatternFrequency() error = %v", err)
+	}
+
+	// Verify
+	updated, _ := patternStore.GetPatternByID(ctx, id)
+	if updated.Frequency != 2 {
+		t.Errorf("Expected frequency 2, got %d", updated.Frequency)
+	}
+	if len(updated.Projects) != 2 {
+		t.Errorf("Expected 2 projects, got %d", len(updated.Projects))
+	}
+	if len(updated.ObservationIDs) != 2 {
+		t.Errorf("Expected 2 observation IDs, got %d", len(updated.ObservationIDs))
+	}
+}
+
+func TestPatternStore_GetPatternStats(t *testing.T) {
+	store := setupPatternTestStore(t)
+
+	patternStore := NewPatternStore(store)
+	ctx := context.Background()
+
+	// Create patterns with different types and statuses
+	bug := createTestPattern("Bug")
+	bug.Type = models.PatternTypeBug
+	bug.Frequency = 5
+
+	refactor := createTestPattern("Refactor")
+	refactor.Type = models.PatternTypeRefactor
+	refactor.Frequency = 3
+
+	deprecated := createTestPattern("Deprecated")
+	deprecated.Type = models.PatternTypeArchitecture
+	deprecated.Status = models.PatternStatusDeprecated
+
+	patternStore.StorePattern(ctx, bug)
+	patternStore.StorePattern(ctx, refactor)
+	patternStore.StorePattern(ctx, deprecated)
+
+	// Get stats
+	stats, err := patternStore.GetPatternStats(ctx)
+	if err != nil {
+		t.Fatalf("GetPatternStats() error = %v", err)
+	}
+
+	if stats.Total != 3 {
+		t.Errorf("Expected total 3, got %d", stats.Total)
+	}
+	if stats.Active != 2 {
+		t.Errorf("Expected 2 active, got %d", stats.Active)
+	}
+	if stats.Deprecated != 1 {
+		t.Errorf("Expected 1 deprecated, got %d", stats.Deprecated)
+	}
+	if stats.Bugs != 1 {
+		t.Errorf("Expected 1 bug, got %d", stats.Bugs)
+	}
+	if stats.Refactors != 1 {
+		t.Errorf("Expected 1 refactor, got %d", stats.Refactors)
+	}
+	if stats.TotalOccurrences != 9 { // 5 + 3 + 1
+		t.Errorf("Expected 9 total occurrences, got %d", stats.TotalOccurrences)
+	}
+}
+
+func TestPatternStore_CleanupCallback(t *testing.T) {
+	store := setupPatternTestStore(t)
+
+	patternStore := NewPatternStore(store)
+	ctx := context.Background()
+
+	var deletedIDs []int64
+	patternStore.SetCleanupFunc(func(ctx context.Context, ids []int64) {
+		deletedIDs = ids
+	})
+
+	// Create and delete pattern
+	pattern := createTestPattern("Cleanup Test")
+	id, _ := patternStore.StorePattern(ctx, pattern)
+
+	patternStore.DeletePattern(ctx, id)
+
+	if len(deletedIDs) != 1 || deletedIDs[0] != id {
+		t.Errorf("Expected cleanup callback with ID %d, got %v", id, deletedIDs)
+	}
+}
+
+// Helper function to create a test pattern
+func createTestPattern(name string) *models.Pattern {
+	now := time.Now()
+	return &models.Pattern{
+		Name:           name,
+		Type:           models.PatternTypeBug,
+		Description:    sql.NullString{String: "Test description", Valid: true},
+		Signature:      []string{"test", "pattern"},
+		Recommendation: sql.NullString{String: "Test recommendation", Valid: true},
+		Frequency:      1,
+		Projects:       []string{"test-project"},
+		ObservationIDs: []int64{1},
+		Status:         models.PatternStatusActive,
+		Confidence:     0.5,
+		LastSeenAt:     now.Format(time.RFC3339),
+		LastSeenEpoch:  now.UnixMilli(),
+		CreatedAt:      now.Format(time.RFC3339),
+		CreatedAtEpoch: now.UnixMilli(),
+	}
+}

internal/db/sqlite/prompt.go

@@ -102,13 +102,13 @@ func (s *PromptStore) CleanupOldPrompts(ctx context.Context) ([]int64, error) {
 	var toDelete []int64
 	for rows.Next() {
 		var id int64
-		if err := rows.Scan(&id); err != nil {
-			return nil, err
+		if scanErr := rows.Scan(&id); scanErr != nil {
+			return nil, scanErr
 		}
 		toDelete = append(toDelete, id)
 	}
-	if err := rows.Err(); err != nil {
-		return nil, err
+	if rowsErr := rows.Err(); rowsErr != nil {
+		return nil, rowsErr
 	}
 
 	if len(toDelete) == 0 {

internal/db/sqlite/prompt_test.go

@@ -261,14 +261,14 @@ func TestPromptStore_SaveMultiplePrompts(t *testing.T) {
 
 	tests := []struct {
 		claudeSessionID string
-		promptNum       int
 		text            string
+		promptNum       int
 		matches         int
 	}{
-		{"claude-1", 1, "First prompt", 5},
-		{"claude-1", 2, "Second prompt", 3},
-		{"claude-2", 1, "Third prompt", 0},
-		{"claude-1", 3, "Fourth prompt", 10},
+		{claudeSessionID: "claude-1", promptNum: 1, text: "First prompt", matches: 5},
+		{claudeSessionID: "claude-1", promptNum: 2, text: "Second prompt", matches: 3},
+		{claudeSessionID: "claude-2", promptNum: 1, text: "Third prompt", matches: 0},
+		{claudeSessionID: "claude-1", promptNum: 3, text: "Fourth prompt", matches: 10},
 	}
 
 	for _, tt := range tests {

internal/db/sqlite/relation.go

@@ -0,0 +1,377 @@
+// Package sqlite provides SQLite database operations for claude-mnemonic.
+package sqlite
+
+import (
+	"context"
+	"database/sql"
+
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+)
+
+// RelationStore provides relation-related database operations.
+type RelationStore struct {
+	store *Store
+}
+
+// NewRelationStore creates a new relation store.
+func NewRelationStore(store *Store) *RelationStore {
+	return &RelationStore{store: store}
+}
+
+// StoreRelation stores a new observation relation.
+// Uses INSERT OR IGNORE to handle duplicate (source_id, target_id, relation_type) combinations.
+func (s *RelationStore) StoreRelation(ctx context.Context, relation *models.ObservationRelation) (int64, error) {
+	const query = `
+		INSERT OR IGNORE INTO observation_relations
+		(source_id, target_id, relation_type, confidence, detection_source, reason, created_at, created_at_epoch)
+		VALUES (?, ?, ?, ?, ?, ?, ?, ?)
+	`
+
+	result, err := s.store.ExecContext(ctx, query,
+		relation.SourceID, relation.TargetID,
+		string(relation.RelationType), relation.Confidence,
+		string(relation.DetectionSource), relation.Reason,
+		relation.CreatedAt, relation.CreatedAtEpoch,
+	)
+	if err != nil {
+		return 0, err
+	}
+
+	return result.LastInsertId()
+}
+
+// StoreRelations stores multiple relations in a single transaction.
+func (s *RelationStore) StoreRelations(ctx context.Context, relations []*models.ObservationRelation) error {
+	if len(relations) == 0 {
+		return nil
+	}
+
+	tx, err := s.store.db.BeginTx(ctx, nil)
+	if err != nil {
+		return err
+	}
+	defer func() {
+		if err != nil {
+			_ = tx.Rollback()
+		}
+	}()
+
+	const query = `
+		INSERT OR IGNORE INTO observation_relations
+		(source_id, target_id, relation_type, confidence, detection_source, reason, created_at, created_at_epoch)
+		VALUES (?, ?, ?, ?, ?, ?, ?, ?)
+	`
+
+	stmt, err := tx.PrepareContext(ctx, query)
+	if err != nil {
+		return err
+	}
+	defer stmt.Close()
+
+	for _, rel := range relations {
+		_, err = stmt.ExecContext(ctx,
+			rel.SourceID, rel.TargetID,
+			string(rel.RelationType), rel.Confidence,
+			string(rel.DetectionSource), rel.Reason,
+			rel.CreatedAt, rel.CreatedAtEpoch,
+		)
+		if err != nil {
+			return err
+		}
+	}
+
+	return tx.Commit()
+}
+
+// GetRelationsByObservationID retrieves all relations involving an observation (as source or target).
+func (s *RelationStore) GetRelationsByObservationID(ctx context.Context, obsID int64) ([]*models.ObservationRelation, error) {
+	const query = `
+		SELECT id, source_id, target_id, relation_type, confidence, detection_source, reason,
+		       created_at, created_at_epoch
+		FROM observation_relations
+		WHERE source_id = ? OR target_id = ?
+		ORDER BY confidence DESC, created_at_epoch DESC
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, obsID, obsID)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return s.scanRelationRows(rows)
+}
+
+// GetOutgoingRelations retrieves relations where the observation is the source.
+func (s *RelationStore) GetOutgoingRelations(ctx context.Context, obsID int64) ([]*models.ObservationRelation, error) {
+	const query = `
+		SELECT id, source_id, target_id, relation_type, confidence, detection_source, reason,
+		       created_at, created_at_epoch
+		FROM observation_relations
+		WHERE source_id = ?
+		ORDER BY confidence DESC, created_at_epoch DESC
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, obsID)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return s.scanRelationRows(rows)
+}
+
+// GetIncomingRelations retrieves relations where the observation is the target.
+func (s *RelationStore) GetIncomingRelations(ctx context.Context, obsID int64) ([]*models.ObservationRelation, error) {
+	const query = `
+		SELECT id, source_id, target_id, relation_type, confidence, detection_source, reason,
+		       created_at, created_at_epoch
+		FROM observation_relations
+		WHERE target_id = ?
+		ORDER BY confidence DESC, created_at_epoch DESC
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, obsID)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return s.scanRelationRows(rows)
+}
+
+// GetRelationsByType retrieves all relations of a specific type.
+func (s *RelationStore) GetRelationsByType(ctx context.Context, relationType models.RelationType, limit int) ([]*models.ObservationRelation, error) {
+	const query = `
+		SELECT id, source_id, target_id, relation_type, confidence, detection_source, reason,
+		       created_at, created_at_epoch
+		FROM observation_relations
+		WHERE relation_type = ?
+		ORDER BY confidence DESC, created_at_epoch DESC
+		LIMIT ?
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, string(relationType), limit)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return s.scanRelationRows(rows)
+}
+
+// GetRelationsWithDetails retrieves relations with observation titles for display.
+func (s *RelationStore) GetRelationsWithDetails(ctx context.Context, obsID int64) ([]*models.RelationWithDetails, error) {
+	const query = `
+		SELECT r.id, r.source_id, r.target_id, r.relation_type, r.confidence, r.detection_source, r.reason,
+		       r.created_at, r.created_at_epoch,
+		       COALESCE(src.title, '') as source_title,
+		       COALESCE(tgt.title, '') as target_title,
+		       src.type as source_type,
+		       tgt.type as target_type
+		FROM observation_relations r
+		JOIN observations src ON src.id = r.source_id
+		JOIN observations tgt ON tgt.id = r.target_id
+		WHERE r.source_id = ? OR r.target_id = ?
+		ORDER BY r.confidence DESC, r.created_at_epoch DESC
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, obsID, obsID)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	var results []*models.RelationWithDetails
+	for rows.Next() {
+		var r models.ObservationRelation
+		var rwd models.RelationWithDetails
+		var reason sql.NullString
+		if err := rows.Scan(
+			&r.ID, &r.SourceID, &r.TargetID,
+			&r.RelationType, &r.Confidence, &r.DetectionSource, &reason,
+			&r.CreatedAt, &r.CreatedAtEpoch,
+			&rwd.SourceTitle, &rwd.TargetTitle,
+			&rwd.SourceType, &rwd.TargetType,
+		); err != nil {
+			return nil, err
+		}
+		if reason.Valid {
+			r.Reason = reason.String
+		}
+		rwd.Relation = &r
+		results = append(results, &rwd)
+	}
+	return results, rows.Err()
+}
+
+// GetRelationGraph retrieves a relation graph centered on an observation.
+// This returns all observations within N hops from the center.
+func (s *RelationStore) GetRelationGraph(ctx context.Context, centerID int64, maxDepth int) (*models.RelationGraph, error) {
+	// Get all relations involving the center observation
+	relations, err := s.GetRelationsWithDetails(ctx, centerID)
+	if err != nil {
+		return nil, err
+	}
+
+	graph := &models.RelationGraph{
+		CenterID:  centerID,
+		Relations: relations,
+	}
+
+	// If depth > 1, recursively get relations for connected observations
+	if maxDepth > 1 {
+		visited := map[int64]bool{centerID: true}
+		toVisit := make([]int64, 0)
+
+		// Collect IDs of directly connected observations
+		for _, r := range relations {
+			if !visited[r.Relation.SourceID] {
+				toVisit = append(toVisit, r.Relation.SourceID)
+				visited[r.Relation.SourceID] = true
+			}
+			if !visited[r.Relation.TargetID] {
+				toVisit = append(toVisit, r.Relation.TargetID)
+				visited[r.Relation.TargetID] = true
+			}
+		}
+
+		// Get relations for connected observations (depth - 1)
+		for depth := 1; depth < maxDepth && len(toVisit) > 0; depth++ {
+			nextLevel := make([]int64, 0)
+			for _, obsID := range toVisit {
+				moreRelations, err := s.GetRelationsWithDetails(ctx, obsID)
+				if err != nil {
+					continue
+				}
+				for _, r := range moreRelations {
+					// Avoid duplicates
+					exists := false
+					for _, existing := range graph.Relations {
+						if existing.Relation.ID == r.Relation.ID {
+							exists = true
+							break
+						}
+					}
+					if !exists {
+						graph.Relations = append(graph.Relations, r)
+					}
+
+					// Queue next level
+					if !visited[r.Relation.SourceID] {
+						nextLevel = append(nextLevel, r.Relation.SourceID)
+						visited[r.Relation.SourceID] = true
+					}
+					if !visited[r.Relation.TargetID] {
+						nextLevel = append(nextLevel, r.Relation.TargetID)
+						visited[r.Relation.TargetID] = true
+					}
+				}
+			}
+			toVisit = nextLevel
+		}
+	}
+
+	return graph, nil
+}
+
+// DeleteRelationsByObservationID deletes all relations involving an observation.
+// Called when an observation is deleted.
+func (s *RelationStore) DeleteRelationsByObservationID(ctx context.Context, obsID int64) error {
+	const query = `DELETE FROM observation_relations WHERE source_id = ? OR target_id = ?`
+	_, err := s.store.ExecContext(ctx, query, obsID, obsID)
+	return err
+}
+
+// GetRelationCount returns the count of relations for an observation.
+func (s *RelationStore) GetRelationCount(ctx context.Context, obsID int64) (int, error) {
+	const query = `
+		SELECT COUNT(*) FROM observation_relations
+		WHERE source_id = ? OR target_id = ?
+	`
+	var count int
+	err := s.store.QueryRowContext(ctx, query, obsID, obsID).Scan(&count)
+	return count, err
+}
+
+// GetTotalRelationCount returns the total count of all relations.
+func (s *RelationStore) GetTotalRelationCount(ctx context.Context) (int, error) {
+	const query = `SELECT COUNT(*) FROM observation_relations`
+	var count int
+	err := s.store.QueryRowContext(ctx, query).Scan(&count)
+	return count, err
+}
+
+// GetHighConfidenceRelations retrieves relations with confidence above threshold.
+func (s *RelationStore) GetHighConfidenceRelations(ctx context.Context, minConfidence float64, limit int) ([]*models.ObservationRelation, error) {
+	const query = `
+		SELECT id, source_id, target_id, relation_type, confidence, detection_source, reason,
+		       created_at, created_at_epoch
+		FROM observation_relations
+		WHERE confidence >= ?
+		ORDER BY confidence DESC, created_at_epoch DESC
+		LIMIT ?
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, minConfidence, limit)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return s.scanRelationRows(rows)
+}
+
+// UpdateRelationConfidence updates the confidence of a relation.
+func (s *RelationStore) UpdateRelationConfidence(ctx context.Context, relationID int64, newConfidence float64) error {
+	const query = `UPDATE observation_relations SET confidence = ? WHERE id = ?`
+	_, err := s.store.ExecContext(ctx, query, newConfidence, relationID)
+	return err
+}
+
+// GetRelatedObservationIDs returns IDs of observations related to the given one.
+// This is useful for expanding search results.
+func (s *RelationStore) GetRelatedObservationIDs(ctx context.Context, obsID int64, minConfidence float64) ([]int64, error) {
+	const query = `
+		SELECT DISTINCT CASE WHEN source_id = ? THEN target_id ELSE source_id END as related_id
+		FROM observation_relations
+		WHERE (source_id = ? OR target_id = ?) AND confidence >= ?
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, obsID, obsID, obsID, minConfidence)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	var ids []int64
+	for rows.Next() {
+		var id int64
+		if err := rows.Scan(&id); err != nil {
+			return nil, err
+		}
+		ids = append(ids, id)
+	}
+	return ids, rows.Err()
+}
+
+// scanRelationRows scans multiple relations from rows.
+func (s *RelationStore) scanRelationRows(rows *sql.Rows) ([]*models.ObservationRelation, error) {
+	var relations []*models.ObservationRelation
+	for rows.Next() {
+		var r models.ObservationRelation
+		var reason sql.NullString
+		if err := rows.Scan(
+			&r.ID, &r.SourceID, &r.TargetID,
+			&r.RelationType, &r.Confidence, &r.DetectionSource, &reason,
+			&r.CreatedAt, &r.CreatedAtEpoch,
+		); err != nil {
+			return nil, err
+		}
+		if reason.Valid {
+			r.Reason = reason.String
+		}
+		relations = append(relations, &r)
+	}
+	return relations, rows.Err()
+}

internal/db/sqlite/scoring.go

@@ -0,0 +1,324 @@
+// Package sqlite provides SQLite database operations for claude-mnemonic.
+package sqlite
+
+import (
+	"context"
+	"database/sql"
+	"time"
+
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+)
+
+// UpdateObservationFeedback updates the user feedback for an observation.
+// Feedback values: -1 (thumbs down), 0 (neutral), 1 (thumbs up).
+func (s *ObservationStore) UpdateObservationFeedback(ctx context.Context, id int64, feedback int) error {
+	const query = `
+		UPDATE observations
+		SET user_feedback = ?, score_updated_at_epoch = ?
+		WHERE id = ?
+	`
+	_, err := s.store.ExecContext(ctx, query, feedback, time.Now().UnixMilli(), id)
+	return err
+}
+
+// IncrementRetrievalCount increments the retrieval counter for the given observation IDs.
+// This is called when observations are returned in search results.
+func (s *ObservationStore) IncrementRetrievalCount(ctx context.Context, ids []int64) error {
+	if len(ids) == 0 {
+		return nil
+	}
+
+	now := time.Now().UnixMilli()
+
+	// Build query with placeholders
+	// #nosec G202 -- query uses parameterized placeholders, not user input
+	query := `
+		UPDATE observations
+		SET retrieval_count = COALESCE(retrieval_count, 0) + 1,
+		    last_retrieved_at_epoch = ?
+		WHERE id IN (?` + repeatPlaceholders(len(ids)-1) + `)
+	`
+
+	args := make([]interface{}, 0, len(ids)+1)
+	args = append(args, now)
+	for _, id := range ids {
+		args = append(args, id)
+	}
+
+	_, err := s.store.db.ExecContext(ctx, query, args...)
+	return err
+}
+
+// UpdateImportanceScore updates the importance score for a single observation.
+func (s *ObservationStore) UpdateImportanceScore(ctx context.Context, id int64, score float64) error {
+	const query = `
+		UPDATE observations
+		SET importance_score = ?, score_updated_at_epoch = ?
+		WHERE id = ?
+	`
+	_, err := s.store.ExecContext(ctx, query, score, time.Now().UnixMilli(), id)
+	return err
+}
+
+// UpdateImportanceScores bulk updates importance scores for multiple observations.
+// This is more efficient than individual updates for batch recalculation.
+func (s *ObservationStore) UpdateImportanceScores(ctx context.Context, scores map[int64]float64) error {
+	if len(scores) == 0 {
+		return nil
+	}
+
+	tx, err := s.store.db.BeginTx(ctx, nil)
+	if err != nil {
+		return err
+	}
+	defer func() { _ = tx.Rollback() }()
+
+	now := time.Now().UnixMilli()
+	stmt, err := tx.PrepareContext(ctx, `
+		UPDATE observations
+		SET importance_score = ?, score_updated_at_epoch = ?
+		WHERE id = ?
+	`)
+	if err != nil {
+		return err
+	}
+	defer stmt.Close()
+
+	for id, score := range scores {
+		if _, err := stmt.ExecContext(ctx, score, now, id); err != nil {
+			return err
+		}
+	}
+
+	return tx.Commit()
+}
+
+// GetObservationsNeedingScoreUpdate returns observations that need their importance score recalculated.
+// Returns observations where score_updated_at_epoch is NULL or older than the threshold.
+func (s *ObservationStore) GetObservationsNeedingScoreUpdate(ctx context.Context, threshold time.Duration, limit int) ([]*models.Observation, error) {
+	cutoff := time.Now().Add(-threshold).UnixMilli()
+
+	query := `SELECT ` + observationColumns + `
+		FROM observations
+		WHERE score_updated_at_epoch IS NULL OR score_updated_at_epoch < ?
+		ORDER BY created_at_epoch DESC
+		LIMIT ?
+	`
+
+	rows, err := s.store.QueryContext(ctx, query, cutoff, limit)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return scanObservationRows(rows)
+}
+
+// GetConceptWeights returns all concept weights from the database.
+func (s *ObservationStore) GetConceptWeights(ctx context.Context) (map[string]float64, error) {
+	const query = `SELECT concept, weight FROM concept_weights`
+
+	rows, err := s.store.QueryContext(ctx, query)
+	if err != nil {
+		// Table might not exist in older databases
+		if err == sql.ErrNoRows {
+			return models.DefaultConceptWeights, nil
+		}
+		return nil, err
+	}
+	defer rows.Close()
+
+	weights := make(map[string]float64)
+	for rows.Next() {
+		var concept string
+		var weight float64
+		if err := rows.Scan(&concept, &weight); err != nil {
+			return nil, err
+		}
+		weights[concept] = weight
+	}
+
+	if err := rows.Err(); err != nil {
+		return nil, err
+	}
+
+	// If no weights found, use defaults
+	if len(weights) == 0 {
+		return models.DefaultConceptWeights, nil
+	}
+
+	return weights, nil
+}
+
+// UpdateConceptWeight updates a single concept weight.
+func (s *ObservationStore) UpdateConceptWeight(ctx context.Context, concept string, weight float64) error {
+	const query = `
+		INSERT INTO concept_weights (concept, weight, updated_at)
+		VALUES (?, ?, datetime('now'))
+		ON CONFLICT(concept) DO UPDATE SET weight = excluded.weight, updated_at = excluded.updated_at
+	`
+	_, err := s.store.ExecContext(ctx, query, concept, weight)
+	return err
+}
+
+// UpdateConceptWeights bulk updates multiple concept weights.
+func (s *ObservationStore) UpdateConceptWeights(ctx context.Context, weights map[string]float64) error {
+	if len(weights) == 0 {
+		return nil
+	}
+
+	tx, err := s.store.db.BeginTx(ctx, nil)
+	if err != nil {
+		return err
+	}
+	defer func() { _ = tx.Rollback() }()
+
+	stmt, err := tx.PrepareContext(ctx, `
+		INSERT INTO concept_weights (concept, weight, updated_at)
+		VALUES (?, ?, datetime('now'))
+		ON CONFLICT(concept) DO UPDATE SET weight = excluded.weight, updated_at = excluded.updated_at
+	`)
+	if err != nil {
+		return err
+	}
+	defer stmt.Close()
+
+	for concept, weight := range weights {
+		if _, err := stmt.ExecContext(ctx, concept, weight); err != nil {
+			return err
+		}
+	}
+
+	return tx.Commit()
+}
+
+// GetObservationFeedbackStats returns statistics about user feedback.
+func (s *ObservationStore) GetObservationFeedbackStats(ctx context.Context, project string) (*FeedbackStats, error) {
+	var query string
+	var args []interface{}
+
+	if project == "" {
+		query = `
+			SELECT
+				COUNT(*) as total,
+				COALESCE(SUM(CASE WHEN user_feedback = 1 THEN 1 ELSE 0 END), 0) as positive,
+				COALESCE(SUM(CASE WHEN user_feedback = -1 THEN 1 ELSE 0 END), 0) as negative,
+				COALESCE(SUM(CASE WHEN user_feedback = 0 THEN 1 ELSE 0 END), 0) as neutral,
+				COALESCE(AVG(COALESCE(importance_score, 1.0)), 0) as avg_score,
+				COALESCE(AVG(COALESCE(retrieval_count, 0)), 0) as avg_retrieval
+			FROM observations
+		`
+	} else {
+		query = `
+			SELECT
+				COUNT(*) as total,
+				COALESCE(SUM(CASE WHEN user_feedback = 1 THEN 1 ELSE 0 END), 0) as positive,
+				COALESCE(SUM(CASE WHEN user_feedback = -1 THEN 1 ELSE 0 END), 0) as negative,
+				COALESCE(SUM(CASE WHEN user_feedback = 0 THEN 1 ELSE 0 END), 0) as neutral,
+				COALESCE(AVG(COALESCE(importance_score, 1.0)), 0) as avg_score,
+				COALESCE(AVG(COALESCE(retrieval_count, 0)), 0) as avg_retrieval
+			FROM observations
+			WHERE project = ? OR scope = 'global'
+		`
+		args = append(args, project)
+	}
+
+	var stats FeedbackStats
+	err := s.store.QueryRowContext(ctx, query, args...).Scan(
+		&stats.Total,
+		&stats.Positive,
+		&stats.Negative,
+		&stats.Neutral,
+		&stats.AvgScore,
+		&stats.AvgRetrieval,
+	)
+	if err != nil {
+		return nil, err
+	}
+
+	return &stats, nil
+}
+
+// FeedbackStats contains statistics about observation feedback and scoring.
+type FeedbackStats struct {
+	Total        int     `json:"total"`
+	Positive     int     `json:"positive"`
+	Negative     int     `json:"negative"`
+	Neutral      int     `json:"neutral"`
+	AvgScore     float64 `json:"avg_score"`
+	AvgRetrieval float64 `json:"avg_retrieval"`
+}
+
+// GetTopScoringObservations returns the highest-scoring observations.
+func (s *ObservationStore) GetTopScoringObservations(ctx context.Context, project string, limit int) ([]*models.Observation, error) {
+	var query string
+	var args []interface{}
+
+	if project == "" {
+		query = `SELECT ` + observationColumns + `
+			FROM observations
+			ORDER BY COALESCE(importance_score, 1.0) DESC, created_at_epoch DESC
+			LIMIT ?
+		`
+		args = append(args, limit)
+	} else {
+		query = `SELECT ` + observationColumns + `
+			FROM observations
+			WHERE project = ? OR scope = 'global'
+			ORDER BY COALESCE(importance_score, 1.0) DESC, created_at_epoch DESC
+			LIMIT ?
+		`
+		args = append(args, project, limit)
+	}
+
+	rows, err := s.store.QueryContext(ctx, query, args...)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return scanObservationRows(rows)
+}
+
+// GetMostRetrievedObservations returns the most frequently retrieved observations.
+func (s *ObservationStore) GetMostRetrievedObservations(ctx context.Context, project string, limit int) ([]*models.Observation, error) {
+	var query string
+	var args []interface{}
+
+	if project == "" {
+		query = `SELECT ` + observationColumns + `
+			FROM observations
+			WHERE retrieval_count > 0
+			ORDER BY retrieval_count DESC, created_at_epoch DESC
+			LIMIT ?
+		`
+		args = append(args, limit)
+	} else {
+		query = `SELECT ` + observationColumns + `
+			FROM observations
+			WHERE (project = ? OR scope = 'global') AND retrieval_count > 0
+			ORDER BY retrieval_count DESC, created_at_epoch DESC
+			LIMIT ?
+		`
+		args = append(args, project, limit)
+	}
+
+	rows, err := s.store.QueryContext(ctx, query, args...)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return scanObservationRows(rows)
+}
+
+// ResetObservationScores resets all observation scores to their default values.
+// This is useful for testing or when changing the scoring algorithm.
+func (s *ObservationStore) ResetObservationScores(ctx context.Context) error {
+	const query = `
+		UPDATE observations
+		SET importance_score = 1.0, score_updated_at_epoch = NULL
+	`
+	_, err := s.store.ExecContext(ctx, query)
+	return err
+}

internal/db/sqlite/scoring_test.go

@@ -0,0 +1,698 @@
+// Package sqlite provides SQLite database operations for claude-mnemonic.
+package sqlite
+
+import (
+	"context"
+	"database/sql"
+	"testing"
+	"time"
+
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"github.com/stretchr/testify/suite"
+)
+
+// testScoringObservationStore creates an ObservationStore with scoring columns for testing.
+func testScoringObservationStore(t *testing.T) (*ObservationStore, *Store, func()) {
+	t.Helper()
+
+	db, _, cleanup := testDB(t)
+	createBaseTables(t, db)
+	createConceptWeightsTable(t, db)
+
+	// Add importance index if not exists (columns already in createBaseTables)
+	if _, err := db.Exec(`CREATE INDEX IF NOT EXISTS idx_observations_importance ON observations(importance_score DESC, created_at_epoch DESC)`); err != nil {
+		t.Fatalf("create importance index: %v", err)
+	}
+
+	store := newStoreFromDB(db)
+	obsStore := NewObservationStore(store)
+
+	return obsStore, store, cleanup
+}
+
+// createConceptWeightsTable creates the concept_weights table for testing.
+func createConceptWeightsTable(t *testing.T, db *sql.DB) {
+	t.Helper()
+
+	_, err := db.Exec(`
+		CREATE TABLE IF NOT EXISTS concept_weights (
+			concept TEXT PRIMARY KEY,
+			weight REAL NOT NULL DEFAULT 0.1,
+			updated_at TEXT NOT NULL
+		)
+	`)
+	if err != nil {
+		t.Fatalf("create concept_weights: %v", err)
+	}
+}
+
+// ScoringStoreSuite is a test suite for scoring-related database operations.
+type ScoringStoreSuite struct {
+	suite.Suite
+	obsStore *ObservationStore
+	store    *Store
+	cleanup  func()
+	ctx      context.Context
+}
+
+func (s *ScoringStoreSuite) SetupTest() {
+	s.obsStore, s.store, s.cleanup = testScoringObservationStore(s.T())
+	s.ctx = context.Background()
+}
+
+func (s *ScoringStoreSuite) TearDownTest() {
+	if s.cleanup != nil {
+		s.cleanup()
+	}
+}
+
+func TestScoringStoreSuite(t *testing.T) {
+	suite.Run(t, new(ScoringStoreSuite))
+}
+
+// =============================================================================
+// FEEDBACK TESTS
+// =============================================================================
+
+func (s *ScoringStoreSuite) TestUpdateObservationFeedback_Positive() {
+	// Create observation
+	obs := &models.ParsedObservation{
+		Type:  models.ObsTypeBugfix,
+		Title: "Test feedback",
+	}
+	id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, 1, 100)
+	s.NoError(err)
+
+	// Update feedback to positive
+	err = s.obsStore.UpdateObservationFeedback(s.ctx, id, 1)
+	s.NoError(err)
+
+	// Verify
+	retrieved, err := s.obsStore.GetObservationByID(s.ctx, id)
+	s.NoError(err)
+	s.Equal(1, retrieved.UserFeedback)
+	s.True(retrieved.ScoreUpdatedAt.Valid)
+}
+
+func (s *ScoringStoreSuite) TestUpdateObservationFeedback_Negative() {
+	obs := &models.ParsedObservation{
+		Type: models.ObsTypeDiscovery,
+	}
+	id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, 1, 100)
+	s.NoError(err)
+
+	err = s.obsStore.UpdateObservationFeedback(s.ctx, id, -1)
+	s.NoError(err)
+
+	retrieved, err := s.obsStore.GetObservationByID(s.ctx, id)
+	s.NoError(err)
+	s.Equal(-1, retrieved.UserFeedback)
+}
+
+func (s *ScoringStoreSuite) TestUpdateObservationFeedback_Neutral() {
+	obs := &models.ParsedObservation{
+		Type: models.ObsTypeChange,
+	}
+	id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, 1, 100)
+	s.NoError(err)
+
+	// First set to positive
+	err = s.obsStore.UpdateObservationFeedback(s.ctx, id, 1)
+	s.NoError(err)
+
+	// Then reset to neutral
+	err = s.obsStore.UpdateObservationFeedback(s.ctx, id, 0)
+	s.NoError(err)
+
+	retrieved, err := s.obsStore.GetObservationByID(s.ctx, id)
+	s.NoError(err)
+	s.Equal(0, retrieved.UserFeedback)
+}
+
+func (s *ScoringStoreSuite) TestUpdateObservationFeedback_NonExistent() {
+	// Updating non-existent observation should not fail (just no rows affected)
+	err := s.obsStore.UpdateObservationFeedback(s.ctx, 99999, 1)
+	s.NoError(err)
+}
+
+// =============================================================================
+// RETRIEVAL COUNT TESTS
+// =============================================================================
+
+func (s *ScoringStoreSuite) TestIncrementRetrievalCount_Single() {
+	obs := &models.ParsedObservation{
+		Type: models.ObsTypeBugfix,
+	}
+	id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, 1, 100)
+	s.NoError(err)
+
+	err = s.obsStore.IncrementRetrievalCount(s.ctx, []int64{id})
+	s.NoError(err)
+
+	retrieved, err := s.obsStore.GetObservationByID(s.ctx, id)
+	s.NoError(err)
+	s.Equal(1, retrieved.RetrievalCount)
+	s.True(retrieved.LastRetrievedAt.Valid)
+}
+
+func (s *ScoringStoreSuite) TestIncrementRetrievalCount_Multiple() {
+	var ids []int64
+	for i := 0; i < 3; i++ {
+		obs := &models.ParsedObservation{
+			Type: models.ObsTypeDiscovery,
+		}
+		id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, i, 100)
+		s.NoError(err)
+		ids = append(ids, id)
+	}
+
+	err := s.obsStore.IncrementRetrievalCount(s.ctx, ids)
+	s.NoError(err)
+
+	for _, id := range ids {
+		retrieved, err := s.obsStore.GetObservationByID(s.ctx, id)
+		s.NoError(err)
+		s.Equal(1, retrieved.RetrievalCount)
+	}
+}
+
+func (s *ScoringStoreSuite) TestIncrementRetrievalCount_Cumulative() {
+	obs := &models.ParsedObservation{
+		Type: models.ObsTypeBugfix,
+	}
+	id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, 1, 100)
+	s.NoError(err)
+
+	// Increment multiple times
+	for i := 0; i < 5; i++ {
+		err = s.obsStore.IncrementRetrievalCount(s.ctx, []int64{id})
+		s.NoError(err)
+	}
+
+	retrieved, err := s.obsStore.GetObservationByID(s.ctx, id)
+	s.NoError(err)
+	s.Equal(5, retrieved.RetrievalCount)
+}
+
+func (s *ScoringStoreSuite) TestIncrementRetrievalCount_Empty() {
+	err := s.obsStore.IncrementRetrievalCount(s.ctx, []int64{})
+	s.NoError(err)
+}
+
+// =============================================================================
+// IMPORTANCE SCORE TESTS
+// =============================================================================
+
+func (s *ScoringStoreSuite) TestUpdateImportanceScore_Single() {
+	obs := &models.ParsedObservation{
+		Type: models.ObsTypeBugfix,
+	}
+	id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, 1, 100)
+	s.NoError(err)
+
+	err = s.obsStore.UpdateImportanceScore(s.ctx, id, 1.5)
+	s.NoError(err)
+
+	retrieved, err := s.obsStore.GetObservationByID(s.ctx, id)
+	s.NoError(err)
+	s.InDelta(1.5, retrieved.ImportanceScore, 0.001)
+}
+
+func (s *ScoringStoreSuite) TestUpdateImportanceScores_Batch() {
+	var ids []int64
+	for i := 0; i < 5; i++ {
+		obs := &models.ParsedObservation{
+			Type: models.ObsTypeDiscovery,
+		}
+		id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, i, 100)
+		s.NoError(err)
+		ids = append(ids, id)
+	}
+
+	scores := map[int64]float64{
+		ids[0]: 1.5,
+		ids[1]: 0.8,
+		ids[2]: 1.2,
+		ids[3]: 0.5,
+		ids[4]: 2.0,
+	}
+
+	err := s.obsStore.UpdateImportanceScores(s.ctx, scores)
+	s.NoError(err)
+
+	for id, expectedScore := range scores {
+		retrieved, err := s.obsStore.GetObservationByID(s.ctx, id)
+		s.NoError(err)
+		s.InDelta(expectedScore, retrieved.ImportanceScore, 0.001)
+	}
+}
+
+func (s *ScoringStoreSuite) TestUpdateImportanceScores_Empty() {
+	err := s.obsStore.UpdateImportanceScores(s.ctx, map[int64]float64{})
+	s.NoError(err)
+}
+
+// =============================================================================
+// OBSERVATIONS NEEDING SCORE UPDATE TESTS
+// =============================================================================
+
+func (s *ScoringStoreSuite) TestGetObservationsNeedingScoreUpdate_NeverUpdated() {
+	// Observations without score_updated_at_epoch should need update
+	obs := &models.ParsedObservation{
+		Type: models.ObsTypeBugfix,
+	}
+	id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, 1, 100)
+	s.NoError(err)
+
+	observations, err := s.obsStore.GetObservationsNeedingScoreUpdate(s.ctx, 6*time.Hour, 100)
+	s.NoError(err)
+	s.Len(observations, 1)
+	s.Equal(id, observations[0].ID)
+}
+
+func (s *ScoringStoreSuite) TestGetObservationsNeedingScoreUpdate_RecentlyUpdated() {
+	obs := &models.ParsedObservation{
+		Type: models.ObsTypeBugfix,
+	}
+	id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, 1, 100)
+	s.NoError(err)
+
+	// Update score (this sets score_updated_at_epoch)
+	err = s.obsStore.UpdateImportanceScore(s.ctx, id, 1.5)
+	s.NoError(err)
+
+	// Should not need update (just updated)
+	observations, err := s.obsStore.GetObservationsNeedingScoreUpdate(s.ctx, 6*time.Hour, 100)
+	s.NoError(err)
+	s.Empty(observations)
+}
+
+func (s *ScoringStoreSuite) TestGetObservationsNeedingScoreUpdate_Limit() {
+	// Create 10 observations
+	for i := 0; i < 10; i++ {
+		obs := &models.ParsedObservation{
+			Type: models.ObsTypeDiscovery,
+		}
+		_, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, i, 100)
+		s.NoError(err)
+	}
+
+	// Request only 5
+	observations, err := s.obsStore.GetObservationsNeedingScoreUpdate(s.ctx, 6*time.Hour, 5)
+	s.NoError(err)
+	s.Len(observations, 5)
+}
+
+// =============================================================================
+// CONCEPT WEIGHTS TESTS
+// =============================================================================
+
+func (s *ScoringStoreSuite) TestGetConceptWeights_Empty() {
+	weights, err := s.obsStore.GetConceptWeights(s.ctx)
+	s.NoError(err)
+	s.Equal(models.DefaultConceptWeights, weights)
+}
+
+func (s *ScoringStoreSuite) TestUpdateConceptWeight_NewConcept() {
+	err := s.obsStore.UpdateConceptWeight(s.ctx, "new-concept", 0.42)
+	s.NoError(err)
+
+	weights, err := s.obsStore.GetConceptWeights(s.ctx)
+	s.NoError(err)
+	s.Equal(0.42, weights["new-concept"])
+}
+
+func (s *ScoringStoreSuite) TestUpdateConceptWeight_UpdateExisting() {
+	// Insert first
+	err := s.obsStore.UpdateConceptWeight(s.ctx, "test-concept", 0.1)
+	s.NoError(err)
+
+	// Update
+	err = s.obsStore.UpdateConceptWeight(s.ctx, "test-concept", 0.9)
+	s.NoError(err)
+
+	weights, err := s.obsStore.GetConceptWeights(s.ctx)
+	s.NoError(err)
+	s.Equal(0.9, weights["test-concept"])
+}
+
+func (s *ScoringStoreSuite) TestUpdateConceptWeights_Batch() {
+	weightsToSet := map[string]float64{
+		"security":    0.5,
+		"performance": 0.3,
+		"testing":     0.2,
+	}
+
+	err := s.obsStore.UpdateConceptWeights(s.ctx, weightsToSet)
+	s.NoError(err)
+
+	retrieved, err := s.obsStore.GetConceptWeights(s.ctx)
+	s.NoError(err)
+
+	for concept, expected := range weightsToSet {
+		s.Equal(expected, retrieved[concept])
+	}
+}
+
+func (s *ScoringStoreSuite) TestUpdateConceptWeights_Empty() {
+	err := s.obsStore.UpdateConceptWeights(s.ctx, map[string]float64{})
+	s.NoError(err)
+}
+
+// =============================================================================
+// FEEDBACK STATS TESTS
+// =============================================================================
+
+func (s *ScoringStoreSuite) TestGetObservationFeedbackStats_Empty() {
+	stats, err := s.obsStore.GetObservationFeedbackStats(s.ctx, "")
+	s.NoError(err)
+	s.Equal(0, stats.Total)
+	s.Equal(0, stats.Positive)
+	s.Equal(0, stats.Negative)
+	s.Equal(0, stats.Neutral)
+}
+
+func (s *ScoringStoreSuite) TestGetObservationFeedbackStats_WithData() {
+	// Create observations with different feedback
+	feedbacks := []int{1, 1, 1, -1, -1, 0, 0, 0, 0, 0}
+	for i, fb := range feedbacks {
+		obs := &models.ParsedObservation{
+			Type: models.ObsTypeDiscovery,
+		}
+		id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, i, 100)
+		s.NoError(err)
+		if fb != 0 {
+			err = s.obsStore.UpdateObservationFeedback(s.ctx, id, fb)
+			s.NoError(err)
+		}
+	}
+
+	stats, err := s.obsStore.GetObservationFeedbackStats(s.ctx, "")
+	s.NoError(err)
+	s.Equal(10, stats.Total)
+	s.Equal(3, stats.Positive)
+	s.Equal(2, stats.Negative)
+	s.Equal(5, stats.Neutral)
+}
+
+func (s *ScoringStoreSuite) TestGetObservationFeedbackStats_ByProject() {
+	// Project A observations
+	for i := 0; i < 5; i++ {
+		obs := &models.ParsedObservation{
+			Type: models.ObsTypeBugfix,
+		}
+		id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, i, 100)
+		s.NoError(err)
+		_ = s.obsStore.UpdateObservationFeedback(s.ctx, id, 1)
+	}
+
+	// Project B observations
+	for i := 0; i < 3; i++ {
+		obs := &models.ParsedObservation{
+			Type: models.ObsTypeFeature,
+		}
+		id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-b", obs, i, 100)
+		s.NoError(err)
+		_ = s.obsStore.UpdateObservationFeedback(s.ctx, id, -1)
+	}
+
+	// Check project A stats
+	statsA, err := s.obsStore.GetObservationFeedbackStats(s.ctx, "project-a")
+	s.NoError(err)
+	s.Equal(5, statsA.Total)
+	s.Equal(5, statsA.Positive)
+
+	// Check project B stats
+	statsB, err := s.obsStore.GetObservationFeedbackStats(s.ctx, "project-b")
+	s.NoError(err)
+	s.Equal(3, statsB.Total)
+	s.Equal(3, statsB.Negative)
+}
+
+// =============================================================================
+// TOP SCORING OBSERVATIONS TESTS
+// =============================================================================
+
+func (s *ScoringStoreSuite) TestGetTopScoringObservations() {
+	// Create observations with different scores
+	for i := 0; i < 5; i++ {
+		obs := &models.ParsedObservation{
+			Type: models.ObsTypeDiscovery,
+		}
+		id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, i, 100)
+		s.NoError(err)
+		// Set different scores
+		err = s.obsStore.UpdateImportanceScore(s.ctx, id, float64(i+1)*0.5)
+		s.NoError(err)
+	}
+
+	// Get top 3
+	top, err := s.obsStore.GetTopScoringObservations(s.ctx, "", 3)
+	s.NoError(err)
+	s.Len(top, 3)
+
+	// Verify ordered by score descending
+	s.GreaterOrEqual(top[0].ImportanceScore, top[1].ImportanceScore)
+	s.GreaterOrEqual(top[1].ImportanceScore, top[2].ImportanceScore)
+}
+
+func (s *ScoringStoreSuite) TestGetTopScoringObservations_ByProject() {
+	// Project A with high scores
+	for i := 0; i < 3; i++ {
+		obs := &models.ParsedObservation{
+			Type: models.ObsTypeBugfix,
+		}
+		id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, i, 100)
+		s.NoError(err)
+		_ = s.obsStore.UpdateImportanceScore(s.ctx, id, 2.0)
+	}
+
+	// Project B with low scores
+	for i := 0; i < 3; i++ {
+		obs := &models.ParsedObservation{
+			Type: models.ObsTypeChange,
+		}
+		id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-b", obs, i, 100)
+		s.NoError(err)
+		_ = s.obsStore.UpdateImportanceScore(s.ctx, id, 0.5)
+	}
+
+	// Get top for project A
+	topA, err := s.obsStore.GetTopScoringObservations(s.ctx, "project-a", 10)
+	s.NoError(err)
+	s.Len(topA, 3)
+	for _, obs := range topA {
+		s.Equal("project-a", obs.Project)
+	}
+}
+
+// =============================================================================
+// MOST RETRIEVED OBSERVATIONS TESTS
+// =============================================================================
+
+func (s *ScoringStoreSuite) TestGetMostRetrievedObservations() {
+	// Create observations with different retrieval counts
+	var ids []int64
+	for i := 0; i < 5; i++ {
+		obs := &models.ParsedObservation{
+			Type: models.ObsTypeDiscovery,
+		}
+		id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, i, 100)
+		s.NoError(err)
+		ids = append(ids, id)
+	}
+
+	// Set different retrieval counts
+	for i := 0; i < 10; i++ {
+		_ = s.obsStore.IncrementRetrievalCount(s.ctx, []int64{ids[0]}) // 10 retrievals
+	}
+	for i := 0; i < 5; i++ {
+		_ = s.obsStore.IncrementRetrievalCount(s.ctx, []int64{ids[1]}) // 5 retrievals
+	}
+	for i := 0; i < 3; i++ {
+		_ = s.obsStore.IncrementRetrievalCount(s.ctx, []int64{ids[2]}) // 3 retrievals
+	}
+	// ids[3] and ids[4] have 0 retrievals
+
+	// Get top 3
+	most, err := s.obsStore.GetMostRetrievedObservations(s.ctx, "", 3)
+	s.NoError(err)
+	s.Len(most, 3)
+
+	// Verify ordered by retrieval count descending
+	s.Equal(10, most[0].RetrievalCount)
+	s.Equal(5, most[1].RetrievalCount)
+	s.Equal(3, most[2].RetrievalCount)
+}
+
+func (s *ScoringStoreSuite) TestGetMostRetrievedObservations_NoRetrievals() {
+	// Create observations without any retrievals
+	obs := &models.ParsedObservation{
+		Type: models.ObsTypeDiscovery,
+	}
+	_, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, 1, 100)
+	s.NoError(err)
+
+	most, err := s.obsStore.GetMostRetrievedObservations(s.ctx, "", 10)
+	s.NoError(err)
+	s.Empty(most) // No observations with retrieval_count > 0
+}
+
+// =============================================================================
+// RESET OBSERVATION SCORES TESTS
+// =============================================================================
+
+func (s *ScoringStoreSuite) TestResetObservationScores() {
+	// Create observations with various scores
+	for i := 0; i < 5; i++ {
+		obs := &models.ParsedObservation{
+			Type: models.ObsTypeDiscovery,
+		}
+		id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, i, 100)
+		s.NoError(err)
+		_ = s.obsStore.UpdateImportanceScore(s.ctx, id, float64(i+1))
+	}
+
+	// Reset all scores
+	err := s.obsStore.ResetObservationScores(s.ctx)
+	s.NoError(err)
+
+	// Verify all scores are reset to 1.0
+	observations, err := s.obsStore.GetAllRecentObservations(s.ctx, 100)
+	s.NoError(err)
+	for _, obs := range observations {
+		s.InDelta(1.0, obs.ImportanceScore, 0.001)
+		s.False(obs.ScoreUpdatedAt.Valid)
+	}
+}
+
+// =============================================================================
+// EDGE CASES
+// =============================================================================
+
+func (s *ScoringStoreSuite) TestScoring_ZeroScore() {
+	obs := &models.ParsedObservation{
+		Type: models.ObsTypeBugfix,
+	}
+	id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, 1, 100)
+	s.NoError(err)
+
+	// Set score to 0
+	err = s.obsStore.UpdateImportanceScore(s.ctx, id, 0.0)
+	s.NoError(err)
+
+	retrieved, err := s.obsStore.GetObservationByID(s.ctx, id)
+	s.NoError(err)
+	s.InDelta(0.0, retrieved.ImportanceScore, 0.001)
+}
+
+func (s *ScoringStoreSuite) TestScoring_NegativeScore() {
+	obs := &models.ParsedObservation{
+		Type: models.ObsTypeBugfix,
+	}
+	id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, 1, 100)
+	s.NoError(err)
+
+	// Set negative score (calculator shouldn't produce this, but test DB handling)
+	err = s.obsStore.UpdateImportanceScore(s.ctx, id, -0.5)
+	s.NoError(err)
+
+	retrieved, err := s.obsStore.GetObservationByID(s.ctx, id)
+	s.NoError(err)
+	s.InDelta(-0.5, retrieved.ImportanceScore, 0.001)
+}
+
+func (s *ScoringStoreSuite) TestScoring_LargeScore() {
+	obs := &models.ParsedObservation{
+		Type: models.ObsTypeBugfix,
+	}
+	id, _, err := s.obsStore.StoreObservation(s.ctx, "session-1", "project-a", obs, 1, 100)
+	s.NoError(err)
+
+	// Set very large score
+	err = s.obsStore.UpdateImportanceScore(s.ctx, id, 999.999)
+	s.NoError(err)
+
+	retrieved, err := s.obsStore.GetObservationByID(s.ctx, id)
+	s.NoError(err)
+	s.InDelta(999.999, retrieved.ImportanceScore, 0.001)
+}
+
+func (s *ScoringStoreSuite) TestConceptWeight_ZeroWeight() {
+	err := s.obsStore.UpdateConceptWeight(s.ctx, "zero-concept", 0.0)
+	s.NoError(err)
+
+	weights, err := s.obsStore.GetConceptWeights(s.ctx)
+	s.NoError(err)
+	s.Equal(0.0, weights["zero-concept"])
+}
+
+func (s *ScoringStoreSuite) TestConceptWeight_ExactBoundary() {
+	err := s.obsStore.UpdateConceptWeight(s.ctx, "max-concept", 1.0)
+	s.NoError(err)
+
+	weights, err := s.obsStore.GetConceptWeights(s.ctx)
+	s.NoError(err)
+	s.Equal(1.0, weights["max-concept"])
+}
+
+// =============================================================================
+// STANDALONE TESTS
+// =============================================================================
+
+func TestFeedbackStats_Structure(t *testing.T) {
+	stats := FeedbackStats{
+		Total:        100,
+		Positive:     30,
+		Negative:     10,
+		Neutral:      60,
+		AvgScore:     1.5,
+		AvgRetrieval: 5.0,
+	}
+
+	assert.Equal(t, 100, stats.Total)
+	assert.Equal(t, 30, stats.Positive)
+	assert.Equal(t, 10, stats.Negative)
+	assert.Equal(t, 60, stats.Neutral)
+	assert.Equal(t, 1.5, stats.AvgScore)
+	assert.Equal(t, 5.0, stats.AvgRetrieval)
+}
+
+func TestScoringStore_Integration(t *testing.T) {
+	obsStore, _, cleanup := testScoringObservationStore(t)
+	defer cleanup()
+
+	ctx := context.Background()
+
+	// Full integration test: store, feedback, retrieval, score update
+	obs := &models.ParsedObservation{
+		Type:     models.ObsTypeBugfix,
+		Title:    "Integration test observation",
+		Concepts: []string{"security"},
+	}
+	id, _, err := obsStore.StoreObservation(ctx, "session-int", "project-int", obs, 1, 100)
+	require.NoError(t, err)
+
+	// Add feedback
+	err = obsStore.UpdateObservationFeedback(ctx, id, 1)
+	require.NoError(t, err)
+
+	// Increment retrieval
+	err = obsStore.IncrementRetrievalCount(ctx, []int64{id})
+	require.NoError(t, err)
+
+	// Update score
+	err = obsStore.UpdateImportanceScore(ctx, id, 1.75)
+	require.NoError(t, err)
+
+	// Verify final state
+	retrieved, err := obsStore.GetObservationByID(ctx, id)
+	require.NoError(t, err)
+	assert.Equal(t, 1, retrieved.UserFeedback)
+	assert.Equal(t, 1, retrieved.RetrievalCount)
+	assert.InDelta(t, 1.75, retrieved.ImportanceScore, 0.001)
+	assert.True(t, retrieved.ScoreUpdatedAt.Valid)
+	assert.True(t, retrieved.LastRetrievedAt.Valid)
+}

internal/db/sqlite/session_test.go

@@ -211,8 +211,8 @@ func (s *SessionStoreSuite) TestFindAnySDKSession_Scenarios() {
 	tests := []struct {
 		name            string
 		claudeSessionID string
-		wantFound       bool
 		wantProject     string
+		wantFound       bool
 	}{
 		{
 			name:            "find existing session 1",

internal/db/sqlite/store.go

@@ -94,8 +94,8 @@ func (s *Store) GetStmt(query string) (*sql.Stmt, error) {
 	defer s.stmtMu.Unlock()
 
 	// Double-check after acquiring write lock
-	if stmt, ok := s.stmtCache[query]; ok {
-		return stmt, nil
+	if cachedStmt, ok := s.stmtCache[query]; ok {
+		return cachedStmt, nil
 	}
 
 	stmt, err := s.db.Prepare(query)

internal/db/sqlite/store_test.go

@@ -130,13 +130,13 @@ func (s *StoreSuite) TestQueryContext() {
 	seedSession(s.T(), s.db, "claude-1", "sdk-1", "project-a")
 
 	tests := []struct {
+		setupFunc  func()
+		assertFunc func(rows *sql.Rows)
 		name       string
 		query      string
 		args       []interface{}
-		wantErr    bool
 		wantRows   int
-		setupFunc  func()
-		assertFunc func(rows *sql.Rows)
+		wantErr    bool
 	}{
 		{
 			name:     "query existing session",
@@ -366,8 +366,8 @@ func (s *HelpersSuite) TestNullInt() {
 func (s *HelpersSuite) TestRepeatPlaceholders() {
 	tests := []struct {
 		name     string
-		input    int
 		expected string
+		input    int
 	}{
 		{
 			name:     "zero",
@@ -438,10 +438,10 @@ func TestBuildGetByIDsQuery(t *testing.T) {
 	tests := []struct {
 		name      string
 		baseQuery string
-		ids       []int64
 		orderBy   string
-		limit     int
 		wantQuery string
+		ids       []int64
+		limit     int
 		wantArgs  int
 	}{
 		{
@@ -483,10 +483,10 @@ func TestEnsureSessionExists(t *testing.T) {
 	ctx := context.Background()
 
 	tests := []struct {
+		setup        func()
 		name         string
 		sdkSessionID string
 		project      string
-		setup        func()
 		wantErr      bool
 	}{
 		{

internal/db/sqlite/testhelpers_test.go

@@ -103,6 +103,12 @@ func createBaseTables(t *testing.T, db *sql.DB) {
 			discovery_tokens INTEGER DEFAULT 0,
 			created_at TEXT NOT NULL,
 			created_at_epoch INTEGER NOT NULL,
+			importance_score REAL DEFAULT 1.0,
+			user_feedback INTEGER DEFAULT 0,
+			retrieval_count INTEGER DEFAULT 0,
+			last_retrieved_at_epoch INTEGER,
+			score_updated_at_epoch INTEGER,
+			is_superseded INTEGER DEFAULT 0,
 			FOREIGN KEY(sdk_session_id) REFERENCES sdk_sessions(sdk_session_id) ON DELETE CASCADE
 		)
 	`)
@@ -110,6 +116,27 @@ func createBaseTables(t *testing.T, db *sql.DB) {
 		t.Fatalf("create observations: %v", err)
 	}
 
+	// Create observation_conflicts table for conflict detection
+	_, err = db.Exec(`
+		CREATE TABLE IF NOT EXISTS observation_conflicts (
+			id INTEGER PRIMARY KEY AUTOINCREMENT,
+			newer_obs_id INTEGER NOT NULL,
+			older_obs_id INTEGER NOT NULL,
+			conflict_type TEXT NOT NULL CHECK(conflict_type IN ('superseded', 'contradicts', 'outdated_pattern')),
+			resolution TEXT NOT NULL CHECK(resolution IN ('prefer_newer', 'prefer_older', 'manual')),
+			reason TEXT,
+			detected_at TEXT NOT NULL,
+			detected_at_epoch INTEGER NOT NULL,
+			resolved INTEGER DEFAULT 0,
+			resolved_at TEXT,
+			FOREIGN KEY(newer_obs_id) REFERENCES observations(id) ON DELETE CASCADE,
+			FOREIGN KEY(older_obs_id) REFERENCES observations(id) ON DELETE CASCADE
+		)
+	`)
+	if err != nil {
+		t.Fatalf("create observation_conflicts: %v", err)
+	}
+
 	_, err = db.Exec(`
 		CREATE TABLE IF NOT EXISTS session_summaries (
 			id INTEGER PRIMARY KEY AUTOINCREMENT,
@@ -150,6 +177,31 @@ func createBaseTables(t *testing.T, db *sql.DB) {
 		t.Fatalf("create user_prompts: %v", err)
 	}
 
+	_, err = db.Exec(`
+		CREATE TABLE IF NOT EXISTS patterns (
+			id INTEGER PRIMARY KEY AUTOINCREMENT,
+			name TEXT NOT NULL,
+			type TEXT NOT NULL CHECK(type IN ('bug', 'refactor', 'architecture', 'anti-pattern', 'best-practice')),
+			description TEXT,
+			signature TEXT,
+			recommendation TEXT,
+			frequency INTEGER DEFAULT 1,
+			projects TEXT,
+			observation_ids TEXT,
+			status TEXT DEFAULT 'active' CHECK(status IN ('active', 'deprecated', 'merged')),
+			merged_into_id INTEGER,
+			confidence REAL DEFAULT 0.5,
+			last_seen_at TEXT NOT NULL,
+			last_seen_at_epoch INTEGER NOT NULL,
+			created_at TEXT NOT NULL,
+			created_at_epoch INTEGER NOT NULL,
+			FOREIGN KEY(merged_into_id) REFERENCES patterns(id) ON DELETE SET NULL
+		)
+	`)
+	if err != nil {
+		t.Fatalf("create patterns: %v", err)
+	}
+
 	indexes := []string{
 		`CREATE INDEX IF NOT EXISTS idx_sdk_sessions_claude_id ON sdk_sessions(claude_session_id)`,
 		`CREATE INDEX IF NOT EXISTS idx_sdk_sessions_sdk_id ON sdk_sessions(sdk_session_id)`,

internal/embedding/model.go

@@ -21,15 +21,10 @@ const (
 // ONNXConfig describes ONNX-specific model configuration.
 // This allows different models to specify their tensor names and pooling needs.
 type ONNXConfig struct {
-	// InputNames are the ONNX input tensor names in order.
-	InputNames []string
-	// OutputNames are the ONNX output tensor names.
+	Pooling     PoolingStrategy
+	InputNames  []string
 	OutputNames []string
-	// Pooling specifies how to convert token embeddings to sentence embeddings.
-	// If PoolingNone, the model outputs sentence embeddings directly.
-	Pooling PoolingStrategy
-	// HiddenSize is the embedding dimension (used for pooling calculations).
-	HiddenSize int
+	HiddenSize  int
 }
 
 // EmbeddingModel represents a text embedding model.
@@ -62,11 +57,11 @@ type ONNXConfigurer interface {
 
 // ModelMetadata describes an embedding model for UI/config.
 type ModelMetadata struct {
-	Name        string `json:"name"`        // Human-readable name
-	Version     string `json:"version"`     // Short ID for DB storage
-	Dimensions  int    `json:"dimensions"`  // Vector size
-	Description string `json:"description"` // Brief description
-	Default     bool   `json:"default"`     // Is this the default model?
+	Name        string `json:"name"`
+	Version     string `json:"version"`
+	Description string `json:"description"`
+	Dimensions  int    `json:"dimensions"`
+	Default     bool   `json:"default"`
 }
 
 // ModelFactory creates a new instance of an embedding model.
@@ -74,10 +69,10 @@ type ModelFactory func() (EmbeddingModel, error)
 
 // ModelRegistry provides model lookup by version.
 type ModelRegistry struct {
-	mu           sync.RWMutex
 	models       map[string]ModelFactory
 	metadata     map[string]ModelMetadata
 	defaultModel string
+	mu           sync.RWMutex
 }
 
 // NewModelRegistry creates a new model registry.

internal/embedding/service.go

@@ -46,9 +46,9 @@ var bgeONNXConfig = ONNXConfig{
 type bgeModel struct {
 	tk      *tokenizer.Tokenizer
 	session *ort.DynamicAdvancedSession
+	libDir  string
+	config  ONNXConfig
 	mu      sync.Mutex
-	libDir  string     // temp directory containing extracted libraries
-	config  ONNXConfig // ONNX configuration for this model
 }
 
 // Compile-time check that bgeModel implements EmbeddingModel
@@ -70,8 +70,8 @@ func newBGEModel() (EmbeddingModel, error) {
 	ort.SetSharedLibraryPath(libPath)
 
 	// Initialize ONNX runtime
-	if err := ort.InitializeEnvironment(); err != nil {
-		return nil, fmt.Errorf("initialize ONNX runtime: %w", err)
+	if initErr := ort.InitializeEnvironment(); initErr != nil {
+		return nil, fmt.Errorf("initialize ONNX runtime: %w", initErr)
 	}
 
 	// Load tokenizer from embedded data
@@ -292,19 +292,19 @@ func (m *bgeModel) computeBatch(sentences []string) ([][]float32, error) {
 	if err != nil {
 		return nil, fmt.Errorf("create input_ids tensor: %w", err)
 	}
-	defer inputIdsTensor.Destroy()
+	defer func() { _ = inputIdsTensor.Destroy() }()
 
 	attentionMaskTensor, err := ort.NewTensor(inputShape, attentionMaskData)
 	if err != nil {
 		return nil, fmt.Errorf("create attention_mask tensor: %w", err)
 	}
-	defer attentionMaskTensor.Destroy()
+	defer func() { _ = attentionMaskTensor.Destroy() }()
 
 	tokenTypeIdsTensor, err := ort.NewTensor(inputShape, tokenTypeIdsData)
 	if err != nil {
 		return nil, fmt.Errorf("create token_type_ids tensor: %w", err)
 	}
-	defer tokenTypeIdsTensor.Destroy()
+	defer func() { _ = tokenTypeIdsTensor.Destroy() }()
 
 	// Create output tensor based on pooling strategy
 	var outputShape ort.Shape
@@ -324,7 +324,7 @@ func (m *bgeModel) computeBatch(sentences []string) ([][]float32, error) {
 	if err != nil {
 		return nil, fmt.Errorf("create output tensor: %w", err)
 	}
-	defer outputTensor.Destroy()
+	defer func() { _ = outputTensor.Destroy() }()
 
 	// Run inference
 	inputTensors := []ort.Value{inputIdsTensor, attentionMaskTensor, tokenTypeIdsTensor}

internal/graph/edge_detector.go

@@ -0,0 +1,417 @@
+package graph
+
+import (
+	"context"
+	"fmt"
+	"math"
+
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+	"github.com/rs/zerolog/log"
+)
+
+const (
+	// SemanticSimilarityThreshold for creating semantic edges
+	SemanticSimilarityThreshold = 0.85
+
+	// MinFileOverlapForEdge minimum file overlap ratio to create edge
+	MinFileOverlapForEdge = 0.3
+
+	// MaxEdgesPerNode prevents creating too many edges
+	MaxEdgesPerNode = 20
+)
+
+// DetectEdges identifies relationships between observations
+func DetectEdges(ctx context.Context, observations []*models.Observation) ([]Edge, error) {
+	if len(observations) < 2 {
+		return nil, nil
+	}
+
+	edges := make([]Edge, 0)
+
+	// Build lookup maps for efficient detection
+	sessionMap := buildSessionMap(observations)
+	conceptMap := buildConceptMap(observations)
+	fileMap := buildFileMap(observations)
+
+	log.Info().
+		Int("observations", len(observations)).
+		Int("sessions", len(sessionMap)).
+		Int("concepts", len(conceptMap)).
+		Msg("Starting edge detection")
+
+	// Detect temporal edges (same session)
+	temporalEdges := detectTemporalEdges(sessionMap)
+	edges = append(edges, temporalEdges...)
+
+	// Detect concept edges (shared tags)
+	conceptEdges := detectConceptEdges(conceptMap)
+	edges = append(edges, conceptEdges...)
+
+	// Detect file overlap edges
+	fileEdges := detectFileOverlapEdges(fileMap, observations)
+	edges = append(edges, fileEdges...)
+
+	// Prune excessive edges per node
+	edges = pruneEdges(edges, MaxEdgesPerNode)
+
+	log.Info().
+		Int("temporal_edges", len(temporalEdges)).
+		Int("concept_edges", len(conceptEdges)).
+		Int("file_edges", len(fileEdges)).
+		Int("total_edges", len(edges)).
+		Msg("Edge detection complete")
+
+	return edges, nil
+}
+
+// buildSessionMap groups observations by SDK session
+func buildSessionMap(observations []*models.Observation) map[string][]int64 {
+	sessionMap := make(map[string][]int64)
+
+	for _, obs := range observations {
+		if obs.SDKSessionID != "" {
+			sessionMap[obs.SDKSessionID] = append(sessionMap[obs.SDKSessionID], obs.ID)
+		}
+	}
+
+	return sessionMap
+}
+
+// buildConceptMap groups observations by concept tags
+func buildConceptMap(observations []*models.Observation) map[string][]int64 {
+	conceptMap := make(map[string][]int64)
+
+	for _, obs := range observations {
+		for _, concept := range obs.Concepts {
+			conceptMap[concept] = append(conceptMap[concept], obs.ID)
+		}
+	}
+
+	return conceptMap
+}
+
+// buildFileMap maps files to observations (from both FilesRead and FilesModified)
+func buildFileMap(observations []*models.Observation) map[string][]int64 {
+	fileMap := make(map[string][]int64)
+
+	for _, obs := range observations {
+		// Add files from FilesRead
+		for _, file := range obs.FilesRead {
+			fileMap[file] = append(fileMap[file], obs.ID)
+		}
+		// Add files from FilesModified
+		for _, file := range obs.FilesModified {
+			fileMap[file] = append(fileMap[file], obs.ID)
+		}
+	}
+
+	return fileMap
+}
+
+// detectTemporalEdges creates edges between observations in the same session
+func detectTemporalEdges(sessionMap map[string][]int64) []Edge {
+	edges := make([]Edge, 0)
+
+	for _, obsIDs := range sessionMap {
+		if len(obsIDs) < 2 {
+			continue
+		}
+
+		// Create edges between consecutive observations in session
+		for i := 0; i < len(obsIDs)-1; i++ {
+			edges = append(edges, Edge{
+				FromID:   obsIDs[i],
+				ToID:     obsIDs[i+1],
+				Relation: RelationTemporal,
+				Weight:   0.8, // High weight for temporal proximity
+			})
+		}
+	}
+
+	return edges
+}
+
+// detectConceptEdges creates edges between observations sharing concepts
+func detectConceptEdges(conceptMap map[string][]int64) []Edge {
+	edges := make([]Edge, 0)
+	seen := make(map[string]bool)
+
+	for concept, obsIDs := range conceptMap {
+		if len(obsIDs) < 2 {
+			continue
+		}
+
+		// Create edges between all observations sharing this concept
+		for i := 0; i < len(obsIDs); i++ {
+			for j := i + 1; j < len(obsIDs); j++ {
+				// Use sorted pair as key to avoid duplicates
+				pairKey := edgeKey(obsIDs[i], obsIDs[j])
+				if seen[pairKey] {
+					continue
+				}
+				seen[pairKey] = true
+
+				// Weight based on concept specificity (longer = more specific)
+				weight := float32(0.5 + 0.3*math.Min(1.0, float64(len(concept))/20.0))
+
+				edges = append(edges, Edge{
+					FromID:   obsIDs[i],
+					ToID:     obsIDs[j],
+					Relation: RelationConcept,
+					Weight:   weight,
+				})
+			}
+		}
+	}
+
+	return edges
+}
+
+// detectFileOverlapEdges creates edges based on file references
+func detectFileOverlapEdges(fileMap map[string][]int64, observations []*models.Observation) []Edge {
+	edges := make([]Edge, 0)
+	seen := make(map[string]bool)
+
+	// Build observation ID to observation map for quick lookup
+	obsMap := make(map[int64]*models.Observation)
+	for _, obs := range observations {
+		obsMap[obs.ID] = obs
+	}
+
+	for _, obsIDs := range fileMap {
+		if len(obsIDs) < 2 {
+			continue
+		}
+
+		// Create edges between observations referencing same files
+		for i := 0; i < len(obsIDs); i++ {
+			for j := i + 1; j < len(obsIDs); j++ {
+				pairKey := edgeKey(obsIDs[i], obsIDs[j])
+				if seen[pairKey] {
+					continue
+				}
+				seen[pairKey] = true
+
+				// Calculate file overlap ratio
+				obs1, ok1 := obsMap[obsIDs[i]]
+				obs2, ok2 := obsMap[obsIDs[j]]
+
+				if !ok1 || !ok2 {
+					continue
+				}
+
+				// Merge FilesRead and FilesModified for both observations
+				files1 := append([]string{}, obs1.FilesRead...)
+				files1 = append(files1, obs1.FilesModified...)
+				files2 := append([]string{}, obs2.FilesRead...)
+				files2 = append(files2, obs2.FilesModified...)
+
+				overlap := calculateFileOverlap(files1, files2)
+				if overlap < MinFileOverlapForEdge {
+					continue
+				}
+
+				edges = append(edges, Edge{
+					FromID:   obsIDs[i],
+					ToID:     obsIDs[j],
+					Relation: RelationFileOverlap,
+					Weight:   overlap,
+				})
+			}
+		}
+	}
+
+	return edges
+}
+
+// calculateFileOverlap computes Jaccard similarity of file sets
+func calculateFileOverlap(files1, files2 []string) float32 {
+	if len(files1) == 0 || len(files2) == 0 {
+		return 0.0
+	}
+
+	// Convert to sets
+	set1 := make(map[string]bool)
+	for _, f := range files1 {
+		set1[f] = true
+	}
+
+	set2 := make(map[string]bool)
+	for _, f := range files2 {
+		set2[f] = true
+	}
+
+	// Count intersection
+	intersection := 0
+	for f := range set1 {
+		if set2[f] {
+			intersection++
+		}
+	}
+
+	// Jaccard similarity = intersection / union
+	union := len(set1) + len(set2) - intersection
+	if union == 0 {
+		return 0.0
+	}
+
+	return float32(intersection) / float32(union)
+}
+
+// pruneEdges limits edges per node to prevent graph explosion
+func pruneEdges(edges []Edge, maxPerNode int) []Edge {
+	if maxPerNode <= 0 {
+		return edges
+	}
+
+	// Count edges per node
+	outEdges := make(map[int64][]Edge)
+	inEdges := make(map[int64][]Edge)
+
+	for _, edge := range edges {
+		outEdges[edge.FromID] = append(outEdges[edge.FromID], edge)
+		inEdges[edge.ToID] = append(inEdges[edge.ToID], edge)
+	}
+
+	// Prune low-weight edges if node has too many
+	pruned := make([]Edge, 0, len(edges))
+	processed := make(map[string]bool)
+
+	for _, edge := range edges {
+		pairKey := edgeKey(edge.FromID, edge.ToID)
+		if processed[pairKey] {
+			continue
+		}
+		processed[pairKey] = true
+
+		// Check if either node has too many edges
+		fromCount := len(outEdges[edge.FromID])
+		toCount := len(inEdges[edge.ToID])
+
+		if fromCount <= maxPerNode && toCount <= maxPerNode {
+			pruned = append(pruned, edge)
+			continue
+		}
+
+		// Keep edge if it's high-weight (top edges for this node)
+		if shouldKeepEdge(edge, outEdges[edge.FromID], maxPerNode) {
+			pruned = append(pruned, edge)
+		}
+	}
+
+	if len(pruned) < len(edges) {
+		log.Debug().
+			Int("original", len(edges)).
+			Int("pruned", len(pruned)).
+			Int("removed", len(edges)-len(pruned)).
+			Msg("Pruned excessive edges")
+	}
+
+	return pruned
+}
+
+// shouldKeepEdge determines if edge should be kept during pruning
+func shouldKeepEdge(edge Edge, nodeEdges []Edge, maxPerNode int) bool {
+	// Sort node's edges by weight descending
+	sortedEdges := make([]Edge, len(nodeEdges))
+	copy(sortedEdges, nodeEdges)
+
+	sortEdgesByWeight(sortedEdges)
+
+	// Keep edge if it's in top maxPerNode
+	for i := 0; i < maxPerNode && i < len(sortedEdges); i++ {
+		if sortedEdges[i].FromID == edge.FromID && sortedEdges[i].ToID == edge.ToID {
+			return true
+		}
+	}
+
+	return false
+}
+
+// sortEdgesByWeight sorts edges by weight descending
+func sortEdgesByWeight(edges []Edge) {
+	// Simple bubble sort (edges are typically small per node)
+	n := len(edges)
+	for i := 0; i < n-1; i++ {
+		for j := 0; j < n-i-1; j++ {
+			if edges[j].Weight < edges[j+1].Weight {
+				edges[j], edges[j+1] = edges[j+1], edges[j]
+			}
+		}
+	}
+}
+
+// edgeKey creates a unique key for an edge pair (sorted)
+func edgeKey(id1, id2 int64) string {
+	if id1 < id2 {
+		return fmt.Sprintf("%d-%d", id1, id2)
+	}
+	return fmt.Sprintf("%d-%d", id2, id1)
+}
+
+// DetectSemanticEdges creates edges based on semantic similarity
+// This requires embeddings and is called separately when available
+func DetectSemanticEdges(ctx context.Context, observations []*models.Observation, embeddings map[int64][]float32) []Edge {
+	edges := make([]Edge, 0)
+	seen := make(map[string]bool)
+
+	// Compare all pairs (expensive, but necessary for semantic similarity)
+	for i := 0; i < len(observations); i++ {
+		emb1, ok1 := embeddings[observations[i].ID]
+		if !ok1 {
+			continue
+		}
+
+		for j := i + 1; j < len(observations); j++ {
+			emb2, ok2 := embeddings[observations[j].ID]
+			if !ok2 {
+				continue
+			}
+
+			similarity := cosineSimilarity(emb1, emb2)
+			if similarity < SemanticSimilarityThreshold {
+				continue
+			}
+
+			pairKey := edgeKey(observations[i].ID, observations[j].ID)
+			if seen[pairKey] {
+				continue
+			}
+			seen[pairKey] = true
+
+			edges = append(edges, Edge{
+				FromID:   observations[i].ID,
+				ToID:     observations[j].ID,
+				Relation: RelationSemantic,
+				Weight:   similarity,
+			})
+		}
+	}
+
+	log.Info().
+		Int("semantic_edges", len(edges)).
+		Float32("threshold", SemanticSimilarityThreshold).
+		Msg("Detected semantic edges")
+
+	return edges
+}
+
+// cosineSimilarity computes cosine similarity between two vectors
+func cosineSimilarity(a, b []float32) float32 {
+	if len(a) != len(b) {
+		return 0.0
+	}
+
+	var dotProduct, normA, normB float32
+	for i := range a {
+		dotProduct += a[i] * b[i]
+		normA += a[i] * a[i]
+		normB += b[i] * b[i]
+	}
+
+	if normA == 0 || normB == 0 {
+		return 0.0
+	}
+
+	return dotProduct / float32(math.Sqrt(float64(normA))*math.Sqrt(float64(normB)))
+}

internal/graph/observation_graph.go

@@ -0,0 +1,423 @@
+// Package graph provides observation relationship graphs for LEANN Phase 2.
+//
+// This package implements graph-based selective recomputation where observation
+// relationships (file overlap, semantic similarity, temporal proximity) form a
+// graph structure. Hub nodes (high-degree observations) store embeddings, while
+// leaf nodes recompute on-demand.
+package graph
+
+import (
+	"context"
+	"fmt"
+	"math"
+	"sort"
+	"sync"
+	"time"
+
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+	"github.com/rs/zerolog/log"
+)
+
+// RelationType defines the type of relationship between observations
+type RelationType int
+
+const (
+	// RelationFileOverlap indicates observations reference overlapping files
+	RelationFileOverlap RelationType = iota
+	// RelationSemantic indicates high semantic similarity (cosine > 0.85)
+	RelationSemantic
+	// RelationTemporal indicates observations from same session
+	RelationTemporal
+	// RelationConcept indicates shared concept tags
+	RelationConcept
+)
+
+// Edge represents a relationship between two observations
+type Edge struct {
+	FromID   int64
+	ToID     int64
+	Relation RelationType
+	Weight   float32 // 0.0-1.0, higher = stronger relationship
+}
+
+// Node represents an observation in the graph
+type Node struct {
+	Metadata    NodeMetadata
+	LastAccess  time.Time
+	StoredEmb   []float32 // Nil if recomputed on-demand
+	ID          int64
+	Degree      int // Number of edges (hub detection)
+	AccessCount int
+}
+
+// NodeMetadata contains observation metadata
+type NodeMetadata struct {
+	CreatedAt    time.Time
+	Project      string
+	Type         string
+	Title        string
+	IsSuperseded bool
+}
+
+// CSRGraph represents a graph in Compressed Sparse Row format for memory efficiency
+type CSRGraph struct {
+	RowPtr  []int32   // Node adjacency list pointers
+	ColIdx  []int32   // Edge destination IDs
+	Weights []float32 // Edge weights
+	mu      sync.RWMutex
+}
+
+// ObservationGraph manages the observation relationship graph
+type ObservationGraph struct {
+	nodes   map[int64]*Node
+	csr     *CSRGraph
+	edges   []Edge
+	nodesMu sync.RWMutex
+	edgesMu sync.RWMutex
+}
+
+// NewObservationGraph creates a new empty observation graph
+func NewObservationGraph() *ObservationGraph {
+	return &ObservationGraph{
+		nodes: make(map[int64]*Node),
+		edges: make([]Edge, 0),
+		csr:   &CSRGraph{},
+	}
+}
+
+// AddNode adds or updates a node in the graph
+func (g *ObservationGraph) AddNode(node *Node) {
+	g.nodesMu.Lock()
+	defer g.nodesMu.Unlock()
+
+	g.nodes[node.ID] = node
+}
+
+// AddEdge adds an edge to the graph
+func (g *ObservationGraph) AddEdge(edge Edge) {
+	g.edgesMu.Lock()
+	defer g.edgesMu.Unlock()
+
+	g.edges = append(g.edges, edge)
+
+	// Update degree counts
+	g.nodesMu.Lock()
+	if fromNode, ok := g.nodes[edge.FromID]; ok {
+		fromNode.Degree++
+	}
+	if toNode, ok := g.nodes[edge.ToID]; ok {
+		toNode.Degree++
+	}
+	g.nodesMu.Unlock()
+}
+
+// BuildCSR converts edge list to CSR format for efficient traversal
+func (g *ObservationGraph) BuildCSR() error {
+	g.edgesMu.RLock()
+	g.nodesMu.RLock()
+	defer g.edgesMu.RUnlock()
+	defer g.nodesMu.RUnlock()
+
+	if len(g.nodes) == 0 {
+		return fmt.Errorf("no nodes in graph")
+	}
+
+	// Create node ID to index mapping
+	nodeIDs := make([]int64, 0, len(g.nodes))
+	for id := range g.nodes {
+		nodeIDs = append(nodeIDs, id)
+	}
+	sort.Slice(nodeIDs, func(i, j int) bool {
+		return nodeIDs[i] < nodeIDs[j]
+	})
+
+	idToIdx := make(map[int64]int32)
+	for idx, id := range nodeIDs {
+		// #nosec G115 - observation count will never exceed int32 max (2.1B) in practice
+		idToIdx[id] = int32(idx)
+	}
+
+	// Count edges per node
+	edgeCounts := make([]int, len(nodeIDs))
+	for _, edge := range g.edges {
+		if fromIdx, ok := idToIdx[edge.FromID]; ok {
+			edgeCounts[fromIdx]++
+		}
+	}
+
+	// Build row pointers
+	rowPtr := make([]int32, len(nodeIDs)+1)
+	rowPtr[0] = 0
+	for i := 0; i < len(nodeIDs); i++ {
+		// #nosec G115 - edge counts per node will not exceed int32 max
+		rowPtr[i+1] = rowPtr[i] + int32(edgeCounts[i])
+	}
+
+	// Build column indices and weights
+	totalEdges := rowPtr[len(nodeIDs)]
+	colIdx := make([]int32, totalEdges)
+	weights := make([]float32, totalEdges)
+
+	// Temporary counter for filling CSR
+	currentPos := make([]int32, len(nodeIDs))
+	copy(currentPos, rowPtr[:len(nodeIDs)])
+
+	for _, edge := range g.edges {
+		fromIdx, fromOk := idToIdx[edge.FromID]
+		toIdx, toOk := idToIdx[edge.ToID]
+
+		if fromOk && toOk {
+			pos := currentPos[fromIdx]
+			colIdx[pos] = toIdx
+			weights[pos] = edge.Weight
+			currentPos[fromIdx]++
+		}
+	}
+
+	g.csr.mu.Lock()
+	g.csr.RowPtr = rowPtr
+	g.csr.ColIdx = colIdx
+	g.csr.Weights = weights
+	g.csr.mu.Unlock()
+
+	log.Info().
+		Int("nodes", len(nodeIDs)).
+		Int("edges", int(totalEdges)).
+		Msg("Built CSR graph representation")
+
+	return nil
+}
+
+// GetNeighbors returns neighboring nodes and their edge weights
+func (g *ObservationGraph) GetNeighbors(nodeID int64) ([]int64, []float32, error) {
+	g.csr.mu.RLock()
+	defer g.csr.mu.RUnlock()
+
+	// Find node index in CSR
+	g.nodesMu.RLock()
+	nodeIDs := make([]int64, 0, len(g.nodes))
+	for id := range g.nodes {
+		nodeIDs = append(nodeIDs, id)
+	}
+	g.nodesMu.RUnlock()
+
+	sort.Slice(nodeIDs, func(i, j int) bool {
+		return nodeIDs[i] < nodeIDs[j]
+	})
+
+	nodeIdx := sort.Search(len(nodeIDs), func(i int) bool {
+		return nodeIDs[i] >= nodeID
+	})
+
+	if nodeIdx >= len(nodeIDs) || nodeIDs[nodeIdx] != nodeID {
+		return nil, nil, fmt.Errorf("node %d not found", nodeID)
+	}
+
+	// Extract neighbors from CSR
+	startIdx := g.csr.RowPtr[nodeIdx]
+	endIdx := g.csr.RowPtr[nodeIdx+1]
+
+	neighborCount := endIdx - startIdx
+	neighbors := make([]int64, neighborCount)
+	weights := make([]float32, neighborCount)
+
+	for i := int32(0); i < neighborCount; i++ {
+		neighborIdx := g.csr.ColIdx[startIdx+i]
+		neighbors[i] = nodeIDs[neighborIdx]
+		weights[i] = g.csr.Weights[startIdx+i]
+	}
+
+	return neighbors, weights, nil
+}
+
+// GetNode retrieves a node by ID
+func (g *ObservationGraph) GetNode(nodeID int64) (*Node, error) {
+	g.nodesMu.RLock()
+	defer g.nodesMu.RUnlock()
+
+	node, ok := g.nodes[nodeID]
+	if !ok {
+		return nil, fmt.Errorf("node %d not found", nodeID)
+	}
+
+	return node, nil
+}
+
+// FindHubs identifies hub nodes (high degree) in the graph
+func (g *ObservationGraph) FindHubs(percentile float64) []int64 {
+	g.nodesMu.RLock()
+	defer g.nodesMu.RUnlock()
+
+	if len(g.nodes) == 0 {
+		return nil
+	}
+
+	// Collect all degrees
+	degrees := make([]int, 0, len(g.nodes))
+	nodeIDs := make([]int64, 0, len(g.nodes))
+
+	for id, node := range g.nodes {
+		degrees = append(degrees, node.Degree)
+		nodeIDs = append(nodeIDs, id)
+	}
+
+	// Sort by degree
+	type nodeDegree struct {
+		ID     int64
+		Degree int
+	}
+
+	nodeDegrees := make([]nodeDegree, len(nodeIDs))
+	for i := range nodeIDs {
+		nodeDegrees[i] = nodeDegree{
+			ID:     nodeIDs[i],
+			Degree: degrees[i],
+		}
+	}
+
+	sort.Slice(nodeDegrees, func(i, j int) bool {
+		return nodeDegrees[i].Degree > nodeDegrees[j].Degree
+	})
+
+	// Return top percentile
+	cutoff := int(math.Ceil(float64(len(nodeDegrees)) * (1.0 - percentile)))
+	if cutoff > len(nodeDegrees) {
+		cutoff = len(nodeDegrees)
+	}
+
+	hubs := make([]int64, cutoff)
+	for i := 0; i < cutoff; i++ {
+		hubs[i] = nodeDegrees[i].ID
+	}
+
+	log.Info().
+		Int("total_nodes", len(g.nodes)).
+		Int("hubs", len(hubs)).
+		Float64("percentile", percentile).
+		Msg("Identified hub nodes")
+
+	return hubs
+}
+
+// Stats returns graph statistics
+func (g *ObservationGraph) Stats() GraphStats {
+	g.nodesMu.RLock()
+	g.edgesMu.RLock()
+	defer g.nodesMu.RUnlock()
+	defer g.edgesMu.RUnlock()
+
+	stats := GraphStats{
+		NodeCount: len(g.nodes),
+		EdgeCount: len(g.edges),
+	}
+
+	if len(g.nodes) > 0 {
+		degrees := make([]int, 0, len(g.nodes))
+		for _, node := range g.nodes {
+			degrees = append(degrees, node.Degree)
+		}
+
+		sort.Ints(degrees)
+		stats.AvgDegree = float64(sum(degrees)) / float64(len(degrees))
+		stats.MaxDegree = degrees[len(degrees)-1]
+		stats.MinDegree = degrees[0]
+
+		// Median
+		mid := len(degrees) / 2
+		if len(degrees)%2 == 0 {
+			stats.MedianDegree = float64(degrees[mid-1]+degrees[mid]) / 2.0
+		} else {
+			stats.MedianDegree = float64(degrees[mid])
+		}
+	}
+
+	// Count edge types
+	stats.EdgeTypes = make(map[RelationType]int)
+	for _, edge := range g.edges {
+		stats.EdgeTypes[edge.Relation]++
+	}
+
+	return stats
+}
+
+// GraphStats contains graph statistics
+type GraphStats struct {
+	EdgeTypes    map[RelationType]int
+	AvgDegree    float64
+	MedianDegree float64
+	NodeCount    int
+	EdgeCount    int
+	MaxDegree    int
+	MinDegree    int
+}
+
+// BuildFromObservations constructs a graph from a list of observations
+func BuildFromObservations(ctx context.Context, observations []*models.Observation) (*ObservationGraph, error) {
+	graph := NewObservationGraph()
+
+	// Add nodes
+	for _, obs := range observations {
+		// Extract title from sql.NullString
+		title := ""
+		if obs.Title.Valid {
+			title = obs.Title.String
+		}
+
+		node := &Node{
+			ID:     obs.ID,
+			Degree: 0,
+			Metadata: NodeMetadata{
+				Project:      obs.Project,
+				Type:         string(obs.Type),
+				Title:        title,
+				CreatedAt:    time.UnixMilli(obs.CreatedAtEpoch),
+				IsSuperseded: obs.IsSuperseded,
+			},
+			LastAccess:  time.Now(),
+			AccessCount: 0,
+		}
+		graph.AddNode(node)
+	}
+
+	// Detect edges (will be implemented in edge_detector.go)
+	edges, err := DetectEdges(ctx, observations)
+	if err != nil {
+		return nil, fmt.Errorf("detect edges: %w", err)
+	}
+
+	for _, edge := range edges {
+		graph.AddEdge(edge)
+	}
+
+	// Build CSR representation
+	if err := graph.BuildCSR(); err != nil {
+		return nil, fmt.Errorf("build CSR: %w", err)
+	}
+
+	return graph, nil
+}
+
+// Helper function to sum integers
+func sum(values []int) int {
+	total := 0
+	for _, v := range values {
+		total += v
+	}
+	return total
+}
+
+// String returns a human-readable representation of RelationType
+func (r RelationType) String() string {
+	switch r {
+	case RelationFileOverlap:
+		return "file_overlap"
+	case RelationSemantic:
+		return "semantic"
+	case RelationTemporal:
+		return "temporal"
+	case RelationConcept:
+		return "concept"
+	default:
+		return "unknown"
+	}
+}

internal/mcp/server.go

@@ -15,10 +15,10 @@ import (
 
 // Server is the MCP server that exposes search tools.
 type Server struct {
-	searchMgr *search.Manager
-	version   string
 	stdin     io.Reader
 	stdout    io.Writer
+	searchMgr *search.Manager
+	version   string
 }
 
 // NewServer creates a new MCP server.
@@ -41,17 +41,17 @@ type Request struct {
 
 // Response represents a JSON-RPC response.
 type Response struct {
-	JSONRPC string `json:"jsonrpc"`
 	ID      any    `json:"id"`
 	Result  any    `json:"result,omitempty"`
 	Error   *Error `json:"error,omitempty"`
+	JSONRPC string `json:"jsonrpc"`
 }
 
 // Error represents a JSON-RPC error.
 type Error struct {
-	Code    int    `json:"code"`
-	Message string `json:"message"`
 	Data    any    `json:"data,omitempty"`
+	Message string `json:"message"`
+	Code    int    `json:"code"`
 }
 
 // ToolCallParams represents parameters for tools/call method.
@@ -62,9 +62,9 @@ type ToolCallParams struct {
 
 // Tool represents an MCP tool definition.
 type Tool struct {
+	InputSchema map[string]any `json:"inputSchema"`
 	Name        string         `json:"name"`
 	Description string         `json:"description"`
-	InputSchema map[string]any `json:"inputSchema"`
 }
 
 // Run starts the MCP server loop.
@@ -440,17 +440,17 @@ func (s *Server) callTool(ctx context.Context, name string, args json.RawMessage
 
 // TimelineParams represents parameters for timeline operations.
 type TimelineParams struct {
-	AnchorID  int64  `json:"anchor_id"`
 	Query     string `json:"query"`
-	Before    int    `json:"before"`
-	After     int    `json:"after"`
 	Project   string `json:"project"`
 	ObsType   string `json:"obs_type"`
 	Concepts  string `json:"concepts"`
 	Files     string `json:"files"`
+	Format    string `json:"format"`
+	AnchorID  int64  `json:"anchor_id"`
+	Before    int    `json:"before"`
+	After     int    `json:"after"`
 	DateStart int64  `json:"dateStart"`
 	DateEnd   int64  `json:"dateEnd"`
-	Format    string `json:"format"`
 }
 
 // handleTimeline handles timeline requests.

internal/mcp/server_test.go

@@ -34,8 +34,8 @@ func (s *ServerSuite) TestNewServer() {
 func TestRequest(t *testing.T) {
 	tests := []struct {
 		name     string
-		req      Request
 		expected string
+		req      Request
 	}{
 		{
 			name: "initialize request",
@@ -137,10 +137,10 @@ func TestResponse(t *testing.T) {
 
 // TestError tests Error struct.
 func TestError(t *testing.T) {
-	tests := []struct {
-		name     string
+	tests := []struct { //nolint:govet
 		err      Error
 		expected string
+		name     string
 	}{
 		{
 			name: "parse error",
@@ -365,11 +365,11 @@ func TestHandleRequest(t *testing.T) {
 	ctx := context.Background()
 
 	tests := []struct {
-		name         string
 		req          *Request
-		expectError  bool
-		errorCode    int
+		name         string
 		errorMessage string
+		errorCode    int
+		expectError  bool
 	}{
 		{
 			name: "initialize method",
@@ -683,10 +683,8 @@ func TestRun_MultipleRequests(t *testing.T) {
 func TestHandleTimeline_Defaults(t *testing.T) {
 	// Test that handleTimeline sets default before/after values
 	params := TimelineParams{
-		AnchorID: 0,
-		Query:    "",
-		Before:   0,
-		After:    0,
+		Before: 0,
+		After:  0,
 	}
 
 	// Simulate the default value assignment from handleTimeline
@@ -753,13 +751,13 @@ func TestServerStdinStdoutConfig(t *testing.T) {
 // TestResponseIDTypes tests that response IDs can be various types.
 func TestResponseIDTypes(t *testing.T) {
 	tests := []struct {
-		name string
 		id   any
+		name string
 	}{
-		{"integer id", 1},
-		{"string id", "abc-123"},
-		{"float id", 1.5},
-		{"null id", nil},
+		{name: "integer id", id: 1},
+		{name: "string id", id: "abc-123"},
+		{name: "float id", id: 1.5},
+		{name: "null id", id: nil},
 	}
 
 	for _, tt := range tests {

internal/pattern/detector.go

@@ -0,0 +1,432 @@
+// Package pattern provides pattern detection and recognition functionality.
+package pattern
+
+import (
+	"context"
+	"sync"
+	"time"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/db/sqlite"
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+	"github.com/rs/zerolog/log"
+)
+
+// DetectorConfig contains configuration for the pattern detector.
+type DetectorConfig struct {
+	// MinMatchScore is the minimum similarity score to consider a match (0.0-1.0).
+	MinMatchScore float64
+	// MinFrequencyForPattern is the minimum occurrences before creating a pattern.
+	MinFrequencyForPattern int
+	// AnalysisInterval is how often to run background pattern analysis.
+	AnalysisInterval time.Duration
+	// MaxPatternsToTrack is the maximum number of active patterns.
+	MaxPatternsToTrack int
+}
+
+// DefaultConfig returns the default detector configuration.
+func DefaultConfig() DetectorConfig {
+	return DetectorConfig{
+		MinMatchScore:          0.3, // 30% similarity threshold
+		MinFrequencyForPattern: 2,   // At least 2 occurrences to form a pattern
+		AnalysisInterval:       5 * time.Minute,
+		MaxPatternsToTrack:     1000,
+	}
+}
+
+// PatternSyncFunc is a callback for syncing patterns to vector store.
+type PatternSyncFunc func(pattern *models.Pattern)
+
+// Detector detects and tracks recurring patterns across observations.
+type Detector struct {
+	ctx              context.Context
+	patternStore     *sqlite.PatternStore
+	observationStore *sqlite.ObservationStore
+	syncFunc         PatternSyncFunc
+	candidates       map[string]*candidatePattern
+	cancel           context.CancelFunc
+	config           DetectorConfig
+	wg               sync.WaitGroup
+	candidatesMu     sync.RWMutex
+}
+
+// SetSyncFunc sets the callback for syncing patterns to vector store.
+func (d *Detector) SetSyncFunc(fn PatternSyncFunc) {
+	d.syncFunc = fn
+}
+
+// candidatePattern tracks a potential pattern before it reaches frequency threshold.
+type candidatePattern struct {
+	patternType    models.PatternType
+	title          string
+	signature      []string
+	observationIDs []int64
+	projects       []string
+	lastSeenEpoch  int64
+}
+
+// NewDetector creates a new pattern detector.
+func NewDetector(patternStore *sqlite.PatternStore, observationStore *sqlite.ObservationStore, config DetectorConfig) *Detector {
+	ctx, cancel := context.WithCancel(context.Background())
+	return &Detector{
+		config:           config,
+		patternStore:     patternStore,
+		observationStore: observationStore,
+		candidates:       make(map[string]*candidatePattern),
+		ctx:              ctx,
+		cancel:           cancel,
+	}
+}
+
+// Start begins background pattern analysis.
+func (d *Detector) Start() {
+	d.wg.Add(1)
+	go d.backgroundAnalysis()
+	log.Info().Dur("interval", d.config.AnalysisInterval).Msg("Pattern detector started")
+}
+
+// Stop stops background pattern analysis.
+func (d *Detector) Stop() {
+	d.cancel()
+	d.wg.Wait()
+	log.Info().Msg("Pattern detector stopped")
+}
+
+// backgroundAnalysis runs periodic pattern analysis.
+func (d *Detector) backgroundAnalysis() {
+	defer d.wg.Done()
+
+	ticker := time.NewTicker(d.config.AnalysisInterval)
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-d.ctx.Done():
+			return
+		case <-ticker.C:
+			if err := d.AnalyzeRecentObservations(d.ctx); err != nil {
+				log.Warn().Err(err).Msg("Background pattern analysis failed")
+			}
+		}
+	}
+}
+
+// AnalyzeObservation processes a new observation for pattern detection.
+// This is called synchronously when a new observation is stored.
+func (d *Detector) AnalyzeObservation(ctx context.Context, obs *models.Observation) (*DetectionResult, error) {
+	result := &DetectionResult{}
+
+	// Extract signature from observation
+	signature := models.ExtractSignature(
+		obs.Concepts,
+		obs.Title.String,
+		obs.Narrative.String,
+	)
+	if len(signature) == 0 {
+		return result, nil // Nothing to detect
+	}
+
+	// Check against existing patterns
+	matches, err := d.patternStore.FindMatchingPatterns(ctx, signature, d.config.MinMatchScore)
+	if err != nil {
+		return nil, err
+	}
+
+	if len(matches) > 0 {
+		// Found existing pattern match
+		bestMatch := matches[0]
+		for _, m := range matches[1:] {
+			if models.CalculateMatchScore(signature, m.Signature) > models.CalculateMatchScore(signature, bestMatch.Signature) {
+				bestMatch = m
+			}
+		}
+
+		// Update the pattern with new occurrence
+		if err := d.patternStore.IncrementPatternFrequency(ctx, bestMatch.ID, obs.Project, obs.ID); err != nil {
+			log.Warn().Err(err).Int64("pattern_id", bestMatch.ID).Msg("Failed to update pattern frequency")
+		}
+
+		result.MatchedPattern = bestMatch
+		result.MatchScore = models.CalculateMatchScore(signature, bestMatch.Signature)
+		result.IsNewPattern = false
+
+		log.Debug().
+			Int64("pattern_id", bestMatch.ID).
+			Str("pattern_name", bestMatch.Name).
+			Float64("score", result.MatchScore).
+			Msg("Observation matched existing pattern")
+
+		return result, nil
+	}
+
+	// No existing pattern match - check candidates
+	candidateKey := generateCandidateKey(signature)
+	d.candidatesMu.Lock()
+	defer d.candidatesMu.Unlock()
+
+	if candidate, exists := d.candidates[candidateKey]; exists {
+		// Update existing candidate
+		candidate.observationIDs = append(candidate.observationIDs, obs.ID)
+
+		// Add project if not already present
+		found := false
+		for _, p := range candidate.projects {
+			if p == obs.Project {
+				found = true
+				break
+			}
+		}
+		if !found {
+			candidate.projects = append(candidate.projects, obs.Project)
+		}
+		candidate.lastSeenEpoch = time.Now().UnixMilli()
+
+		// Check if candidate should be promoted to pattern
+		if len(candidate.observationIDs) >= d.config.MinFrequencyForPattern {
+			pattern, err := d.promoteCandidate(ctx, candidateKey, candidate)
+			if err != nil {
+				log.Warn().Err(err).Msg("Failed to promote candidate to pattern")
+			} else {
+				result.MatchedPattern = pattern
+				result.IsNewPattern = true
+				log.Info().
+					Str("pattern_name", pattern.Name).
+					Int("frequency", pattern.Frequency).
+					Msg("New pattern detected and stored")
+			}
+		}
+	} else {
+		// Create new candidate
+		patternType := models.DetectPatternType(obs.Concepts, obs.Title.String, obs.Narrative.String)
+		d.candidates[candidateKey] = &candidatePattern{
+			signature:      signature,
+			observationIDs: []int64{obs.ID},
+			projects:       []string{obs.Project},
+			patternType:    patternType,
+			title:          obs.Title.String,
+			lastSeenEpoch:  time.Now().UnixMilli(),
+		}
+		log.Debug().
+			Str("candidate_key", candidateKey).
+			Strs("signature", signature).
+			Msg("New pattern candidate created")
+	}
+
+	return result, nil
+}
+
+// promoteCandidate converts a candidate to a stored pattern.
+func (d *Detector) promoteCandidate(ctx context.Context, key string, candidate *candidatePattern) (*models.Pattern, error) {
+	// Generate pattern name from signature
+	name := generatePatternName(candidate.patternType, candidate.signature, candidate.title)
+
+	// Create base pattern using NewPattern with first observation
+	firstProject := ""
+	if len(candidate.projects) > 0 {
+		firstProject = candidate.projects[0]
+	}
+	var firstObsID int64
+	if len(candidate.observationIDs) > 0 {
+		firstObsID = candidate.observationIDs[0]
+	}
+	pattern := models.NewPattern(
+		name,
+		candidate.patternType,
+		"Automatically detected pattern from recurring observations",
+		candidate.signature,
+		firstProject,
+		firstObsID,
+	)
+
+	// Add remaining projects and observations
+	for i := 1; i < len(candidate.projects); i++ {
+		pattern.Projects = append(pattern.Projects, candidate.projects[i])
+	}
+	for i := 1; i < len(candidate.observationIDs); i++ {
+		pattern.ObservationIDs = append(pattern.ObservationIDs, candidate.observationIDs[i])
+	}
+	pattern.Frequency = len(candidate.observationIDs)
+
+	id, err := d.patternStore.StorePattern(ctx, pattern)
+	if err != nil {
+		return nil, err
+	}
+	pattern.ID = id
+
+	// Sync to vector store if callback is set
+	if d.syncFunc != nil {
+		d.syncFunc(pattern)
+	}
+
+	// Remove from candidates
+	delete(d.candidates, key)
+
+	return pattern, nil
+}
+
+// AnalyzeRecentObservations analyzes recent observations for pattern detection.
+// This is used for background batch analysis.
+func (d *Detector) AnalyzeRecentObservations(ctx context.Context) error {
+	// Get observations from the last 24 hours that haven't been analyzed
+	observations, err := d.observationStore.GetRecentObservations(ctx, "", 100)
+	if err != nil {
+		return err
+	}
+
+	analyzed := 0
+	patternsFound := 0
+	for _, obs := range observations {
+		result, err := d.AnalyzeObservation(ctx, obs)
+		if err != nil {
+			log.Warn().Err(err).Int64("obs_id", obs.ID).Msg("Failed to analyze observation")
+			continue
+		}
+		analyzed++
+		if result.MatchedPattern != nil {
+			patternsFound++
+		}
+	}
+
+	if analyzed > 0 {
+		log.Info().
+			Int("analyzed", analyzed).
+			Int("patterns_found", patternsFound).
+			Msg("Background pattern analysis completed")
+	}
+
+	// Clean up old candidates (older than 7 days)
+	d.cleanupOldCandidates()
+
+	return nil
+}
+
+// cleanupOldCandidates removes candidates that haven't been seen recently.
+func (d *Detector) cleanupOldCandidates() {
+	d.candidatesMu.Lock()
+	defer d.candidatesMu.Unlock()
+
+	threshold := time.Now().Add(-7 * 24 * time.Hour).UnixMilli()
+	for key, candidate := range d.candidates {
+		if candidate.lastSeenEpoch < threshold {
+			delete(d.candidates, key)
+		}
+	}
+}
+
+// GetPatternInsight returns a formatted insight string for a pattern.
+func (d *Detector) GetPatternInsight(ctx context.Context, patternID int64) (string, error) {
+	pattern, err := d.patternStore.GetPatternByID(ctx, patternID)
+	if err != nil {
+		return "", err
+	}
+
+	return formatPatternInsight(pattern), nil
+}
+
+// DetectionResult contains the result of pattern detection.
+type DetectionResult struct {
+	MatchedPattern *models.Pattern
+	MatchScore     float64
+	IsNewPattern   bool
+}
+
+// generateCandidateKey creates a unique key for a signature.
+func generateCandidateKey(signature []string) string {
+	if len(signature) == 0 {
+		return ""
+	}
+	key := ""
+	for _, s := range signature {
+		key += s + "|"
+	}
+	return key
+}
+
+// generatePatternName creates a human-readable name for a pattern.
+func generatePatternName(patternType models.PatternType, signature []string, title string) string {
+	// Use title if available and meaningful
+	if title != "" && len(title) < 60 {
+		return title
+	}
+
+	// Otherwise generate from type and signature
+	prefix := ""
+	switch patternType {
+	case models.PatternTypeBug:
+		prefix = "Bug Pattern: "
+	case models.PatternTypeRefactor:
+		prefix = "Refactor Pattern: "
+	case models.PatternTypeArchitecture:
+		prefix = "Architecture Pattern: "
+	case models.PatternTypeAntiPattern:
+		prefix = "Anti-Pattern: "
+	case models.PatternTypeBestPractice:
+		prefix = "Best Practice: "
+	}
+
+	// Use first few signature elements
+	if len(signature) > 0 {
+		name := prefix
+		for i, s := range signature {
+			if i >= 3 {
+				break
+			}
+			if i > 0 {
+				name += ", "
+			}
+			name += s
+		}
+		return name
+	}
+
+	return prefix + "Unnamed"
+}
+
+// formatPatternInsight creates a human-readable insight from a pattern.
+func formatPatternInsight(pattern *models.Pattern) string {
+	insight := "I've encountered this pattern " +
+		itoa(pattern.Frequency) + " times"
+
+	if len(pattern.Projects) > 1 {
+		insight += " across " + itoa(len(pattern.Projects)) + " projects"
+	}
+
+	insight += ". "
+
+	if pattern.Recommendation.Valid && pattern.Recommendation.String != "" {
+		insight += "What works: " + pattern.Recommendation.String
+	} else {
+		switch pattern.Type {
+		case models.PatternTypeBug:
+			insight += "This appears to be a recurring bug pattern."
+		case models.PatternTypeAntiPattern:
+			insight += "This is an identified anti-pattern to avoid."
+		case models.PatternTypeBestPractice:
+			insight += "This is a validated best practice."
+		default:
+			insight += "This is a recognized pattern in the codebase."
+		}
+	}
+
+	return insight
+}
+
+// itoa converts int to string without importing strconv.
+func itoa(n int) string {
+	if n == 0 {
+		return "0"
+	}
+	negative := false
+	if n < 0 {
+		negative = true
+		n = -n
+	}
+	var digits []byte
+	for n > 0 {
+		digits = append([]byte{byte('0' + n%10)}, digits...)
+		n /= 10
+	}
+	if negative {
+		digits = append([]byte{'-'}, digits...)
+	}
+	return string(digits)
+}

internal/pattern/detector_test.go

@@ -0,0 +1,450 @@
+package pattern
+
+import (
+	"context"
+	"database/sql"
+	"os"
+	"testing"
+	"time"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/db/sqlite"
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+)
+
+func TestNewDetector(t *testing.T) {
+	store := setupTestStore(t)
+	defer store.Close()
+
+	patternStore := sqlite.NewPatternStore(store)
+	observationStore := sqlite.NewObservationStore(store)
+	config := DefaultConfig()
+
+	detector := NewDetector(patternStore, observationStore, config)
+	if detector == nil {
+		t.Fatal("Expected non-nil detector")
+	}
+
+	if detector.config.MinMatchScore != config.MinMatchScore {
+		t.Errorf("Expected MinMatchScore %f, got %f",
+			config.MinMatchScore, detector.config.MinMatchScore)
+	}
+}
+
+func TestDetector_StartStop(t *testing.T) {
+	store := setupTestStore(t)
+	defer store.Close()
+
+	patternStore := sqlite.NewPatternStore(store)
+	observationStore := sqlite.NewObservationStore(store)
+	config := DefaultConfig()
+	config.AnalysisInterval = 100 * time.Millisecond // Short interval for testing
+
+	detector := NewDetector(patternStore, observationStore, config)
+
+	// Start
+	detector.Start()
+
+	// Wait a bit to ensure background goroutine is running
+	time.Sleep(50 * time.Millisecond)
+
+	// Stop
+	detector.Stop()
+
+	// Verify we can stop without hanging
+	// (if this test hangs, the Stop logic is broken)
+}
+
+func TestDetector_AnalyzeObservation_NewCandidate(t *testing.T) {
+	store := setupTestStore(t)
+	defer store.Close()
+
+	patternStore := sqlite.NewPatternStore(store)
+	observationStore := sqlite.NewObservationStore(store)
+	config := DefaultConfig()
+	config.MinFrequencyForPattern = 2
+
+	detector := NewDetector(patternStore, observationStore, config)
+	ctx := context.Background()
+
+	// Create an observation
+	obs := createTestObservation(1, "nil-handling", []string{"nil", "error-handling"})
+
+	// Analyze first observation
+	result, err := detector.AnalyzeObservation(ctx, obs)
+	if err != nil {
+		t.Fatalf("AnalyzeObservation() error = %v", err)
+	}
+
+	// First observation should create a candidate, not a pattern
+	if result.MatchedPattern != nil {
+		t.Errorf("Expected no pattern match for first observation")
+	}
+	if result.IsNewPattern {
+		t.Errorf("Expected IsNewPattern to be false for first observation")
+	}
+}
+
+func TestDetector_AnalyzeObservation_PromoteToPattern(t *testing.T) {
+	store := setupTestStore(t)
+	defer store.Close()
+
+	patternStore := sqlite.NewPatternStore(store)
+	observationStore := sqlite.NewObservationStore(store)
+	config := DefaultConfig()
+	config.MinFrequencyForPattern = 2
+
+	detector := NewDetector(patternStore, observationStore, config)
+	ctx := context.Background()
+
+	// Create two similar observations
+	obs1 := createTestObservation(1, "Nil pointer handling", []string{"nil", "error-handling"})
+	obs2 := createTestObservation(2, "Nil pointer handling", []string{"nil", "error-handling"})
+
+	// Analyze first observation
+	_, err := detector.AnalyzeObservation(ctx, obs1)
+	if err != nil {
+		t.Fatalf("AnalyzeObservation(obs1) error = %v", err)
+	}
+
+	// Analyze second observation - should promote to pattern
+	result, err := detector.AnalyzeObservation(ctx, obs2)
+	if err != nil {
+		t.Fatalf("AnalyzeObservation(obs2) error = %v", err)
+	}
+
+	if result.MatchedPattern == nil {
+		t.Fatal("Expected pattern to be created after second occurrence")
+	}
+	if !result.IsNewPattern {
+		t.Errorf("Expected IsNewPattern to be true for newly promoted pattern")
+	}
+	if result.MatchedPattern.Frequency != 2 {
+		t.Errorf("Expected frequency 2, got %d", result.MatchedPattern.Frequency)
+	}
+}
+
+func TestDetector_AnalyzeObservation_MatchExisting(t *testing.T) {
+	store := setupTestStore(t)
+	defer store.Close()
+
+	patternStore := sqlite.NewPatternStore(store)
+	observationStore := sqlite.NewObservationStore(store)
+	config := DefaultConfig()
+
+	detector := NewDetector(patternStore, observationStore, config)
+	ctx := context.Background()
+
+	// Create an existing pattern
+	pattern := &models.Pattern{
+		Name:           "Existing Pattern",
+		Type:           models.PatternTypeBug,
+		Signature:      []string{"nil", "error-handling", "pointer"},
+		Frequency:      5,
+		Projects:       []string{"proj1"},
+		ObservationIDs: []int64{1, 2, 3, 4, 5},
+		Status:         models.PatternStatusActive,
+		Confidence:     0.7,
+		LastSeenAt:     time.Now().Format(time.RFC3339),
+		LastSeenEpoch:  time.Now().UnixMilli(),
+		CreatedAt:      time.Now().Format(time.RFC3339),
+		CreatedAtEpoch: time.Now().UnixMilli(),
+	}
+	patternStore.StorePattern(ctx, pattern)
+
+	// Create observation with similar signature
+	obs := createTestObservation(10, "Nil check", []string{"nil", "error-handling"})
+
+	// Analyze - should match existing pattern
+	result, err := detector.AnalyzeObservation(ctx, obs)
+	if err != nil {
+		t.Fatalf("AnalyzeObservation() error = %v", err)
+	}
+
+	if result.MatchedPattern == nil {
+		t.Fatal("Expected to match existing pattern")
+	}
+	if result.IsNewPattern {
+		t.Errorf("Expected IsNewPattern to be false for existing pattern")
+	}
+	if result.MatchScore < 0.3 {
+		t.Errorf("Expected match score >= 0.3, got %f", result.MatchScore)
+	}
+}
+
+func TestDetector_AnalyzeObservation_NoMatch(t *testing.T) {
+	store := setupTestStore(t)
+	defer store.Close()
+
+	patternStore := sqlite.NewPatternStore(store)
+	observationStore := sqlite.NewObservationStore(store)
+	config := DefaultConfig()
+	config.MinMatchScore = 0.5 // Higher threshold
+
+	detector := NewDetector(patternStore, observationStore, config)
+	ctx := context.Background()
+
+	// Create an existing pattern with specific signature
+	pattern := &models.Pattern{
+		Name:           "Specific Pattern",
+		Type:           models.PatternTypeBug,
+		Signature:      []string{"database", "connection", "pool"},
+		Frequency:      3,
+		Projects:       []string{"proj1"},
+		ObservationIDs: []int64{1, 2, 3},
+		Status:         models.PatternStatusActive,
+		Confidence:     0.6,
+		LastSeenAt:     time.Now().Format(time.RFC3339),
+		LastSeenEpoch:  time.Now().UnixMilli(),
+		CreatedAt:      time.Now().Format(time.RFC3339),
+		CreatedAtEpoch: time.Now().UnixMilli(),
+	}
+	patternStore.StorePattern(ctx, pattern)
+
+	// Create observation with completely different signature
+	obs := createTestObservation(10, "UI Component", []string{"frontend", "react", "component"})
+
+	// Analyze - should not match
+	result, err := detector.AnalyzeObservation(ctx, obs)
+	if err != nil {
+		t.Fatalf("AnalyzeObservation() error = %v", err)
+	}
+
+	if result.MatchedPattern != nil {
+		t.Errorf("Expected no match for unrelated observation")
+	}
+}
+
+func TestDetector_CandidateCleanup(t *testing.T) {
+	store := setupTestStore(t)
+	defer store.Close()
+
+	patternStore := sqlite.NewPatternStore(store)
+	observationStore := sqlite.NewObservationStore(store)
+	config := DefaultConfig()
+	config.MinFrequencyForPattern = 3 // Higher threshold
+
+	detector := NewDetector(patternStore, observationStore, config)
+
+	// Add an old candidate manually
+	oldKey := "old|candidate|"
+	detector.candidates[oldKey] = &candidatePattern{
+		signature:      []string{"old", "candidate"},
+		observationIDs: []int64{1},
+		projects:       []string{"proj1"},
+		patternType:    models.PatternTypeBug,
+		title:          "Old Candidate",
+		lastSeenEpoch:  time.Now().Add(-8 * 24 * time.Hour).UnixMilli(), // 8 days ago
+	}
+
+	// Add a recent candidate
+	recentKey := "recent|candidate|"
+	detector.candidates[recentKey] = &candidatePattern{
+		signature:      []string{"recent", "candidate"},
+		observationIDs: []int64{2},
+		projects:       []string{"proj1"},
+		patternType:    models.PatternTypeBug,
+		title:          "Recent Candidate",
+		lastSeenEpoch:  time.Now().UnixMilli(),
+	}
+
+	// Run cleanup
+	detector.cleanupOldCandidates()
+
+	// Old candidate should be removed
+	if _, exists := detector.candidates[oldKey]; exists {
+		t.Errorf("Expected old candidate to be cleaned up")
+	}
+
+	// Recent candidate should remain
+	if _, exists := detector.candidates[recentKey]; !exists {
+		t.Errorf("Expected recent candidate to remain")
+	}
+}
+
+func TestDetector_GetPatternInsight(t *testing.T) {
+	store := setupTestStore(t)
+	defer store.Close()
+
+	patternStore := sqlite.NewPatternStore(store)
+	observationStore := sqlite.NewObservationStore(store)
+	config := DefaultConfig()
+
+	detector := NewDetector(patternStore, observationStore, config)
+	ctx := context.Background()
+
+	// Create pattern with recommendation
+	pattern := &models.Pattern{
+		Name:           "Insight Test Pattern",
+		Type:           models.PatternTypeBestPractice,
+		Signature:      []string{"test"},
+		Recommendation: sql.NullString{String: "Always write tests first", Valid: true},
+		Frequency:      12,
+		Projects:       []string{"proj1", "proj2", "proj3"},
+		ObservationIDs: []int64{1},
+		Status:         models.PatternStatusActive,
+		Confidence:     0.8,
+		LastSeenAt:     time.Now().Format(time.RFC3339),
+		LastSeenEpoch:  time.Now().UnixMilli(),
+		CreatedAt:      time.Now().Format(time.RFC3339),
+		CreatedAtEpoch: time.Now().UnixMilli(),
+	}
+	id, _ := patternStore.StorePattern(ctx, pattern)
+
+	// Get insight
+	insight, err := detector.GetPatternInsight(ctx, id)
+	if err != nil {
+		t.Fatalf("GetPatternInsight() error = %v", err)
+	}
+
+	// Verify insight contains expected elements
+	if insight == "" {
+		t.Error("Expected non-empty insight")
+	}
+	if !containsString(insight, "12 times") {
+		t.Errorf("Expected insight to contain frequency, got: %s", insight)
+	}
+	if !containsString(insight, "3 projects") {
+		t.Errorf("Expected insight to contain project count, got: %s", insight)
+	}
+	if !containsString(insight, "Always write tests first") {
+		t.Errorf("Expected insight to contain recommendation, got: %s", insight)
+	}
+}
+
+func TestDefaultConfig(t *testing.T) {
+	config := DefaultConfig()
+
+	if config.MinMatchScore <= 0 || config.MinMatchScore > 1 {
+		t.Errorf("Invalid MinMatchScore: %f", config.MinMatchScore)
+	}
+	if config.MinFrequencyForPattern < 1 {
+		t.Errorf("Invalid MinFrequencyForPattern: %d", config.MinFrequencyForPattern)
+	}
+	if config.AnalysisInterval <= 0 {
+		t.Errorf("Invalid AnalysisInterval: %v", config.AnalysisInterval)
+	}
+	if config.MaxPatternsToTrack <= 0 {
+		t.Errorf("Invalid MaxPatternsToTrack: %d", config.MaxPatternsToTrack)
+	}
+}
+
+func TestGeneratePatternName(t *testing.T) {
+	tests := []struct {
+		patternType models.PatternType
+		title       string
+		wantPrefix  string
+		signature   []string
+	}{
+		{patternType: models.PatternTypeBug, signature: []string{"nil", "error"}, title: "", wantPrefix: "Bug Pattern:"},
+		{patternType: models.PatternTypeRefactor, signature: []string{"extract"}, title: "", wantPrefix: "Refactor Pattern:"},
+		{patternType: models.PatternTypeArchitecture, signature: []string{"service"}, title: "", wantPrefix: "Architecture Pattern:"},
+		{patternType: models.PatternTypeAntiPattern, signature: []string{"god-class"}, title: "", wantPrefix: "Anti-Pattern:"},
+		{patternType: models.PatternTypeBestPractice, signature: []string{"testing"}, title: "", wantPrefix: "Best Practice:"},
+		{patternType: models.PatternTypeBug, signature: []string{}, title: "Short Title", wantPrefix: "Short Title"}, // Use title directly
+	}
+
+	for _, tt := range tests {
+		name := generatePatternName(tt.patternType, tt.signature, tt.title)
+		if !hasPrefix(name, tt.wantPrefix) {
+			t.Errorf("generatePatternName(%v, %v, %q) = %q, want prefix %q",
+				tt.patternType, tt.signature, tt.title, name, tt.wantPrefix)
+		}
+	}
+}
+
+func TestFormatPatternInsight(t *testing.T) {
+	// Pattern without recommendation
+	pattern1 := &models.Pattern{
+		Type:      models.PatternTypeBug,
+		Frequency: 5,
+		Projects:  []string{"proj1"},
+	}
+	insight1 := formatPatternInsight(pattern1)
+	if !containsString(insight1, "5 times") {
+		t.Errorf("Expected insight to contain frequency")
+	}
+	if !containsString(insight1, "recurring bug pattern") {
+		t.Errorf("Expected bug pattern description")
+	}
+
+	// Pattern with recommendation
+	pattern2 := &models.Pattern{
+		Type:           models.PatternTypeBestPractice,
+		Frequency:      10,
+		Projects:       []string{"proj1", "proj2"},
+		Recommendation: sql.NullString{String: "Do this", Valid: true},
+	}
+	insight2 := formatPatternInsight(pattern2)
+	if !containsString(insight2, "10 times") {
+		t.Errorf("Expected insight to contain frequency")
+	}
+	if !containsString(insight2, "2 projects") {
+		t.Errorf("Expected insight to contain project count")
+	}
+	if !containsString(insight2, "Do this") {
+		t.Errorf("Expected insight to contain recommendation")
+	}
+}
+
+// Helper functions
+
+func setupTestStore(t *testing.T) *sqlite.Store {
+	t.Helper()
+
+	// Create temp database file
+	tmpFile, err := os.CreateTemp("", "pattern_test_*.db")
+	if err != nil {
+		t.Fatalf("Failed to create temp file: %v", err)
+	}
+	tmpFile.Close()
+
+	t.Cleanup(func() {
+		os.Remove(tmpFile.Name())
+	})
+
+	store, err := sqlite.NewStore(sqlite.StoreConfig{
+		Path:     tmpFile.Name(),
+		MaxConns: 1,
+		WALMode:  true,
+	})
+	if err != nil {
+		// Check if this is an FTS5 related error
+		if containsString(err.Error(), "fts5") || containsString(err.Error(), "no such module") {
+			t.Skip("FTS5 not available in this SQLite build")
+		}
+		t.Fatalf("Failed to create store: %v", err)
+	}
+
+	return store
+}
+
+func createTestObservation(id int64, title string, concepts []string) *models.Observation {
+	return &models.Observation{
+		ID:             id,
+		SDKSessionID:   "test-session",
+		Project:        "test-project",
+		Scope:          models.ScopeProject,
+		Type:           models.ObsTypeBugfix,
+		Title:          sql.NullString{String: title, Valid: true},
+		Narrative:      sql.NullString{String: "Test narrative", Valid: true},
+		Concepts:       concepts,
+		CreatedAt:      time.Now().Format(time.RFC3339),
+		CreatedAtEpoch: time.Now().UnixMilli(),
+	}
+}
+
+func containsString(s, substr string) bool {
+	for i := 0; i <= len(s)-len(substr); i++ {
+		if s[i:i+len(substr)] == substr {
+			return true
+		}
+	}
+	return false
+}
+
+func hasPrefix(s, prefix string) bool {
+	if len(s) < len(prefix) {
+		return false
+	}
+	return s[:len(prefix)] == prefix
+}

internal/reranking/assets.go

@@ -0,0 +1,14 @@
+// Package reranking provides cross-encoder reranking for search results.
+package reranking
+
+import (
+	_ "embed"
+)
+
+// Cross-encoder model and tokenizer files - embedded for all platforms
+//
+//go:embed assets/model.onnx
+var crossEncoderModelData []byte
+
+//go:embed assets/tokenizer.json
+var crossEncoderTokenizerData []byte

internal/reranking/assets/tokenizer_config.json

@@ -0,0 +1,57 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "101": {
+      "content": "[CLS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "102": {
+      "content": "[SEP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "103": {
+      "content": "[MASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": true,
+  "cls_token": "[CLS]",
+  "do_basic_tokenize": true,
+  "do_lower_case": true,
+  "mask_token": "[MASK]",
+  "model_max_length": 512,
+  "never_split": null,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "unk_token": "[UNK]"
+}

internal/reranking/service.go

@@ -0,0 +1,380 @@
+// Package reranking provides cross-encoder reranking for search results.
+// Uses MS-MARCO MiniLM L6 v2 cross-encoder model for relevance scoring.
+package reranking
+
+import (
+	"bytes"
+	"fmt"
+	"math"
+	"sort"
+	"sync"
+
+	"github.com/rs/zerolog/log"
+	"github.com/sugarme/tokenizer"
+	"github.com/sugarme/tokenizer/pretrained"
+	ort "github.com/yalue/onnxruntime_go"
+)
+
+const (
+	// ModelName is the human-readable name for the cross-encoder model
+	ModelName = "ms-marco-MiniLM-L6-v2"
+	// ModelVersion is the short version string for identification
+	ModelVersion = "msmarco-v2"
+	// MaxSequenceLength is the maximum combined query+document token length
+	MaxSequenceLength = 512
+	// DefaultCandidateLimit is the default number of candidates to rerank
+	DefaultCandidateLimit = 100
+	// DefaultResultLimit is the default number of results to return after reranking
+	DefaultResultLimit = 10
+)
+
+// Candidate represents a search result candidate for reranking.
+type Candidate struct {
+	Metadata   map[string]any
+	RerankInfo map[string]float64
+	ID         string
+	Content    string
+	Score      float64
+}
+
+// RerankResult represents a reranked search result.
+type RerankResult struct {
+	Metadata        map[string]any
+	ID              string
+	Content         string
+	OriginalScore   float64
+	RerankScore     float64
+	CombinedScore   float64
+	OriginalRank    int
+	RerankRank      int
+	RankImprovement int
+}
+
+// Service provides cross-encoder reranking functionality.
+type Service struct {
+	tk      *tokenizer.Tokenizer
+	session *ort.DynamicAdvancedSession
+	mu      sync.Mutex
+
+	// Weight for combining scores: combined = alpha*rerank + (1-alpha)*original
+	// Default 0.7 favors cross-encoder score
+	Alpha float64
+}
+
+// Config holds configuration for the reranking service.
+type Config struct {
+	// Alpha is the weight for combining scores (0.0-1.0)
+	// Higher values favor cross-encoder scores, lower values favor bi-encoder scores
+	Alpha float64
+}
+
+// DefaultConfig returns sensible defaults for reranking.
+func DefaultConfig() Config {
+	return Config{
+		Alpha: 0.7, // Favor cross-encoder by default
+	}
+}
+
+// NewService creates a new cross-encoder reranking service.
+// Note: ONNX runtime must be initialized before calling this (via embedding.NewService).
+func NewService(cfg Config) (*Service, error) {
+	// Load tokenizer from embedded data
+	tk, err := pretrained.FromReader(bytes.NewReader(crossEncoderTokenizerData))
+	if err != nil {
+		return nil, fmt.Errorf("load cross-encoder tokenizer: %w", err)
+	}
+
+	// Configure tokenizer for sequence classification (pairs)
+	tk.WithTruncation(&tokenizer.TruncationParams{
+		MaxLength: MaxSequenceLength,
+		Strategy:  tokenizer.LongestFirst,
+		Stride:    0,
+	})
+
+	// Cross-encoder outputs a single logit for relevance scoring
+	inputNames := []string{"input_ids", "attention_mask", "token_type_ids"}
+	outputNames := []string{"logits"}
+
+	session, err := ort.NewDynamicAdvancedSessionWithONNXData(
+		crossEncoderModelData,
+		inputNames,
+		outputNames,
+		nil,
+	)
+	if err != nil {
+		return nil, fmt.Errorf("create cross-encoder ONNX session: %w", err)
+	}
+
+	alpha := cfg.Alpha
+	if alpha <= 0 || alpha > 1 {
+		alpha = 0.7
+	}
+
+	return &Service{
+		tk:      tk,
+		session: session,
+		Alpha:   alpha,
+	}, nil
+}
+
+// Rerank reranks candidates using the cross-encoder model.
+// Takes a query and list of candidates, returns reranked results.
+func (s *Service) Rerank(query string, candidates []Candidate, limit int) ([]RerankResult, error) {
+	if len(candidates) == 0 {
+		return nil, nil
+	}
+
+	if limit <= 0 {
+		limit = DefaultResultLimit
+	}
+
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	// Score all query-document pairs
+	scores, err := s.scoreAll(query, candidates)
+	if err != nil {
+		return nil, fmt.Errorf("score candidates: %w", err)
+	}
+
+	// Build results with combined scores
+	results := make([]RerankResult, len(candidates))
+	for i, c := range candidates {
+		// Normalize cross-encoder score to 0-1 range using sigmoid
+		normalizedRerank := sigmoid(scores[i])
+
+		results[i] = RerankResult{
+			ID:            c.ID,
+			Content:       c.Content,
+			OriginalScore: c.Score,
+			RerankScore:   normalizedRerank,
+			CombinedScore: s.Alpha*normalizedRerank + (1-s.Alpha)*c.Score,
+			Metadata:      c.Metadata,
+			OriginalRank:  i + 1,
+		}
+	}
+
+	// Sort by combined score (descending)
+	sort.Slice(results, func(i, j int) bool {
+		return results[i].CombinedScore > results[j].CombinedScore
+	})
+
+	// Assign rerank positions and calculate improvement
+	for i := range results {
+		results[i].RerankRank = i + 1
+		results[i].RankImprovement = results[i].OriginalRank - results[i].RerankRank
+	}
+
+	// Apply limit
+	if len(results) > limit {
+		results = results[:limit]
+	}
+
+	log.Debug().
+		Int("candidates", len(candidates)).
+		Int("returned", len(results)).
+		Float64("alpha", s.Alpha).
+		Msg("Cross-encoder reranking completed")
+
+	return results, nil
+}
+
+// RerankByScore reranks candidates and returns sorted by pure cross-encoder score.
+// Useful when you want to completely replace bi-encoder ranking.
+func (s *Service) RerankByScore(query string, candidates []Candidate, limit int) ([]RerankResult, error) {
+	if len(candidates) == 0 {
+		return nil, nil
+	}
+
+	if limit <= 0 {
+		limit = DefaultResultLimit
+	}
+
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	scores, err := s.scoreAll(query, candidates)
+	if err != nil {
+		return nil, fmt.Errorf("score candidates: %w", err)
+	}
+
+	results := make([]RerankResult, len(candidates))
+	for i, c := range candidates {
+		normalizedRerank := sigmoid(scores[i])
+		results[i] = RerankResult{
+			ID:            c.ID,
+			Content:       c.Content,
+			OriginalScore: c.Score,
+			RerankScore:   normalizedRerank,
+			CombinedScore: normalizedRerank, // Use pure rerank score
+			Metadata:      c.Metadata,
+			OriginalRank:  i + 1,
+		}
+	}
+
+	// Sort by rerank score only
+	sort.Slice(results, func(i, j int) bool {
+		return results[i].RerankScore > results[j].RerankScore
+	})
+
+	for i := range results {
+		results[i].RerankRank = i + 1
+		results[i].RankImprovement = results[i].OriginalRank - results[i].RerankRank
+	}
+
+	if len(results) > limit {
+		results = results[:limit]
+	}
+
+	return results, nil
+}
+
+// scoreAll scores all query-document pairs using the cross-encoder.
+// Returns raw logits (before sigmoid normalization).
+func (s *Service) scoreAll(query string, candidates []Candidate) ([]float64, error) {
+	batchSize := len(candidates)
+
+	// Tokenize all query-document pairs
+	pairs := make([]tokenizer.EncodeInput, batchSize)
+	for i, c := range candidates {
+		// Cross-encoder takes query and document as a pair
+		pairs[i] = tokenizer.NewDualEncodeInput(
+			tokenizer.NewRawInputSequence(query),
+			tokenizer.NewRawInputSequence(c.Content),
+		)
+	}
+
+	encodings, err := s.tk.EncodeBatch(pairs, true)
+	if err != nil {
+		return nil, fmt.Errorf("tokenize pairs: %w", err)
+	}
+
+	// Find max sequence length
+	seqLength := 0
+	for _, enc := range encodings {
+		if len(enc.Ids) > seqLength {
+			seqLength = len(enc.Ids)
+		}
+	}
+	if seqLength > MaxSequenceLength {
+		seqLength = MaxSequenceLength
+	}
+
+	inputShape := ort.NewShape(int64(batchSize), int64(seqLength))
+
+	// Create input tensors
+	inputIdsData := make([]int64, batchSize*seqLength)
+	attentionMaskData := make([]int64, batchSize*seqLength)
+	tokenTypeIdsData := make([]int64, batchSize*seqLength)
+
+	for b := 0; b < batchSize; b++ {
+		copyLen := len(encodings[b].Ids)
+		if copyLen > seqLength {
+			copyLen = seqLength
+		}
+		for i := 0; i < copyLen; i++ {
+			inputIdsData[b*seqLength+i] = int64(encodings[b].Ids[i])
+		}
+
+		copyLen = len(encodings[b].AttentionMask)
+		if copyLen > seqLength {
+			copyLen = seqLength
+		}
+		for i := 0; i < copyLen; i++ {
+			attentionMaskData[b*seqLength+i] = int64(encodings[b].AttentionMask[i])
+		}
+
+		copyLen = len(encodings[b].TypeIds)
+		if copyLen > seqLength {
+			copyLen = seqLength
+		}
+		for i := 0; i < copyLen; i++ {
+			tokenTypeIdsData[b*seqLength+i] = int64(encodings[b].TypeIds[i])
+		}
+	}
+
+	inputIdsTensor, err := ort.NewTensor(inputShape, inputIdsData)
+	if err != nil {
+		return nil, fmt.Errorf("create input_ids tensor: %w", err)
+	}
+	defer func() { _ = inputIdsTensor.Destroy() }()
+
+	attentionMaskTensor, err := ort.NewTensor(inputShape, attentionMaskData)
+	if err != nil {
+		return nil, fmt.Errorf("create attention_mask tensor: %w", err)
+	}
+	defer func() { _ = attentionMaskTensor.Destroy() }()
+
+	tokenTypeIdsTensor, err := ort.NewTensor(inputShape, tokenTypeIdsData)
+	if err != nil {
+		return nil, fmt.Errorf("create token_type_ids tensor: %w", err)
+	}
+	defer func() { _ = tokenTypeIdsTensor.Destroy() }()
+
+	// Cross-encoder outputs [batch, 1] logits
+	outputShape := ort.NewShape(int64(batchSize), 1)
+	outputTensor, err := ort.NewEmptyTensor[float32](outputShape)
+	if err != nil {
+		return nil, fmt.Errorf("create output tensor: %w", err)
+	}
+	defer func() { _ = outputTensor.Destroy() }()
+
+	// Run inference
+	inputTensors := []ort.Value{inputIdsTensor, attentionMaskTensor, tokenTypeIdsTensor}
+	outputTensors := []ort.Value{outputTensor}
+
+	if err := s.session.Run(inputTensors, outputTensors); err != nil {
+		return nil, fmt.Errorf("run cross-encoder inference: %w", err)
+	}
+
+	// Extract scores
+	flatOutput := outputTensor.GetData()
+	scores := make([]float64, batchSize)
+	for i := 0; i < batchSize; i++ {
+		scores[i] = float64(flatOutput[i])
+	}
+
+	return scores, nil
+}
+
+// Score scores a single query-document pair.
+// Returns the raw cross-encoder logit and normalized score.
+func (s *Service) Score(query, document string) (rawScore, normalizedScore float64, err error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	scores, err := s.scoreAll(query, []Candidate{{Content: document}})
+	if err != nil {
+		return 0, 0, err
+	}
+
+	rawScore = scores[0]
+	normalizedScore = sigmoid(rawScore)
+	return rawScore, normalizedScore, nil
+}
+
+// Close releases model resources.
+func (s *Service) Close() error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	if s.session != nil {
+		if err := s.session.Destroy(); err != nil {
+			return fmt.Errorf("destroy cross-encoder session: %w", err)
+		}
+		s.session = nil
+	}
+
+	return nil
+}
+
+// sigmoid applies the sigmoid function to normalize scores to 0-1 range.
+func sigmoid(x float64) float64 {
+	if x > 20 {
+		return 1.0
+	}
+	if x < -20 {
+		return 0.0
+	}
+	return 1.0 / (1.0 + math.Exp(-x))
+}

internal/reranking/service_test.go

@@ -0,0 +1,448 @@
+package reranking
+
+import (
+	"sync"
+	"testing"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/embedding"
+)
+
+// initONNX initializes the ONNX runtime via the embedding service.
+// Must be called before creating reranking service.
+func initONNX(t *testing.T) func() {
+	t.Helper()
+	embSvc, err := embedding.NewService()
+	if err != nil {
+		t.Fatalf("Failed to initialize ONNX via embedding service: %v", err)
+	}
+	return func() {
+		embSvc.Close()
+	}
+}
+
+// TestSigmoid tests the sigmoid normalization function.
+func TestSigmoid(t *testing.T) {
+	tests := []struct {
+		name    string
+		input   float64
+		wantMin float64
+		wantMax float64
+	}{
+		{"positive large", 10, 0.9999, 1.0},
+		{"positive small", 1, 0.7, 0.8},
+		{"zero", 0, 0.4999, 0.5001},
+		{"negative small", -1, 0.2, 0.3},
+		{"negative large", -10, 0, 0.0001},
+		{"very positive", 25, 0.999999, 1.0},
+		{"very negative", -25, 0, 0.000001},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := sigmoid(tt.input)
+			if got < tt.wantMin || got > tt.wantMax {
+				t.Errorf("sigmoid(%v) = %v, want in range [%v, %v]",
+					tt.input, got, tt.wantMin, tt.wantMax)
+			}
+		})
+	}
+}
+
+// TestDefaultConfig tests the default configuration values.
+func TestDefaultConfig(t *testing.T) {
+	cfg := DefaultConfig()
+
+	if cfg.Alpha < 0 || cfg.Alpha > 1 {
+		t.Errorf("DefaultConfig().Alpha = %v, want in range [0, 1]", cfg.Alpha)
+	}
+	if cfg.Alpha != 0.7 {
+		t.Errorf("DefaultConfig().Alpha = %v, want 0.7", cfg.Alpha)
+	}
+}
+
+// TestNewService tests service creation.
+func TestNewService(t *testing.T) {
+	cleanup := initONNX(t)
+	defer cleanup()
+
+	cfg := DefaultConfig()
+	svc, err := NewService(cfg)
+	if err != nil {
+		t.Fatalf("NewService() error = %v", err)
+	}
+	defer svc.Close()
+
+	if svc == nil {
+		t.Fatal("NewService() returned nil")
+	}
+
+	if svc.Alpha != cfg.Alpha {
+		t.Errorf("Service.Alpha = %v, want %v", svc.Alpha, cfg.Alpha)
+	}
+}
+
+// TestScore tests single pair scoring.
+func TestScore(t *testing.T) {
+	cleanup := initONNX(t)
+	defer cleanup()
+
+	cfg := DefaultConfig()
+	svc, err := NewService(cfg)
+	if err != nil {
+		t.Fatalf("NewService() error = %v", err)
+	}
+	defer svc.Close()
+
+	query := "What is the capital of France?"
+	relevant := "Paris is the capital and largest city of France."
+	irrelevant := "Dogs are popular pets known for their loyalty."
+
+	// Score relevant document
+	_, relevantNorm, err := svc.Score(query, relevant)
+	if err != nil {
+		t.Fatalf("Score(relevant) error = %v", err)
+	}
+
+	// Score irrelevant document
+	_, irrelevantNorm, err := svc.Score(query, irrelevant)
+	if err != nil {
+		t.Fatalf("Score(irrelevant) error = %v", err)
+	}
+
+	// Relevant document should score higher
+	if relevantNorm <= irrelevantNorm {
+		t.Errorf("Expected relevant (%v) > irrelevant (%v)",
+			relevantNorm, irrelevantNorm)
+	}
+}
+
+// TestRerank tests the reranking functionality.
+func TestRerank(t *testing.T) {
+	cleanup := initONNX(t)
+	defer cleanup()
+
+	cfg := DefaultConfig()
+	svc, err := NewService(cfg)
+	if err != nil {
+		t.Fatalf("NewService() error = %v", err)
+	}
+	defer svc.Close()
+
+	query := "How to handle errors in Go?"
+	candidates := []Candidate{
+		{
+			ID:      "1",
+			Content: "Python exception handling with try/except blocks.",
+			Score:   0.8, // High bi-encoder score but irrelevant
+		},
+		{
+			ID:      "2",
+			Content: "Go error handling uses explicit return values. Functions return error as the last value.",
+			Score:   0.6, // Lower bi-encoder score but relevant
+		},
+		{
+			ID:      "3",
+			Content: "JavaScript uses Promise.catch for async error handling.",
+			Score:   0.7,
+		},
+	}
+
+	results, err := svc.Rerank(query, candidates, 3)
+	if err != nil {
+		t.Fatalf("Rerank() error = %v", err)
+	}
+
+	if len(results) != 3 {
+		t.Fatalf("Rerank() returned %d results, want 3", len(results))
+	}
+
+	// The Go error handling document should rank higher after reranking
+	var goRank int
+	for i, r := range results {
+		if r.ID == "2" {
+			goRank = i + 1
+			break
+		}
+	}
+
+	if goRank == 0 {
+		t.Error("Go document not found in results")
+	}
+
+	// Verify all results have required fields populated
+	for i, r := range results {
+		if r.ID == "" {
+			t.Errorf("Result %d has empty ID", i)
+		}
+		if r.Content == "" {
+			t.Errorf("Result %d has empty Content", i)
+		}
+		if r.RerankRank != i+1 {
+			t.Errorf("Result %d has RerankRank %d, want %d", i, r.RerankRank, i+1)
+		}
+	}
+}
+
+// TestRerankEmpty tests reranking with empty input.
+func TestRerankEmpty(t *testing.T) {
+	cleanup := initONNX(t)
+	defer cleanup()
+
+	cfg := DefaultConfig()
+	svc, err := NewService(cfg)
+	if err != nil {
+		t.Fatalf("NewService() error = %v", err)
+	}
+	defer svc.Close()
+
+	results, err := svc.Rerank("test query", nil, 10)
+	if err != nil {
+		t.Fatalf("Rerank(nil) error = %v", err)
+	}
+
+	if results != nil {
+		t.Errorf("Rerank(nil) = %v, want nil", results)
+	}
+
+	results, err = svc.Rerank("test query", []Candidate{}, 10)
+	if err != nil {
+		t.Fatalf("Rerank([]) error = %v", err)
+	}
+
+	if results != nil {
+		t.Errorf("Rerank([]) = %v, want nil", results)
+	}
+}
+
+// TestRerankLimit tests that limit is respected.
+func TestRerankLimit(t *testing.T) {
+	cleanup := initONNX(t)
+	defer cleanup()
+
+	cfg := DefaultConfig()
+	svc, err := NewService(cfg)
+	if err != nil {
+		t.Fatalf("NewService() error = %v", err)
+	}
+	defer svc.Close()
+
+	candidates := make([]Candidate, 20)
+	for i := range candidates {
+		candidates[i] = Candidate{
+			ID:      string(rune('A' + i)),
+			Content: "Test document content for ranking.",
+			Score:   0.5,
+		}
+	}
+
+	results, err := svc.Rerank("test query", candidates, 5)
+	if err != nil {
+		t.Fatalf("Rerank() error = %v", err)
+	}
+
+	if len(results) != 5 {
+		t.Errorf("Rerank() returned %d results, want 5", len(results))
+	}
+}
+
+// TestRerankByScore tests pure cross-encoder ranking.
+func TestRerankByScore(t *testing.T) {
+	cleanup := initONNX(t)
+	defer cleanup()
+
+	cfg := DefaultConfig()
+	svc, err := NewService(cfg)
+	if err != nil {
+		t.Fatalf("NewService() error = %v", err)
+	}
+	defer svc.Close()
+
+	query := "machine learning algorithms"
+	candidates := []Candidate{
+		{
+			ID:      "1",
+			Content: "Cooking recipes for Italian pasta dishes.",
+			Score:   0.9, // High original score
+		},
+		{
+			ID:      "2",
+			Content: "Neural networks are a type of machine learning algorithm.",
+			Score:   0.3, // Low original score
+		},
+	}
+
+	results, err := svc.RerankByScore(query, candidates, 2)
+	if err != nil {
+		t.Fatalf("RerankByScore() error = %v", err)
+	}
+
+	// Document 2 should rank first since it's about ML
+	if results[0].ID != "2" {
+		t.Errorf("Expected ML document to rank first, got %v", results[0].ID)
+	}
+
+	// CombinedScore should equal RerankScore when using RerankByScore
+	for _, r := range results {
+		if r.CombinedScore != r.RerankScore {
+			t.Errorf("RerankByScore: CombinedScore (%v) != RerankScore (%v)",
+				r.CombinedScore, r.RerankScore)
+		}
+	}
+}
+
+// TestRankImprovement tests that rank improvement is calculated correctly.
+func TestRankImprovement(t *testing.T) {
+	cleanup := initONNX(t)
+	defer cleanup()
+
+	cfg := DefaultConfig()
+	svc, err := NewService(cfg)
+	if err != nil {
+		t.Fatalf("NewService() error = %v", err)
+	}
+	defer svc.Close()
+
+	// Create candidates where we know the expected reranking
+	candidates := []Candidate{
+		{ID: "A", Content: "Unrelated content about weather forecasting.", Score: 0.9},
+		{ID: "B", Content: "How to fix memory leaks in Go programs.", Score: 0.8},
+		{ID: "C", Content: "More unrelated content about gardening tips.", Score: 0.7},
+	}
+
+	results, err := svc.Rerank("debugging memory issues in Go", candidates, 3)
+	if err != nil {
+		t.Fatalf("Rerank() error = %v", err)
+	}
+
+	for _, r := range results {
+		// RankImprovement = OriginalRank - RerankRank
+		// Positive means moved up, negative means moved down
+		expectedImprovement := r.OriginalRank - r.RerankRank
+		if r.RankImprovement != expectedImprovement {
+			t.Errorf("ID %s: RankImprovement = %d, want %d (orig=%d, new=%d)",
+				r.ID, r.RankImprovement, expectedImprovement,
+				r.OriginalRank, r.RerankRank)
+		}
+	}
+}
+
+// TestConcurrentRerank tests concurrent reranking calls.
+func TestConcurrentRerank(t *testing.T) {
+	cleanup := initONNX(t)
+	defer cleanup()
+
+	cfg := DefaultConfig()
+	svc, err := NewService(cfg)
+	if err != nil {
+		t.Fatalf("NewService() error = %v", err)
+	}
+	defer svc.Close()
+
+	candidates := []Candidate{
+		{ID: "1", Content: "Test document one.", Score: 0.5},
+		{ID: "2", Content: "Test document two.", Score: 0.5},
+	}
+
+	var wg sync.WaitGroup
+	errors := make(chan error, 10)
+
+	for i := 0; i < 10; i++ {
+		wg.Add(1)
+		go func(idx int) {
+			defer wg.Done()
+			_, err := svc.Rerank("concurrent test query", candidates, 2)
+			if err != nil {
+				errors <- err
+			}
+		}(i)
+	}
+
+	wg.Wait()
+	close(errors)
+
+	for err := range errors {
+		t.Errorf("Concurrent Rerank error: %v", err)
+	}
+}
+
+// TestClose tests service cleanup.
+func TestClose(t *testing.T) {
+	cleanup := initONNX(t)
+	defer cleanup()
+
+	cfg := DefaultConfig()
+	svc, err := NewService(cfg)
+	if err != nil {
+		t.Fatalf("NewService() error = %v", err)
+	}
+
+	err = svc.Close()
+	if err != nil {
+		t.Errorf("Close() error = %v", err)
+	}
+
+	// Double close should not panic
+	err = svc.Close()
+	if err != nil {
+		t.Errorf("Close() on closed service error = %v", err)
+	}
+}
+
+// TestMetadataPreserved tests that metadata is preserved through reranking.
+func TestMetadataPreserved(t *testing.T) {
+	cleanup := initONNX(t)
+	defer cleanup()
+
+	cfg := DefaultConfig()
+	svc, err := NewService(cfg)
+	if err != nil {
+		t.Fatalf("NewService() error = %v", err)
+	}
+	defer svc.Close()
+
+	candidates := []Candidate{
+		{
+			ID:       "1",
+			Content:  "Test content.",
+			Score:    0.5,
+			Metadata: map[string]any{"custom": "value1", "num": 42},
+		},
+		{
+			ID:       "2",
+			Content:  "Another test.",
+			Score:    0.5,
+			Metadata: map[string]any{"custom": "value2"},
+		},
+	}
+
+	results, err := svc.Rerank("query", candidates, 2)
+	if err != nil {
+		t.Fatalf("Rerank() error = %v", err)
+	}
+
+	for _, r := range results {
+		if r.Metadata == nil {
+			t.Errorf("Result %s has nil metadata", r.ID)
+			continue
+		}
+
+		// Find original candidate
+		var original *Candidate
+		for i := range candidates {
+			if candidates[i].ID == r.ID {
+				original = &candidates[i]
+				break
+			}
+		}
+
+		if original == nil {
+			t.Errorf("Could not find original for result %s", r.ID)
+			continue
+		}
+
+		// Check metadata preserved
+		if original.Metadata["custom"] != r.Metadata["custom"] {
+			t.Errorf("Metadata not preserved for %s", r.ID)
+		}
+	}
+}

internal/scoring/calculator.go

@@ -0,0 +1,168 @@
+// Package scoring provides importance score calculation for observations.
+package scoring
+
+import (
+	"math"
+	"time"
+
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+)
+
+// Calculator computes importance scores for observations.
+type Calculator struct {
+	config *models.ScoringConfig
+}
+
+// NewCalculator creates a new scoring calculator.
+// If config is nil, uses the default configuration.
+func NewCalculator(config *models.ScoringConfig) *Calculator {
+	if config == nil {
+		config = models.DefaultScoringConfig()
+	}
+	return &Calculator{config: config}
+}
+
+// Calculate computes the importance score for an observation at the given time.
+//
+// The scoring formula:
+//
+//	FinalScore = (BaseScore × TypeWeight × RecencyDecay) + FeedbackContrib + ConceptContrib + RetrievalContrib
+//
+// Where:
+//   - BaseScore = 1.0
+//   - TypeWeight = observation type multiplier (e.g., bugfix=1.3, change=0.9)
+//   - RecencyDecay = 0.5^(age_days / half_life_days) - halves every 7 days by default
+//   - FeedbackContrib = user_feedback × feedback_weight
+//   - ConceptContrib = sum(concept_weights) × concept_weight_factor
+//   - RetrievalContrib = log2(retrieval_count + 1) × 0.1 × retrieval_weight
+func (c *Calculator) Calculate(obs *models.Observation, now time.Time) float64 {
+	// 1. Get base type weight
+	typeWeight := models.TypeBaseScore(obs.Type)
+
+	// 2. Calculate recency decay: 0.5^(age_days / half_life_days)
+	ageDays := now.Sub(time.UnixMilli(obs.CreatedAtEpoch)).Hours() / 24.0
+	if ageDays < 0 {
+		ageDays = 0 // Handle future timestamps gracefully
+	}
+	recencyDecay := math.Pow(0.5, ageDays/c.config.RecencyHalfLifeDays)
+
+	// Core score = 1.0 × type_weight × recency_decay
+	coreScore := 1.0 * typeWeight * recencyDecay
+
+	// 3. User feedback contribution: feedback × weight
+	feedbackContrib := float64(obs.UserFeedback) * c.config.FeedbackWeight
+
+	// 4. Concept boost contribution: sum of matching concept weights × factor
+	conceptBoost := 0.0
+	for _, concept := range obs.Concepts {
+		if weight, ok := c.config.ConceptWeights[concept]; ok {
+			conceptBoost += weight
+		}
+	}
+	conceptContrib := conceptBoost * c.config.ConceptWeight
+
+	// 5. Retrieval boost: log2(count + 1) × 0.1 × weight (diminishing returns)
+	retrievalContrib := 0.0
+	if obs.RetrievalCount > 0 {
+		// log2(count + 1) gives diminishing returns: 1→1, 3→2, 7→3, 15→4, etc.
+		retrievalBoost := math.Log2(float64(obs.RetrievalCount)+1) * 0.1
+		retrievalContrib = retrievalBoost * c.config.RetrievalWeight
+	}
+
+	// Final score with minimum threshold
+	finalScore := coreScore + feedbackContrib + conceptContrib + retrievalContrib
+	if finalScore < c.config.MinScore {
+		finalScore = c.config.MinScore
+	}
+
+	return finalScore
+}
+
+// CalculateComponents returns the individual components of the importance score.
+// Useful for debugging and explaining scores to users.
+func (c *Calculator) CalculateComponents(obs *models.Observation, now time.Time) ScoreComponents {
+	typeWeight := models.TypeBaseScore(obs.Type)
+
+	ageDays := now.Sub(time.UnixMilli(obs.CreatedAtEpoch)).Hours() / 24.0
+	if ageDays < 0 {
+		ageDays = 0
+	}
+	recencyDecay := math.Pow(0.5, ageDays/c.config.RecencyHalfLifeDays)
+
+	coreScore := 1.0 * typeWeight * recencyDecay
+	feedbackContrib := float64(obs.UserFeedback) * c.config.FeedbackWeight
+
+	conceptBoost := 0.0
+	for _, concept := range obs.Concepts {
+		if weight, ok := c.config.ConceptWeights[concept]; ok {
+			conceptBoost += weight
+		}
+	}
+	conceptContrib := conceptBoost * c.config.ConceptWeight
+
+	retrievalContrib := 0.0
+	if obs.RetrievalCount > 0 {
+		retrievalBoost := math.Log2(float64(obs.RetrievalCount)+1) * 0.1
+		retrievalContrib = retrievalBoost * c.config.RetrievalWeight
+	}
+
+	finalScore := coreScore + feedbackContrib + conceptContrib + retrievalContrib
+	if finalScore < c.config.MinScore {
+		finalScore = c.config.MinScore
+	}
+
+	return ScoreComponents{
+		TypeWeight:       typeWeight,
+		RecencyDecay:     recencyDecay,
+		CoreScore:        coreScore,
+		FeedbackContrib:  feedbackContrib,
+		ConceptContrib:   conceptContrib,
+		RetrievalContrib: retrievalContrib,
+		FinalScore:       finalScore,
+		AgeDays:          ageDays,
+	}
+}
+
+// ScoreComponents contains the breakdown of an importance score calculation.
+type ScoreComponents struct {
+	TypeWeight       float64 `json:"type_weight"`
+	RecencyDecay     float64 `json:"recency_decay"`
+	CoreScore        float64 `json:"core_score"`
+	FeedbackContrib  float64 `json:"feedback_contrib"`
+	ConceptContrib   float64 `json:"concept_contrib"`
+	RetrievalContrib float64 `json:"retrieval_contrib"`
+	FinalScore       float64 `json:"final_score"`
+	AgeDays          float64 `json:"age_days"`
+}
+
+// BatchCalculate computes scores for multiple observations.
+// Returns a map of observation ID to calculated score.
+func (c *Calculator) BatchCalculate(observations []*models.Observation, now time.Time) map[int64]float64 {
+	scores := make(map[int64]float64, len(observations))
+	for _, obs := range observations {
+		scores[obs.ID] = c.Calculate(obs, now)
+	}
+	return scores
+}
+
+// RecalculateThreshold returns the minimum duration before an observation
+// should have its score recalculated. This prevents excessive recalculation
+// while ensuring scores stay reasonably fresh.
+func (c *Calculator) RecalculateThreshold() time.Duration {
+	// Recalculate at most every 6 hours
+	// This balances freshness with performance
+	return 6 * time.Hour
+}
+
+// UpdateConfig updates the calculator's scoring configuration.
+// This allows runtime tuning of scoring parameters.
+func (c *Calculator) UpdateConfig(config *models.ScoringConfig) {
+	if config != nil {
+		c.config = config
+	}
+}
+
+// GetConfig returns the current scoring configuration.
+func (c *Calculator) GetConfig() *models.ScoringConfig {
+	return c.config
+}

internal/scoring/calculator_test.go

@@ -0,0 +1,638 @@
+// Package scoring provides importance score calculation for observations.
+package scoring
+
+import (
+	"testing"
+	"time"
+
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"github.com/stretchr/testify/suite"
+)
+
+// CalculatorSuite is a test suite for the Calculator.
+type CalculatorSuite struct {
+	suite.Suite
+	calc   *Calculator
+	config *models.ScoringConfig
+	now    time.Time
+}
+
+func (s *CalculatorSuite) SetupTest() {
+	s.config = models.DefaultScoringConfig()
+	s.calc = NewCalculator(s.config)
+	s.now = time.Date(2025, 1, 15, 12, 0, 0, 0, time.UTC)
+}
+
+func TestCalculatorSuite(t *testing.T) {
+	suite.Run(t, new(CalculatorSuite))
+}
+
+// =============================================================================
+// GOOD SCENARIOS - Expected normal operations
+// =============================================================================
+
+func (s *CalculatorSuite) TestCalculate_GoodScenarios_NewObservation() {
+	// A brand new observation should have score close to type weight
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeBugfix,
+		CreatedAtEpoch: s.now.UnixMilli(),
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// Expected: 1.0 × 1.3 (bugfix weight) × 1.0 (no decay) = 1.3
+	s.InDelta(1.3, score, 0.01, "new bugfix should score ~1.3")
+}
+
+func (s *CalculatorSuite) TestCalculate_GoodScenarios_OneWeekOld() {
+	// One week old observation should have half the recency score
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeDiscovery,
+		CreatedAtEpoch: s.now.Add(-7 * 24 * time.Hour).UnixMilli(),
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// Expected: 1.0 × 1.1 (discovery) × 0.5 (7 days half-life) = 0.55
+	s.InDelta(0.55, score, 0.05, "7-day old discovery should score ~0.55")
+}
+
+func (s *CalculatorSuite) TestCalculate_GoodScenarios_TwoWeeksOld() {
+	// Two weeks old should have 1/4 recency score
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeFeature,
+		CreatedAtEpoch: s.now.Add(-14 * 24 * time.Hour).UnixMilli(),
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// Expected: 1.0 × 1.2 (feature) × 0.25 (14 days = 2 half-lives) = 0.30
+	s.InDelta(0.30, score, 0.05, "14-day old feature should score ~0.30")
+}
+
+func (s *CalculatorSuite) TestCalculate_GoodScenarios_PositiveFeedback() {
+	// Positive feedback should boost score
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeChange,
+		CreatedAtEpoch: s.now.UnixMilli(),
+		UserFeedback:   1, // thumbs up
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// Expected: (1.0 × 0.9) + 0.30 (feedback) = 1.20
+	s.InDelta(1.20, score, 0.01, "thumbs up should boost score by 0.30")
+}
+
+func (s *CalculatorSuite) TestCalculate_GoodScenarios_NegativeFeedback() {
+	// Negative feedback should reduce score
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeChange,
+		CreatedAtEpoch: s.now.UnixMilli(),
+		UserFeedback:   -1, // thumbs down
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// Expected: (1.0 × 0.9) - 0.30 (feedback) = 0.60
+	s.InDelta(0.60, score, 0.01, "thumbs down should reduce score by 0.30")
+}
+
+func (s *CalculatorSuite) TestCalculate_GoodScenarios_WithConcepts() {
+	// Observation with valuable concepts should get boost
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeBugfix,
+		CreatedAtEpoch: s.now.UnixMilli(),
+		Concepts:       []string{"security", "gotcha"},
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// Concept boost: (0.30 + 0.25) × 0.20 = 0.11
+	// Expected: 1.3 + 0.11 = 1.41
+	s.InDelta(1.41, score, 0.05, "security+gotcha concepts should boost score")
+}
+
+func (s *CalculatorSuite) TestCalculate_GoodScenarios_WithRetrievals() {
+	// Popular observations should get retrieval boost
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeDiscovery,
+		CreatedAtEpoch: s.now.UnixMilli(),
+		RetrievalCount: 7, // log2(8) = 3
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// Retrieval boost: log2(7+1) × 0.1 × 0.15 = 3 × 0.1 × 0.15 = 0.045
+	// Expected: 1.1 + 0.045 ≈ 1.145
+	s.InDelta(1.145, score, 0.05, "7 retrievals should add small boost")
+}
+
+func (s *CalculatorSuite) TestCalculate_GoodScenarios_CombinedFactors() {
+	// Test with all factors combined
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeBugfix,
+		CreatedAtEpoch: s.now.Add(-7 * 24 * time.Hour).UnixMilli(), // 7 days old
+		UserFeedback:   1,
+		Concepts:       []string{"security"},
+		RetrievalCount: 3,
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// Core: 1.0 × 1.3 × 0.5 = 0.65
+	// Feedback: 0.30
+	// Concept: 0.30 × 0.20 = 0.06
+	// Retrieval: log2(4) × 0.1 × 0.15 = 2 × 0.1 × 0.15 = 0.03
+	// Total ≈ 1.04
+	s.InDelta(1.04, score, 0.1, "combined factors should result in ~1.04")
+}
+
+// =============================================================================
+// WORSE SCENARIOS - Degraded but acceptable operations
+// =============================================================================
+
+func (s *CalculatorSuite) TestCalculate_WorseScenarios_VeryOldObservation() {
+	// Very old observation should have low but non-zero score
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeChange,
+		CreatedAtEpoch: s.now.Add(-90 * 24 * time.Hour).UnixMilli(), // 90 days old
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// 90 days = ~12.86 half-lives → decay ≈ 0.00014
+	// Core: 1.0 × 0.9 × 0.00014 = 0.000126
+	// But minimum score is 0.01
+	s.GreaterOrEqual(score, 0.01, "very old observation should still meet minimum")
+	s.Less(score, 0.1, "very old observation should be low scoring")
+}
+
+func (s *CalculatorSuite) TestCalculate_WorseScenarios_NegativeFeedbackOld() {
+	// Old observation with negative feedback should still have minimum score
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeChange,
+		CreatedAtEpoch: s.now.Add(-60 * 24 * time.Hour).UnixMilli(),
+		UserFeedback:   -1,
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	s.GreaterOrEqual(score, s.config.MinScore, "should never go below minimum score")
+}
+
+func (s *CalculatorSuite) TestCalculate_WorseScenarios_UnknownConcepts() {
+	// Unknown concepts should not affect score negatively
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeDiscovery,
+		CreatedAtEpoch: s.now.UnixMilli(),
+		Concepts:       []string{"unknown-concept", "another-unknown"},
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// Should just be the base score without concept boost
+	s.InDelta(1.1, score, 0.01, "unknown concepts should not affect score")
+}
+
+func (s *CalculatorSuite) TestCalculate_WorseScenarios_MixedConcepts() {
+	// Mix of known and unknown concepts
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeDiscovery,
+		CreatedAtEpoch: s.now.UnixMilli(),
+		Concepts:       []string{"security", "unknown-concept"},
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// Only security should contribute
+	// Expected: 1.1 + (0.30 × 0.20) = 1.16
+	s.InDelta(1.16, score, 0.05, "only known concepts should boost score")
+}
+
+// =============================================================================
+// BAD SCENARIOS - Edge cases and error conditions
+// =============================================================================
+
+func (s *CalculatorSuite) TestCalculate_BadScenarios_FutureTimestamp() {
+	// Observation created in the future (clock skew)
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeBugfix,
+		CreatedAtEpoch: s.now.Add(24 * time.Hour).UnixMilli(), // 1 day in future
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// Should handle gracefully - age should be 0
+	s.InDelta(1.3, score, 0.01, "future timestamp should be treated as now")
+}
+
+func (s *CalculatorSuite) TestCalculate_BadScenarios_ZeroEpoch() {
+	// Missing creation timestamp
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeDiscovery,
+		CreatedAtEpoch: 0, // Missing timestamp
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// This will be treated as very old (1970)
+	s.GreaterOrEqual(score, s.config.MinScore, "should still meet minimum")
+}
+
+func (s *CalculatorSuite) TestCalculate_BadScenarios_EmptyObservation() {
+	// Minimal observation with defaults
+	obs := &models.Observation{
+		ID:             1,
+		Type:           "", // Empty type
+		CreatedAtEpoch: s.now.UnixMilli(),
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// Unknown type should default to 1.0 weight
+	s.InDelta(1.0, score, 0.01, "empty type should use default weight 1.0")
+}
+
+func (s *CalculatorSuite) TestCalculate_BadScenarios_ExtremeRetrievalCount() {
+	// Very high retrieval count
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeDiscovery,
+		CreatedAtEpoch: s.now.UnixMilli(),
+		RetrievalCount: 1000000, // Extreme value
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// log2(1000001) ≈ 19.93, so boost = 19.93 × 0.1 × 0.15 ≈ 0.30
+	// Score should be reasonable, not exploding
+	s.Less(score, 2.0, "extreme retrieval count should not explode score")
+}
+
+func (s *CalculatorSuite) TestCalculate_BadScenarios_NegativeRetrievalCount() {
+	// Negative retrieval count (should not happen but test defensively)
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeDiscovery,
+		CreatedAtEpoch: s.now.UnixMilli(),
+		RetrievalCount: -5,
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// Should not panic and should give base score
+	s.InDelta(1.1, score, 0.01, "negative retrieval should be ignored")
+}
+
+// =============================================================================
+// EDGE CASES - Boundary conditions
+// =============================================================================
+
+func (s *CalculatorSuite) TestCalculate_EdgeCases_ExactlyOneHalfLife() {
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeChange, // 0.9 weight
+		CreatedAtEpoch: s.now.Add(-7 * 24 * time.Hour).UnixMilli(),
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+	s.InDelta(0.45, score, 0.01, "exactly 7 days should give 0.5 decay")
+}
+
+func (s *CalculatorSuite) TestCalculate_EdgeCases_ExactlyTwoHalfLives() {
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeChange,
+		CreatedAtEpoch: s.now.Add(-14 * 24 * time.Hour).UnixMilli(),
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+	s.InDelta(0.225, score, 0.01, "exactly 14 days should give 0.25 decay")
+}
+
+func (s *CalculatorSuite) TestCalculate_EdgeCases_AllTypeWeights() {
+	types := []struct {
+		t      models.ObservationType
+		weight float64
+	}{
+		{models.ObsTypeBugfix, 1.3},
+		{models.ObsTypeFeature, 1.2},
+		{models.ObsTypeDiscovery, 1.1},
+		{models.ObsTypeDecision, 1.1},
+		{models.ObsTypeRefactor, 1.0},
+		{models.ObsTypeChange, 0.9},
+	}
+
+	for _, tt := range types {
+		s.Run(string(tt.t), func() {
+			obs := &models.Observation{
+				ID:             1,
+				Type:           tt.t,
+				CreatedAtEpoch: s.now.UnixMilli(),
+			}
+			score := s.calc.Calculate(obs, s.now)
+			s.InDelta(tt.weight, score, 0.01)
+		})
+	}
+}
+
+func (s *CalculatorSuite) TestCalculate_EdgeCases_MinimumScoreEnforced() {
+	// Create worst case scenario
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeChange,                                    // Lowest weight 0.9
+		CreatedAtEpoch: time.Date(2000, 1, 1, 0, 0, 0, 0, time.UTC).UnixMilli(), // Very old
+		UserFeedback:   -1,                                                      // Negative feedback
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	s.Equal(s.config.MinScore, score, "should be exactly minimum score")
+}
+
+func (s *CalculatorSuite) TestCalculate_EdgeCases_AllConceptsMaxWeight() {
+	// Observation with all high-value concepts
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeBugfix,
+		CreatedAtEpoch: s.now.UnixMilli(),
+		Concepts:       []string{"security", "gotcha", "best-practice", "anti-pattern"},
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// security=0.30, gotcha=0.25, best-practice=0.20, anti-pattern=0.20 = 0.95
+	// Concept contrib: 0.95 × 0.20 = 0.19
+	// Total: 1.3 + 0.19 = 1.49
+	s.InDelta(1.49, score, 0.05, "all high-value concepts should boost significantly")
+}
+
+// =============================================================================
+// CALCULATE COMPONENTS TESTS
+// =============================================================================
+
+func (s *CalculatorSuite) TestCalculateComponents_ReturnsAllComponents() {
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeBugfix,
+		CreatedAtEpoch: s.now.Add(-7 * 24 * time.Hour).UnixMilli(),
+		UserFeedback:   1,
+		Concepts:       []string{"security"},
+		RetrievalCount: 7,
+	}
+
+	components := s.calc.CalculateComponents(obs, s.now)
+
+	s.InDelta(1.3, components.TypeWeight, 0.01)
+	s.InDelta(0.5, components.RecencyDecay, 0.01)
+	s.InDelta(0.65, components.CoreScore, 0.05)
+	s.InDelta(0.30, components.FeedbackContrib, 0.01)
+	s.InDelta(0.06, components.ConceptContrib, 0.02)
+	s.Greater(components.RetrievalContrib, 0.0)
+	s.InDelta(7.0, components.AgeDays, 0.1)
+	s.Greater(components.FinalScore, 0.0)
+}
+
+func (s *CalculatorSuite) TestCalculateComponents_MatchesCalculate() {
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeFeature,
+		CreatedAtEpoch: s.now.Add(-3 * 24 * time.Hour).UnixMilli(),
+		UserFeedback:   -1,
+		Concepts:       []string{"performance", "architecture"},
+		RetrievalCount: 15,
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+	components := s.calc.CalculateComponents(obs, s.now)
+
+	s.InDelta(score, components.FinalScore, 0.001, "Calculate and CalculateComponents should match")
+}
+
+// =============================================================================
+// BATCH CALCULATE TESTS
+// =============================================================================
+
+func (s *CalculatorSuite) TestBatchCalculate_Empty() {
+	scores := s.calc.BatchCalculate(nil, s.now)
+	s.Empty(scores)
+
+	scores = s.calc.BatchCalculate([]*models.Observation{}, s.now)
+	s.Empty(scores)
+}
+
+func (s *CalculatorSuite) TestBatchCalculate_Multiple() {
+	obs := []*models.Observation{
+		{ID: 1, Type: models.ObsTypeBugfix, CreatedAtEpoch: s.now.UnixMilli()},
+		{ID: 2, Type: models.ObsTypeFeature, CreatedAtEpoch: s.now.Add(-7 * 24 * time.Hour).UnixMilli()},
+		{ID: 3, Type: models.ObsTypeChange, CreatedAtEpoch: s.now.Add(-14 * 24 * time.Hour).UnixMilli()},
+	}
+
+	scores := s.calc.BatchCalculate(obs, s.now)
+
+	s.Len(scores, 3)
+	s.Contains(scores, int64(1))
+	s.Contains(scores, int64(2))
+	s.Contains(scores, int64(3))
+
+	s.InDelta(1.3, scores[1], 0.01)   // New bugfix
+	s.InDelta(0.6, scores[2], 0.1)    // 7-day feature
+	s.InDelta(0.225, scores[3], 0.05) // 14-day change
+}
+
+// =============================================================================
+// CONFIGURATION TESTS
+// =============================================================================
+
+func (s *CalculatorSuite) TestNewCalculator_NilConfig() {
+	calc := NewCalculator(nil)
+	s.NotNil(calc)
+	s.NotNil(calc.config)
+	s.Equal(7.0, calc.config.RecencyHalfLifeDays)
+}
+
+func (s *CalculatorSuite) TestUpdateConfig() {
+	newConfig := &models.ScoringConfig{
+		RecencyHalfLifeDays: 14.0, // Changed from 7
+		FeedbackWeight:      0.50,
+		ConceptWeight:       0.10,
+		RetrievalWeight:     0.05,
+		MinScore:            0.001,
+		ConceptWeights:      map[string]float64{"test": 0.5},
+	}
+
+	s.calc.UpdateConfig(newConfig)
+
+	obs := &models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeChange,
+		CreatedAtEpoch: s.now.Add(-14 * 24 * time.Hour).UnixMilli(),
+	}
+
+	score := s.calc.Calculate(obs, s.now)
+
+	// With 14-day half-life, 14 days = exactly one half-life
+	// Expected: 1.0 × 0.9 × 0.5 = 0.45
+	s.InDelta(0.45, score, 0.01)
+}
+
+func (s *CalculatorSuite) TestUpdateConfig_NilIgnored() {
+	originalConfig := s.calc.GetConfig()
+	s.calc.UpdateConfig(nil)
+	s.Equal(originalConfig, s.calc.GetConfig())
+}
+
+func (s *CalculatorSuite) TestGetConfig() {
+	config := s.calc.GetConfig()
+	s.NotNil(config)
+	s.Equal(7.0, config.RecencyHalfLifeDays)
+}
+
+func (s *CalculatorSuite) TestRecalculateThreshold() {
+	threshold := s.calc.RecalculateThreshold()
+	s.Equal(6*time.Hour, threshold)
+}
+
+// =============================================================================
+// STANDALONE TESTS (non-suite)
+// =============================================================================
+
+func TestNewCalculator_DefaultConfig(t *testing.T) {
+	calc := NewCalculator(nil)
+	require.NotNil(t, calc)
+	assert.Equal(t, 7.0, calc.config.RecencyHalfLifeDays)
+	assert.Equal(t, 0.30, calc.config.FeedbackWeight)
+	assert.Equal(t, 0.01, calc.config.MinScore)
+}
+
+func TestCalculator_ConcurrentAccess(t *testing.T) {
+	calc := NewCalculator(nil)
+	now := time.Now()
+
+	// Test that calculator is safe for concurrent reads
+	done := make(chan bool, 10)
+	for i := 0; i < 10; i++ {
+		go func(id int64) {
+			obs := &models.Observation{
+				ID:             id,
+				Type:           models.ObsTypeBugfix,
+				CreatedAtEpoch: now.UnixMilli(),
+			}
+			score := calc.Calculate(obs, now)
+			assert.Greater(t, score, 0.0)
+			done <- true
+		}(int64(i))
+	}
+
+	for i := 0; i < 10; i++ {
+		<-done
+	}
+}
+
+func TestCalculator_DecayPrecision(t *testing.T) {
+	calc := NewCalculator(nil)
+	now := time.Now()
+
+	// Test that decay is mathematically correct
+	testCases := []struct {
+		days          int
+		expectedDecay float64
+	}{
+		{0, 1.0},
+		{7, 0.5},
+		{14, 0.25},
+		{21, 0.125},
+		{28, 0.0625},
+	}
+
+	for _, tc := range testCases {
+		t.Run(string(rune('0'+tc.days/7))+"_half_lives", func(t *testing.T) {
+			obs := &models.Observation{
+				ID:             1,
+				Type:           models.ObsTypeRefactor, // 1.0 weight
+				CreatedAtEpoch: now.Add(-time.Duration(tc.days) * 24 * time.Hour).UnixMilli(),
+			}
+			components := calc.CalculateComponents(obs, now)
+			assert.InDelta(t, tc.expectedDecay, components.RecencyDecay, 0.001)
+		})
+	}
+}
+
+func TestTypeBaseScore_UnknownType(t *testing.T) {
+	score := models.TypeBaseScore("unknown-type")
+	assert.Equal(t, 1.0, score, "unknown type should default to 1.0")
+}
+
+func TestTypeBaseScore_AllKnownTypes(t *testing.T) {
+	expected := map[models.ObservationType]float64{
+		models.ObsTypeBugfix:    1.3,
+		models.ObsTypeFeature:   1.2,
+		models.ObsTypeDiscovery: 1.1,
+		models.ObsTypeDecision:  1.1,
+		models.ObsTypeRefactor:  1.0,
+		models.ObsTypeChange:    0.9,
+	}
+
+	for obsType, expectedScore := range expected {
+		t.Run(string(obsType), func(t *testing.T) {
+			score := models.TypeBaseScore(obsType)
+			assert.Equal(t, expectedScore, score)
+		})
+	}
+}
+
+func TestCalculator_RetrievalBoostDiminishingReturns(t *testing.T) {
+	calc := NewCalculator(nil)
+	now := time.Now()
+
+	// Test that retrieval boost has diminishing returns
+	// When retrieval count doubles, the boost should NOT double (log2 gives diminishing returns)
+
+	// Collect boosts for different counts
+	boosts := make([]float64, 0)
+	retrievalCounts := []int{1, 3, 7, 15, 31, 63, 127}
+
+	for _, count := range retrievalCounts {
+		obs := &models.Observation{
+			ID:             1,
+			Type:           models.ObsTypeRefactor,
+			CreatedAtEpoch: now.UnixMilli(),
+			RetrievalCount: count,
+		}
+		components := calc.CalculateComponents(obs, now)
+		boosts = append(boosts, components.RetrievalContrib)
+	}
+
+	// Verify boost increases but at a decreasing rate
+	for i := 1; i < len(boosts); i++ {
+		// Each boost should be higher than the previous
+		assert.Greater(t, boosts[i], boosts[i-1],
+			"boost should increase with more retrievals")
+
+		// But not proportionally - calculate the ratios
+		if i >= 2 {
+			ratio1 := boosts[i-1] / boosts[i-2]
+			ratio2 := boosts[i] / boosts[i-1]
+			// The growth ratio should be decreasing (diminishing returns)
+			assert.Less(t, ratio2, ratio1+0.01, // Allow small floating point tolerance
+				"growth rate should decrease (diminishing returns)")
+		}
+	}
+}

internal/scoring/recalculator.go

@@ -0,0 +1,186 @@
+// Package scoring provides importance score calculation for observations.
+package scoring
+
+import (
+	"context"
+	"sync"
+	"time"
+
+	"github.com/rs/zerolog"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/db/sqlite"
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+)
+
+// ObservationStore defines the interface for observation storage operations needed by the recalculator.
+type ObservationStore interface {
+	GetObservationsNeedingScoreUpdate(ctx context.Context, threshold time.Duration, limit int) ([]*models.Observation, error)
+	UpdateImportanceScores(ctx context.Context, scores map[int64]float64) error
+	GetConceptWeights(ctx context.Context) (map[string]float64, error)
+}
+
+// Recalculator periodically recalculates importance scores for observations.
+type Recalculator struct {
+	log        zerolog.Logger
+	store      ObservationStore
+	calculator *Calculator
+	stopCh     chan struct{}
+	doneCh     chan struct{}
+	interval   time.Duration
+	batchSize  int
+	mu         sync.Mutex
+	running    bool
+}
+
+// NewRecalculator creates a new background recalculator.
+func NewRecalculator(store ObservationStore, calc *Calculator, log zerolog.Logger) *Recalculator {
+	return &Recalculator{
+		store:      store,
+		calculator: calc,
+		log:        log.With().Str("component", "recalculator").Logger(),
+		interval:   1 * time.Hour, // Run every hour
+		batchSize:  500,           // Process 500 observations at a time
+		stopCh:     make(chan struct{}),
+		doneCh:     make(chan struct{}),
+	}
+}
+
+// Start begins the background recalculation loop.
+// This should be called in a goroutine.
+func (r *Recalculator) Start(ctx context.Context) {
+	r.mu.Lock()
+	if r.running {
+		r.mu.Unlock()
+		return
+	}
+	r.running = true
+	r.mu.Unlock()
+
+	defer func() {
+		r.mu.Lock()
+		r.running = false
+		r.mu.Unlock()
+		close(r.doneCh)
+	}()
+
+	// Initial run
+	r.recalculate(ctx)
+
+	r.mu.Lock()
+	interval := r.interval
+	r.mu.Unlock()
+
+	ticker := time.NewTicker(interval)
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-ctx.Done():
+			r.log.Info().Msg("recalculator shutting down due to context cancellation")
+			return
+		case <-r.stopCh:
+			r.log.Info().Msg("recalculator stopping")
+			return
+		case <-ticker.C:
+			r.recalculate(ctx)
+		}
+	}
+}
+
+// Stop stops the background recalculation loop.
+func (r *Recalculator) Stop() {
+	r.mu.Lock()
+	if !r.running {
+		r.mu.Unlock()
+		return
+	}
+	r.mu.Unlock()
+
+	close(r.stopCh)
+	<-r.doneCh
+}
+
+// recalculate performs a single recalculation batch.
+func (r *Recalculator) recalculate(ctx context.Context) {
+	now := time.Now()
+	threshold := r.calculator.RecalculateThreshold()
+
+	r.mu.Lock()
+	batchSize := r.batchSize
+	r.mu.Unlock()
+
+	observations, err := r.store.GetObservationsNeedingScoreUpdate(ctx, threshold, batchSize)
+	if err != nil {
+		r.log.Error().Err(err).Msg("failed to get observations for score update")
+		return
+	}
+
+	if len(observations) == 0 {
+		return
+	}
+
+	scores := r.calculator.BatchCalculate(observations, now)
+
+	if err := r.store.UpdateImportanceScores(ctx, scores); err != nil {
+		r.log.Error().Err(err).Msg("failed to update importance scores")
+		return
+	}
+
+	r.log.Info().
+		Int("count", len(scores)).
+		Dur("elapsed", time.Since(now)).
+		Msg("recalculated importance scores")
+}
+
+// RecalculateNow triggers an immediate recalculation.
+// This is useful for testing or when scores need to be updated urgently.
+func (r *Recalculator) RecalculateNow(ctx context.Context) error {
+	r.recalculate(ctx)
+	return nil
+}
+
+// RefreshConceptWeights reloads concept weights from the database.
+// Call this after updating concept weights to apply changes.
+func (r *Recalculator) RefreshConceptWeights(ctx context.Context) error {
+	weights, err := r.store.GetConceptWeights(ctx)
+	if err != nil {
+		return err
+	}
+
+	config := r.calculator.GetConfig()
+	config.ConceptWeights = weights
+	r.calculator.UpdateConfig(config)
+
+	r.log.Info().Int("count", len(weights)).Msg("refreshed concept weights")
+	return nil
+}
+
+// Stats returns statistics about the recalculator.
+type Stats struct {
+	Running     bool          `json:"running"`
+	Interval    time.Duration `json:"interval"`
+	BatchSize   int           `json:"batch_size"`
+	HalfLife    float64       `json:"half_life_days"`
+	MinScore    float64       `json:"min_score"`
+	ConceptsLen int           `json:"concepts_count"`
+}
+
+// GetStats returns current recalculator statistics.
+func (r *Recalculator) GetStats() Stats {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+
+	config := r.calculator.GetConfig()
+
+	return Stats{
+		Running:     r.running,
+		Interval:    r.interval,
+		BatchSize:   r.batchSize,
+		HalfLife:    config.RecencyHalfLifeDays,
+		MinScore:    config.MinScore,
+		ConceptsLen: len(config.ConceptWeights),
+	}
+}
+
+// Ensure ObservationStore satisfies the interface
+var _ ObservationStore = (*sqlite.ObservationStore)(nil)

internal/scoring/recalculator_test.go

@@ -0,0 +1,447 @@
+// Package scoring provides importance score calculation for observations.
+package scoring
+
+import (
+	"context"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/rs/zerolog"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+)
+
+// MockObservationStore is a mock implementation of ObservationStore for testing.
+type MockObservationStore struct {
+	updateErr         error
+	getErr            error
+	getConceptsErr    error
+	scores            map[int64]float64
+	conceptWeights    map[string]float64
+	observations      []*models.Observation
+	updateScoresCalls int
+	mu                sync.Mutex
+}
+
+func NewMockObservationStore() *MockObservationStore {
+	return &MockObservationStore{
+		observations:   []*models.Observation{},
+		scores:         make(map[int64]float64),
+		conceptWeights: make(map[string]float64),
+	}
+}
+
+func (m *MockObservationStore) GetObservationsNeedingScoreUpdate(ctx context.Context, threshold time.Duration, limit int) ([]*models.Observation, error) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+
+	if m.getErr != nil {
+		return nil, m.getErr
+	}
+
+	// Return observations that haven't been updated within threshold
+	now := time.Now()
+	var result []*models.Observation
+	for _, obs := range m.observations {
+		if !obs.ScoreUpdatedAt.Valid || now.Sub(time.Unix(obs.ScoreUpdatedAt.Int64, 0)) > threshold {
+			result = append(result, obs)
+			if len(result) >= limit {
+				break
+			}
+		}
+	}
+	return result, nil
+}
+
+func (m *MockObservationStore) UpdateImportanceScores(ctx context.Context, scores map[int64]float64) error {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+
+	m.updateScoresCalls++
+	if m.updateErr != nil {
+		return m.updateErr
+	}
+
+	for id, score := range scores {
+		m.scores[id] = score
+	}
+	return nil
+}
+
+func (m *MockObservationStore) GetConceptWeights(ctx context.Context) (map[string]float64, error) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+
+	if m.getConceptsErr != nil {
+		return nil, m.getConceptsErr
+	}
+	return m.conceptWeights, nil
+}
+
+func (m *MockObservationStore) AddObservation(obs *models.Observation) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	m.observations = append(m.observations, obs)
+}
+
+func (m *MockObservationStore) SetConceptWeights(weights map[string]float64) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	m.conceptWeights = weights
+}
+
+func (m *MockObservationStore) GetScore(id int64) (float64, bool) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	score, ok := m.scores[id]
+	return score, ok
+}
+
+func (m *MockObservationStore) GetUpdateScoresCalls() int {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	return m.updateScoresCalls
+}
+
+func (m *MockObservationStore) SetUpdateError(err error) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	m.updateErr = err
+}
+
+func (m *MockObservationStore) SetGetError(err error) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	m.getErr = err
+}
+
+// =============================================================================
+// RECALCULATOR TESTS
+// =============================================================================
+
+func TestNewRecalculator(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+
+	recalc := NewRecalculator(store, calc, log)
+
+	require.NotNil(t, recalc)
+	assert.NotNil(t, recalc.store)
+	assert.NotNil(t, recalc.calculator)
+	assert.Equal(t, 1*time.Hour, recalc.interval)
+	assert.Equal(t, 500, recalc.batchSize)
+	assert.False(t, recalc.running)
+}
+
+func TestRecalculator_RecalculateNow(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+	recalc := NewRecalculator(store, calc, log)
+
+	// Add observations
+	now := time.Now()
+	store.AddObservation(&models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeBugfix,
+		CreatedAtEpoch: now.UnixMilli(),
+	})
+	store.AddObservation(&models.Observation{
+		ID:             2,
+		Type:           models.ObsTypeFeature,
+		CreatedAtEpoch: now.Add(-7 * 24 * time.Hour).UnixMilli(),
+	})
+
+	ctx := context.Background()
+	err := recalc.RecalculateNow(ctx)
+
+	require.NoError(t, err)
+
+	// Verify scores were calculated
+	score1, ok := store.GetScore(1)
+	assert.True(t, ok)
+	assert.Greater(t, score1, 0.0)
+
+	score2, ok := store.GetScore(2)
+	assert.True(t, ok)
+	assert.Greater(t, score2, 0.0)
+	assert.Less(t, score2, score1, "older observation should have lower score")
+}
+
+func TestRecalculator_RefreshConceptWeights(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+	recalc := NewRecalculator(store, calc, log)
+
+	// Set up custom weights in store
+	customWeights := map[string]float64{
+		"security":    0.50,
+		"performance": 0.25,
+	}
+	store.SetConceptWeights(customWeights)
+
+	ctx := context.Background()
+	err := recalc.RefreshConceptWeights(ctx)
+
+	require.NoError(t, err)
+
+	// Verify weights were updated in calculator config
+	config := calc.GetConfig()
+	assert.Equal(t, 0.50, config.ConceptWeights["security"])
+	assert.Equal(t, 0.25, config.ConceptWeights["performance"])
+}
+
+func TestRecalculator_RefreshConceptWeights_Error(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+	recalc := NewRecalculator(store, calc, log)
+
+	store.getConceptsErr = assert.AnError
+
+	ctx := context.Background()
+	err := recalc.RefreshConceptWeights(ctx)
+
+	require.Error(t, err)
+}
+
+func TestRecalculator_GetStats(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+	recalc := NewRecalculator(store, calc, log)
+
+	// Set fields directly for testing
+	recalc.interval = 2 * time.Hour
+	recalc.batchSize = 250
+
+	stats := recalc.GetStats()
+
+	assert.False(t, stats.Running)
+	assert.Equal(t, 2*time.Hour, stats.Interval)
+	assert.Equal(t, 250, stats.BatchSize)
+	assert.Equal(t, 7.0, stats.HalfLife)
+	assert.Equal(t, 0.01, stats.MinScore)
+}
+
+func TestRecalculator_StartStop(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+	recalc := NewRecalculator(store, calc, log)
+
+	// Use a short interval for testing
+	recalc.interval = 50 * time.Millisecond
+
+	// Add an observation
+	store.AddObservation(&models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeBugfix,
+		CreatedAtEpoch: time.Now().UnixMilli(),
+	})
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	// Start in goroutine
+	go recalc.Start(ctx)
+
+	// Wait for initial run and at least one tick
+	time.Sleep(100 * time.Millisecond)
+
+	// Verify it ran
+	calls := store.GetUpdateScoresCalls()
+	assert.GreaterOrEqual(t, calls, 1, "should have run at least once")
+
+	// Stop via context cancellation
+	cancel()
+	time.Sleep(100 * time.Millisecond)
+
+	// Verify stopped
+	stats := recalc.GetStats()
+	assert.False(t, stats.Running)
+}
+
+func TestRecalculator_StartTwice(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+	recalc := NewRecalculator(store, calc, log)
+
+	recalc.interval = 1 * time.Hour // Long interval so it doesn't tick during test
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	// Start first time
+	go recalc.Start(ctx)
+	time.Sleep(50 * time.Millisecond)
+
+	// Try to start second time (should return immediately)
+	go recalc.Start(ctx)
+	time.Sleep(50 * time.Millisecond)
+
+	// Should still be running (only once)
+	stats := recalc.GetStats()
+	assert.True(t, stats.Running)
+
+	cancel()
+}
+
+func TestRecalculator_StopWhenNotRunning(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+	recalc := NewRecalculator(store, calc, log)
+
+	// Stop without starting - should not panic
+	recalc.Stop()
+
+	stats := recalc.GetStats()
+	assert.False(t, stats.Running)
+}
+
+func TestRecalculator_EmptyStore(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+	recalc := NewRecalculator(store, calc, log)
+
+	ctx := context.Background()
+	err := recalc.RecalculateNow(ctx)
+
+	require.NoError(t, err)
+	assert.Equal(t, 0, store.GetUpdateScoresCalls(), "should not call update with no observations")
+}
+
+func TestRecalculator_GetObservationsError(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+	recalc := NewRecalculator(store, calc, log)
+
+	store.SetGetError(assert.AnError)
+
+	ctx := context.Background()
+	err := recalc.RecalculateNow(ctx)
+
+	// Should not return error (logs it instead)
+	require.NoError(t, err)
+	assert.Equal(t, 0, store.GetUpdateScoresCalls())
+}
+
+func TestRecalculator_UpdateScoresError(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+	recalc := NewRecalculator(store, calc, log)
+
+	store.AddObservation(&models.Observation{
+		ID:             1,
+		Type:           models.ObsTypeBugfix,
+		CreatedAtEpoch: time.Now().UnixMilli(),
+	})
+
+	store.SetUpdateError(assert.AnError)
+
+	ctx := context.Background()
+	err := recalc.RecalculateNow(ctx)
+
+	// Should not return error (logs it instead)
+	require.NoError(t, err)
+	assert.Equal(t, 1, store.GetUpdateScoresCalls())
+}
+
+func TestRecalculator_BatchProcessing(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+	recalc := NewRecalculator(store, calc, log)
+
+	// Set small batch size
+	recalc.batchSize = 3
+
+	// Add 5 observations
+	now := time.Now()
+	for i := 1; i <= 5; i++ {
+		store.AddObservation(&models.Observation{
+			ID:             int64(i),
+			Type:           models.ObsTypeBugfix,
+			CreatedAtEpoch: now.UnixMilli(),
+		})
+	}
+
+	ctx := context.Background()
+	err := recalc.RecalculateNow(ctx)
+
+	require.NoError(t, err)
+
+	// Should only process batch size (3)
+	scores := 0
+	for i := 1; i <= 5; i++ {
+		if _, ok := store.GetScore(int64(i)); ok {
+			scores++
+		}
+	}
+	assert.Equal(t, 3, scores, "should only process batch size observations")
+}
+
+func TestRecalculator_ConcurrentAccess(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+	recalc := NewRecalculator(store, calc, log)
+
+	// Add observations
+	now := time.Now()
+	for i := 1; i <= 10; i++ {
+		store.AddObservation(&models.Observation{
+			ID:             int64(i),
+			Type:           models.ObsTypeBugfix,
+			CreatedAtEpoch: now.UnixMilli(),
+		})
+	}
+
+	ctx := context.Background()
+	var wg sync.WaitGroup
+
+	// Run multiple recalculations concurrently
+	for i := 0; i < 5; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			_ = recalc.RecalculateNow(ctx)
+		}()
+	}
+
+	wg.Wait()
+
+	// Should complete without race conditions
+	// (use -race flag to verify)
+	assert.GreaterOrEqual(t, store.GetUpdateScoresCalls(), 1)
+}
+
+func TestRecalculator_StatsThreadSafe(t *testing.T) {
+	store := NewMockObservationStore()
+	calc := NewCalculator(nil)
+	log := zerolog.Nop()
+	recalc := NewRecalculator(store, calc, log)
+
+	var wg sync.WaitGroup
+
+	// Concurrent reads (use -race flag to verify)
+	for i := 0; i < 10; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			_ = recalc.GetStats()
+		}()
+	}
+
+	wg.Wait()
+}

internal/search/expansion/expander.go

@@ -0,0 +1,448 @@
+// Package expansion provides context-aware query expansion for improved search recall.
+package expansion
+
+import (
+	"context"
+	"regexp"
+	"strings"
+	"sync"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/embedding"
+	"github.com/rs/zerolog/log"
+)
+
+// QueryIntent represents the detected intent of a query.
+type QueryIntent string
+
+const (
+	// IntentQuestion indicates a question-type query (how, why, what, etc.)
+	IntentQuestion QueryIntent = "question"
+	// IntentError indicates an error/debugging query
+	IntentError QueryIntent = "error"
+	// IntentImplementation indicates an implementation/coding query
+	IntentImplementation QueryIntent = "implementation"
+	// IntentArchitecture indicates an architecture/design query
+	IntentArchitecture QueryIntent = "architecture"
+	// IntentGeneral indicates a general lookup query
+	IntentGeneral QueryIntent = "general"
+)
+
+// ExpandedQuery represents a query variant with metadata.
+type ExpandedQuery struct {
+	Query  string      `json:"query"`
+	Source string      `json:"source"`
+	Intent QueryIntent `json:"intent"`
+	Weight float64     `json:"weight"`
+}
+
+// Expander provides context-aware query expansion.
+type Expander struct {
+	embedSvc       *embedding.Service
+	intentPatterns map[QueryIntent][]*regexp.Regexp
+	vocabulary     []VocabEntry
+	vocabVectors   [][]float32
+	vocabMu        sync.RWMutex
+}
+
+// VocabEntry represents a vocabulary term from observations.
+type VocabEntry struct {
+	Term   string
+	Source string
+	Weight float64
+}
+
+// Config holds expander configuration.
+type Config struct {
+	// MaxExpansions limits the number of expanded queries returned
+	MaxExpansions int
+	// MinSimilarity is the minimum similarity score for vocabulary expansion
+	MinSimilarity float64
+	// EnableVocabularyExpansion enables finding related terms from observations
+	EnableVocabularyExpansion bool
+}
+
+// DefaultConfig returns sensible default configuration.
+func DefaultConfig() Config {
+	return Config{
+		MaxExpansions:             4,
+		MinSimilarity:             0.5,
+		EnableVocabularyExpansion: true,
+	}
+}
+
+// NewExpander creates a new query expander.
+func NewExpander(embedSvc *embedding.Service) *Expander {
+	e := &Expander{
+		embedSvc:       embedSvc,
+		intentPatterns: buildIntentPatterns(),
+	}
+	return e
+}
+
+// buildIntentPatterns creates regex patterns for intent detection.
+func buildIntentPatterns() map[QueryIntent][]*regexp.Regexp {
+	patterns := make(map[QueryIntent][]*regexp.Regexp)
+
+	// Question patterns
+	patterns[IntentQuestion] = []*regexp.Regexp{
+		regexp.MustCompile(`(?i)^(how|why|what|when|where|which|who)\b`),
+		regexp.MustCompile(`(?i)\?$`),
+		regexp.MustCompile(`(?i)\b(explain|describe|understand)\b`),
+	}
+
+	// Error/debugging patterns
+	patterns[IntentError] = []*regexp.Regexp{
+		regexp.MustCompile(`(?i)\b(error|bug|issue|problem|fail|crash|exception|panic)\b`),
+		regexp.MustCompile(`(?i)\b(fix|debug|troubleshoot|resolve)\b`),
+		regexp.MustCompile(`(?i)\b(doesn't work|not working|broken)\b`),
+	}
+
+	// Implementation patterns
+	patterns[IntentImplementation] = []*regexp.Regexp{
+		regexp.MustCompile(`(?i)\b(implement|add|create|build|write|code)\b`),
+		regexp.MustCompile(`(?i)\b(function|method|handler|endpoint|api)\b`),
+		regexp.MustCompile(`(?i)\b(feature|functionality)\b`),
+	}
+
+	// Architecture patterns
+	patterns[IntentArchitecture] = []*regexp.Regexp{
+		regexp.MustCompile(`(?i)\b(architecture|design|pattern|structure)\b`),
+		regexp.MustCompile(`(?i)\b(component|module|layer|service)\b`),
+		regexp.MustCompile(`(?i)\b(flow|pipeline|workflow)\b`),
+	}
+
+	return patterns
+}
+
+// DetectIntent analyzes a query to determine its intent.
+func (e *Expander) DetectIntent(query string) QueryIntent {
+	query = strings.TrimSpace(query)
+	if query == "" {
+		return IntentGeneral
+	}
+
+	// Check patterns in priority order
+	intentOrder := []QueryIntent{IntentError, IntentQuestion, IntentImplementation, IntentArchitecture}
+
+	for _, intent := range intentOrder {
+		patterns := e.intentPatterns[intent]
+		for _, pattern := range patterns {
+			if pattern.MatchString(query) {
+				return intent
+			}
+		}
+	}
+
+	return IntentGeneral
+}
+
+// Expand generates expanded query variants based on the original query.
+func (e *Expander) Expand(ctx context.Context, query string, cfg Config) []ExpandedQuery {
+	query = strings.TrimSpace(query)
+	if query == "" {
+		return nil
+	}
+
+	intent := e.DetectIntent(query)
+	expansions := make([]ExpandedQuery, 0, cfg.MaxExpansions)
+
+	// Always include the original query with highest weight
+	expansions = append(expansions, ExpandedQuery{
+		Query:  query,
+		Weight: 1.0,
+		Source: "original",
+		Intent: intent,
+	})
+
+	// Generate intent-based expansions
+	intentExpansions := e.expandByIntent(query, intent)
+	expansions = append(expansions, intentExpansions...)
+
+	// Generate vocabulary-based expansions if enabled and we have vocabulary
+	if cfg.EnableVocabularyExpansion && e.embedSvc != nil && len(e.vocabulary) > 0 {
+		vocabExpansions := e.expandByVocabulary(ctx, query, cfg.MinSimilarity)
+		expansions = append(expansions, vocabExpansions...)
+	}
+
+	// Deduplicate and limit
+	expansions = deduplicateExpansions(expansions)
+	if len(expansions) > cfg.MaxExpansions {
+		expansions = expansions[:cfg.MaxExpansions]
+	}
+
+	log.Debug().
+		Str("query", truncate(query, 50)).
+		Str("intent", string(intent)).
+		Int("expansions", len(expansions)).
+		Msg("Query expanded")
+
+	return expansions
+}
+
+// expandByIntent generates expansions based on detected query intent.
+func (e *Expander) expandByIntent(query string, intent QueryIntent) []ExpandedQuery {
+	var expansions []ExpandedQuery
+
+	// Extract key terms from query for context-aware expansion
+	keyTerms := extractKeyTerms(query)
+
+	switch intent {
+	case IntentQuestion:
+		// For questions, create a declarative variant
+		declarative := makeDeclarative(query)
+		if declarative != query {
+			expansions = append(expansions, ExpandedQuery{
+				Query:  declarative,
+				Weight: 0.85,
+				Source: "intent:declarative",
+				Intent: intent,
+			})
+		}
+
+	case IntentError:
+		// For errors, expand with solution-oriented terms
+		if len(keyTerms) > 0 {
+			solutionQuery := strings.Join(keyTerms, " ") + " solution fix"
+			expansions = append(expansions, ExpandedQuery{
+				Query:  solutionQuery,
+				Weight: 0.8,
+				Source: "intent:solution",
+				Intent: intent,
+			})
+		}
+
+	case IntentImplementation:
+		// For implementation queries, focus on the what/how
+		if len(keyTerms) > 0 {
+			howQuery := "how " + strings.Join(keyTerms, " ")
+			expansions = append(expansions, ExpandedQuery{
+				Query:  howQuery,
+				Weight: 0.75,
+				Source: "intent:how",
+				Intent: intent,
+			})
+		}
+
+	case IntentArchitecture:
+		// For architecture queries, expand with design context
+		if len(keyTerms) > 0 {
+			designQuery := strings.Join(keyTerms, " ") + " design structure"
+			expansions = append(expansions, ExpandedQuery{
+				Query:  designQuery,
+				Weight: 0.75,
+				Source: "intent:design",
+				Intent: intent,
+			})
+		}
+
+	case IntentGeneral:
+		// For general queries, try noun phrase extraction
+		// No additional expansion - rely on vocabulary expansion
+	}
+
+	return expansions
+}
+
+// expandByVocabulary finds similar terms from the observation vocabulary.
+func (e *Expander) expandByVocabulary(ctx context.Context, query string, minSimilarity float64) []ExpandedQuery {
+	e.vocabMu.RLock()
+	defer e.vocabMu.RUnlock()
+
+	if len(e.vocabulary) == 0 || e.embedSvc == nil {
+		return nil
+	}
+
+	// Embed the query
+	queryEmb, err := e.embedSvc.Embed(query)
+	if err != nil {
+		log.Warn().Err(err).Msg("Failed to embed query for vocabulary expansion")
+		return nil
+	}
+
+	// Find similar vocabulary terms
+	type scoredTerm struct {
+		entry VocabEntry
+		score float64
+	}
+
+	var similar []scoredTerm
+	for i, entry := range e.vocabulary {
+		if i >= len(e.vocabVectors) {
+			break
+		}
+
+		score := cosineSimilarity(queryEmb, e.vocabVectors[i])
+		if score >= minSimilarity {
+			similar = append(similar, scoredTerm{entry: entry, score: score})
+		}
+	}
+
+	if len(similar) == 0 {
+		return nil
+	}
+
+	// Sort by score (descending) using bubble sort
+	for i := 0; i < len(similar)-1; i++ {
+		for j := i + 1; j < len(similar); j++ {
+			if similar[j].score > similar[i].score {
+				similar[i], similar[j] = similar[j], similar[i]
+			}
+		}
+	}
+
+	// Create expansion by combining top similar terms with query
+	var expansions []ExpandedQuery
+	if len(similar) > 0 {
+		// Take top 2 similar terms and combine with original key terms
+		keyTerms := extractKeyTerms(query)
+		for i := 0; i < min(2, len(similar)); i++ {
+			term := similar[i].entry.Term
+			// Don't add if term is already in query
+			if strings.Contains(strings.ToLower(query), strings.ToLower(term)) {
+				continue
+			}
+
+			combinedQuery := strings.Join(keyTerms, " ") + " " + term
+			expansions = append(expansions, ExpandedQuery{
+				Query:  combinedQuery,
+				Weight: 0.7 * similar[i].score * similar[i].entry.Weight,
+				Source: "vocabulary:" + term,
+				Intent: IntentGeneral,
+			})
+		}
+	}
+
+	return expansions
+}
+
+// Helper functions
+
+// extractKeyTerms extracts meaningful terms from a query.
+func extractKeyTerms(query string) []string {
+	// Common stop words to filter out
+	stopWords := map[string]bool{
+		"a": true, "an": true, "the": true, "is": true, "are": true,
+		"was": true, "were": true, "be": true, "been": true, "being": true,
+		"have": true, "has": true, "had": true, "do": true, "does": true,
+		"did": true, "will": true, "would": true, "could": true, "should": true,
+		"may": true, "might": true, "must": true, "can": true,
+		"i": true, "me": true, "my": true, "we": true, "our": true,
+		"you": true, "your": true, "it": true, "its": true,
+		"this": true, "that": true, "these": true, "those": true,
+		"what": true, "which": true, "who": true, "whom": true,
+		"how": true, "why": true, "when": true, "where": true,
+		"to": true, "for": true, "with": true, "about": true, "from": true,
+		"in": true, "on": true, "at": true, "by": true, "of": true,
+		"and": true, "or": true, "but": true, "if": true, "then": true,
+	}
+
+	// Split and filter
+	words := strings.Fields(strings.ToLower(query))
+	var terms []string
+
+	for _, word := range words {
+		// Remove punctuation
+		word = strings.Trim(word, ".,?!;:'\"()[]{}")
+		if len(word) < 2 {
+			continue
+		}
+		if stopWords[word] {
+			continue
+		}
+		terms = append(terms, word)
+	}
+
+	return terms
+}
+
+// makeDeclarative converts a question to a declarative statement.
+func makeDeclarative(query string) string {
+	query = strings.TrimSpace(query)
+
+	// Remove question mark
+	query = strings.TrimSuffix(query, "?")
+
+	// Handle common question patterns
+	patterns := []struct {
+		prefix      string
+		replacement string
+	}{
+		{"how do i ", ""},
+		{"how to ", ""},
+		{"how does ", ""},
+		{"how is ", ""},
+		{"what is ", ""},
+		{"what are ", ""},
+		{"why does ", ""},
+		{"why is ", ""},
+		{"where is ", ""},
+		{"where are ", ""},
+		{"when does ", ""},
+		{"when is ", ""},
+	}
+
+	lower := strings.ToLower(query)
+	for _, p := range patterns {
+		if strings.HasPrefix(lower, p.prefix) {
+			return strings.TrimSpace(query[len(p.prefix):])
+		}
+	}
+
+	return query
+}
+
+// deduplicateExpansions removes duplicate queries while preserving order.
+func deduplicateExpansions(expansions []ExpandedQuery) []ExpandedQuery {
+	seen := make(map[string]bool)
+	result := make([]ExpandedQuery, 0, len(expansions))
+
+	for _, exp := range expansions {
+		normalized := strings.ToLower(strings.TrimSpace(exp.Query))
+		if !seen[normalized] {
+			seen[normalized] = true
+			result = append(result, exp)
+		}
+	}
+
+	return result
+}
+
+// cosineSimilarity computes cosine similarity between two vectors.
+func cosineSimilarity(a, b []float32) float64 {
+	if len(a) != len(b) || len(a) == 0 {
+		return 0
+	}
+
+	var dot, normA, normB float64
+	for i := range a {
+		dot += float64(a[i]) * float64(b[i])
+		normA += float64(a[i]) * float64(a[i])
+		normB += float64(b[i]) * float64(b[i])
+	}
+
+	if normA == 0 || normB == 0 {
+		return 0
+	}
+
+	return dot / (sqrt(normA) * sqrt(normB))
+}
+
+// sqrt is a simple square root implementation.
+func sqrt(x float64) float64 {
+	if x <= 0 {
+		return 0
+	}
+	z := x
+	for i := 0; i < 10; i++ {
+		z = (z + x/z) / 2
+	}
+	return z
+}
+
+// truncate truncates a string to maxLen characters.
+func truncate(s string, maxLen int) string {
+	if len(s) <= maxLen {
+		return s
+	}
+	return s[:maxLen] + "..."
+}

internal/search/expansion/expander_test.go

@@ -0,0 +1,518 @@
+// Package expansion provides context-aware query expansion for improved search recall.
+package expansion
+
+import (
+	"context"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/suite"
+)
+
+// ExpanderSuite tests the Expander functionality.
+type ExpanderSuite struct {
+	suite.Suite
+	expander *Expander
+}
+
+func TestExpanderSuite(t *testing.T) {
+	suite.Run(t, new(ExpanderSuite))
+}
+
+func (s *ExpanderSuite) SetupTest() {
+	// Create expander without embedding service for basic tests
+	s.expander = NewExpander(nil)
+}
+
+// TestNewExpander tests expander creation.
+func (s *ExpanderSuite) TestNewExpander() {
+	e := NewExpander(nil)
+	s.NotNil(e)
+	s.NotNil(e.intentPatterns)
+	s.Nil(e.embedSvc)
+}
+
+// TestDetectIntent tests intent detection.
+func (s *ExpanderSuite) TestDetectIntent() {
+	tests := []struct {
+		name     string
+		query    string
+		expected QueryIntent
+	}{
+		// Question intents
+		{"how_question", "how do I implement auth?", IntentQuestion},
+		{"why_question", "why does this fail?", IntentError}, // "fail" triggers error first
+		{"what_question", "what is the purpose of this function?", IntentQuestion},
+		{"question_mark", "the handler for auth?", IntentQuestion},
+		{"explain", "explain the architecture", IntentQuestion},
+
+		// Error intents
+		{"error_word", "authentication error in login", IntentError},
+		{"bug_word", "bug in user registration", IntentError},
+		{"fix_word", "fix the memory leak", IntentError},
+		{"not_working", "login not working", IntentError},
+		{"crash", "application crash on startup", IntentError},
+
+		// Implementation intents
+		{"implement", "implement user authentication", IntentImplementation},
+		{"add_feature", "add new endpoint for users", IntentImplementation},
+		{"create", "create a handler for uploads", IntentImplementation},
+		{"function", "function to validate input", IntentImplementation},
+
+		// Architecture intents
+		{"architecture", "architecture of the system", IntentArchitecture},
+		{"design", "design pattern for observers", IntentArchitecture},
+		{"component", "component structure", IntentArchitecture},
+		{"flow", "data flow in the pipeline", IntentArchitecture},
+
+		// General intents
+		{"general", "user authentication", IntentGeneral},
+		{"empty", "", IntentGeneral},
+		{"simple", "database", IntentGeneral},
+	}
+
+	for _, tt := range tests {
+		s.Run(tt.name, func() {
+			result := s.expander.DetectIntent(tt.query)
+			s.Equal(tt.expected, result, "Query: %s", tt.query)
+		})
+	}
+}
+
+// TestExpand tests basic query expansion.
+func (s *ExpanderSuite) TestExpand() {
+	ctx := context.Background()
+	cfg := DefaultConfig()
+	cfg.EnableVocabularyExpansion = false // Disable for unit test
+
+	tests := []struct {
+		name           string
+		query          string
+		expectedIntent QueryIntent
+		minExpansions  int
+		hasOriginal    bool
+	}{
+		{name: "question", query: "how do I implement auth", expectedIntent: IntentQuestion, minExpansions: 1, hasOriginal: true},
+		{name: "error", query: "fix the bug in login", expectedIntent: IntentError, minExpansions: 1, hasOriginal: true},
+		{name: "implementation", query: "implement user handler", expectedIntent: IntentImplementation, minExpansions: 1, hasOriginal: true},
+		{name: "architecture", query: "architecture design", expectedIntent: IntentArchitecture, minExpansions: 1, hasOriginal: true},
+		{name: "general", query: "database connection", expectedIntent: IntentGeneral, minExpansions: 1, hasOriginal: true},
+		{name: "empty", query: "", expectedIntent: IntentGeneral, minExpansions: 0, hasOriginal: false},
+	}
+
+	for _, tt := range tests {
+		s.Run(tt.name, func() {
+			expansions := s.expander.Expand(ctx, tt.query, cfg)
+
+			if tt.minExpansions == 0 {
+				s.Empty(expansions)
+				return
+			}
+
+			s.GreaterOrEqual(len(expansions), tt.minExpansions)
+
+			if tt.hasOriginal {
+				// First expansion should be the original
+				s.Equal(tt.query, expansions[0].Query)
+				s.Equal(1.0, expansions[0].Weight)
+				s.Equal("original", expansions[0].Source)
+			}
+		})
+	}
+}
+
+// TestExpandWithConfig tests expansion with custom config.
+func (s *ExpanderSuite) TestExpandWithConfig() {
+	ctx := context.Background()
+
+	cfg := Config{
+		MaxExpansions:             2,
+		MinSimilarity:             0.7,
+		EnableVocabularyExpansion: false,
+	}
+
+	expansions := s.expander.Expand(ctx, "how to implement authentication", cfg)
+	s.LessOrEqual(len(expansions), cfg.MaxExpansions)
+}
+
+// TestExpandDeduplication tests that duplicates are removed.
+func (s *ExpanderSuite) TestExpandDeduplication() {
+	ctx := context.Background()
+	cfg := DefaultConfig()
+	cfg.EnableVocabularyExpansion = false
+
+	// Query that might generate duplicate expansions
+	query := "how to fix authentication"
+	expansions := s.expander.Expand(ctx, query, cfg)
+
+	// Check for duplicates
+	seen := make(map[string]bool)
+	for _, exp := range expansions {
+		normalized := exp.Query
+		s.False(seen[normalized], "Duplicate expansion found: %s", exp.Query)
+		seen[normalized] = true
+	}
+}
+
+// TestExtractKeyTerms tests key term extraction.
+func TestExtractKeyTerms(t *testing.T) {
+	tests := []struct {
+		name     string
+		query    string
+		expected []string
+	}{
+		{
+			name:     "simple",
+			query:    "user authentication handler",
+			expected: []string{"user", "authentication", "handler"},
+		},
+		{
+			name:     "with_stop_words",
+			query:    "how to implement the user login",
+			expected: []string{"implement", "user", "login"},
+		},
+		{
+			name:     "with_punctuation",
+			query:    "fix the bug, please!",
+			expected: []string{"fix", "bug", "please"},
+		},
+		{
+			name:     "empty",
+			query:    "",
+			expected: nil,
+		},
+		{
+			name:     "only_stop_words",
+			query:    "the a an is are",
+			expected: nil,
+		},
+		{
+			name:     "short_words_filtered",
+			query:    "a b c auth",
+			expected: []string{"auth"},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := extractKeyTerms(tt.query)
+			if tt.expected == nil {
+				assert.Empty(t, result)
+			} else {
+				assert.Equal(t, tt.expected, result)
+			}
+		})
+	}
+}
+
+// TestMakeDeclarative tests question to declarative conversion.
+func TestMakeDeclarative(t *testing.T) {
+	tests := []struct {
+		name     string
+		query    string
+		expected string
+	}{
+		{
+			name:     "how_do_i",
+			query:    "how do I implement auth?",
+			expected: "implement auth",
+		},
+		{
+			name:     "how_to",
+			query:    "how to fix the bug",
+			expected: "fix the bug",
+		},
+		{
+			name:     "what_is",
+			query:    "what is the purpose of this?",
+			expected: "the purpose of this",
+		},
+		{
+			name:     "why_does",
+			query:    "why does this fail?",
+			expected: "this fail",
+		},
+		{
+			name:     "already_declarative",
+			query:    "user authentication",
+			expected: "user authentication",
+		},
+		{
+			name:     "question_mark_only",
+			query:    "authentication?",
+			expected: "authentication",
+		},
+		{
+			name:     "case_insensitive",
+			query:    "How To Fix Auth?",
+			expected: "Fix Auth",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := makeDeclarative(tt.query)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+// TestDeduplicateExpansions tests deduplication.
+func TestDeduplicateExpansions(t *testing.T) {
+	expansions := []ExpandedQuery{
+		{Query: "auth handler", Weight: 1.0},
+		{Query: "AUTH HANDLER", Weight: 0.8},  // Duplicate (case insensitive)
+		{Query: "auth handler ", Weight: 0.7}, // Duplicate (whitespace)
+		{Query: "user auth", Weight: 0.6},
+	}
+
+	result := deduplicateExpansions(expansions)
+	assert.Len(t, result, 2) // "auth handler" and "user auth"
+	assert.Equal(t, "auth handler", result[0].Query)
+	assert.Equal(t, 1.0, result[0].Weight) // First one preserved
+	assert.Equal(t, "user auth", result[1].Query)
+}
+
+// TestCosineSimilarity tests cosine similarity calculation.
+func TestCosineSimilarity(t *testing.T) {
+	tests := []struct {
+		name     string
+		a        []float32
+		b        []float32
+		expected float64
+		delta    float64
+	}{
+		{
+			name:     "identical_vectors",
+			a:        []float32{1, 0, 0},
+			b:        []float32{1, 0, 0},
+			expected: 1.0,
+			delta:    0.001,
+		},
+		{
+			name:     "orthogonal_vectors",
+			a:        []float32{1, 0, 0},
+			b:        []float32{0, 1, 0},
+			expected: 0.0,
+			delta:    0.001,
+		},
+		{
+			name:     "opposite_vectors",
+			a:        []float32{1, 0, 0},
+			b:        []float32{-1, 0, 0},
+			expected: -1.0,
+			delta:    0.001,
+		},
+		{
+			name:     "similar_vectors",
+			a:        []float32{1, 1, 0},
+			b:        []float32{1, 0, 0},
+			expected: 0.707,
+			delta:    0.01,
+		},
+		{
+			name:     "empty_vectors",
+			a:        []float32{},
+			b:        []float32{},
+			expected: 0.0,
+			delta:    0.001,
+		},
+		{
+			name:     "different_lengths",
+			a:        []float32{1, 0},
+			b:        []float32{1, 0, 0},
+			expected: 0.0,
+			delta:    0.001,
+		},
+		{
+			name:     "zero_vector",
+			a:        []float32{0, 0, 0},
+			b:        []float32{1, 1, 1},
+			expected: 0.0,
+			delta:    0.001,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := cosineSimilarity(tt.a, tt.b)
+			assert.InDelta(t, tt.expected, result, tt.delta)
+		})
+	}
+}
+
+// TestDefaultConfig tests default configuration.
+func TestDefaultConfig(t *testing.T) {
+	cfg := DefaultConfig()
+
+	assert.Equal(t, 4, cfg.MaxExpansions)
+	assert.Equal(t, 0.5, cfg.MinSimilarity)
+	assert.True(t, cfg.EnableVocabularyExpansion)
+}
+
+// TestExpandedQueryStruct tests ExpandedQuery struct.
+func TestExpandedQueryStruct(t *testing.T) {
+	eq := ExpandedQuery{
+		Query:  "test query",
+		Weight: 0.85,
+		Source: "vocabulary:auth",
+		Intent: IntentQuestion,
+	}
+
+	assert.Equal(t, "test query", eq.Query)
+	assert.Equal(t, 0.85, eq.Weight)
+	assert.Equal(t, "vocabulary:auth", eq.Source)
+	assert.Equal(t, IntentQuestion, eq.Intent)
+}
+
+// TestVocabEntry tests VocabEntry struct.
+func TestVocabEntry(t *testing.T) {
+	ve := VocabEntry{
+		Term:   "authentication",
+		Weight: 0.9,
+		Source: "concept",
+	}
+
+	assert.Equal(t, "authentication", ve.Term)
+	assert.Equal(t, 0.9, ve.Weight)
+	assert.Equal(t, "concept", ve.Source)
+}
+
+// TestIntentConstants tests intent constant values.
+func TestIntentConstants(t *testing.T) {
+	assert.Equal(t, QueryIntent("question"), IntentQuestion)
+	assert.Equal(t, QueryIntent("error"), IntentError)
+	assert.Equal(t, QueryIntent("implementation"), IntentImplementation)
+	assert.Equal(t, QueryIntent("architecture"), IntentArchitecture)
+	assert.Equal(t, QueryIntent("general"), IntentGeneral)
+}
+
+// TestTruncate tests the truncate helper.
+func TestTruncate(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    string
+		expected string
+		maxLen   int
+	}{
+		{name: "short", input: "hello", maxLen: 10, expected: "hello"},
+		{name: "exact", input: "hello", maxLen: 5, expected: "hello"},
+		{name: "long", input: "hello world", maxLen: 5, expected: "hello..."},
+		{name: "empty", input: "", maxLen: 10, expected: ""},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := truncate(tt.input, tt.maxLen)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+// TestSqrt tests the sqrt helper.
+func TestSqrt(t *testing.T) {
+	tests := []struct {
+		input    float64
+		expected float64
+		delta    float64
+	}{
+		{4.0, 2.0, 0.001},
+		{9.0, 3.0, 0.001},
+		{16.0, 4.0, 0.001},
+		{2.0, 1.414, 0.01},
+		{0.0, 0.0, 0.001},
+		{-1.0, 0.0, 0.001},
+	}
+
+	for _, tt := range tests {
+		t.Run("", func(t *testing.T) {
+			result := sqrt(tt.input)
+			assert.InDelta(t, tt.expected, result, tt.delta)
+		})
+	}
+}
+
+// TestExpandByIntentError tests error intent expansion.
+func (s *ExpanderSuite) TestExpandByIntentError() {
+	expansions := s.expander.expandByIntent("fix authentication bug", IntentError)
+	s.NotEmpty(expansions)
+
+	// Should have solution-oriented expansion
+	hasSolution := false
+	for _, exp := range expansions {
+		if exp.Source == "intent:solution" {
+			hasSolution = true
+			break
+		}
+	}
+	s.True(hasSolution)
+}
+
+// TestExpandByIntentQuestion tests question intent expansion.
+func (s *ExpanderSuite) TestExpandByIntentQuestion() {
+	expansions := s.expander.expandByIntent("how do I implement auth", IntentQuestion)
+	s.NotEmpty(expansions)
+
+	// Should have declarative expansion
+	hasDeclarative := false
+	for _, exp := range expansions {
+		if exp.Source == "intent:declarative" {
+			hasDeclarative = true
+			break
+		}
+	}
+	s.True(hasDeclarative)
+}
+
+// TestExpandByIntentImplementation tests implementation intent expansion.
+func (s *ExpanderSuite) TestExpandByIntentImplementation() {
+	expansions := s.expander.expandByIntent("implement user handler", IntentImplementation)
+	s.NotEmpty(expansions)
+
+	// Should have how expansion
+	hasHow := false
+	for _, exp := range expansions {
+		if exp.Source == "intent:how" {
+			hasHow = true
+			break
+		}
+	}
+	s.True(hasHow)
+}
+
+// TestExpandByIntentArchitecture tests architecture intent expansion.
+func (s *ExpanderSuite) TestExpandByIntentArchitecture() {
+	expansions := s.expander.expandByIntent("system architecture design", IntentArchitecture)
+	s.NotEmpty(expansions)
+
+	// Should have design expansion
+	hasDesign := false
+	for _, exp := range expansions {
+		if exp.Source == "intent:design" {
+			hasDesign = true
+			break
+		}
+	}
+	s.True(hasDesign)
+}
+
+// TestExpandByIntentGeneral tests general intent returns no expansions.
+func (s *ExpanderSuite) TestExpandByIntentGeneral() {
+	expansions := s.expander.expandByIntent("database", IntentGeneral)
+	s.Empty(expansions) // General intent doesn't add intent-based expansions
+}
+
+// TestEmptyVocabulary tests expansion with empty vocabulary.
+func (s *ExpanderSuite) TestEmptyVocabulary() {
+	ctx := context.Background()
+	expansions := s.expander.expandByVocabulary(ctx, "test query", 0.5)
+	s.Empty(expansions)
+}
+
+// TestIntentPatternsExist tests that all intents have patterns.
+func (s *ExpanderSuite) TestIntentPatternsExist() {
+	s.NotEmpty(s.expander.intentPatterns[IntentQuestion])
+	s.NotEmpty(s.expander.intentPatterns[IntentError])
+	s.NotEmpty(s.expander.intentPatterns[IntentImplementation])
+	s.NotEmpty(s.expander.intentPatterns[IntentArchitecture])
+}

internal/search/integration_test.go

@@ -517,16 +517,16 @@ func TestTruncate_TableDriven(t *testing.T) {
 	tests := []struct {
 		name     string
 		input    string
-		maxLen   int
 		expected string
+		maxLen   int
 	}{
-		{"short_string", "hello", 10, "hello"},
-		{"exact_length", "hello", 5, "hello"},
-		{"long_string", "hello world", 5, "hello..."},
-		{"empty_string", "", 10, ""},
-		{"whitespace_only", "   ", 10, ""},
-		{"with_leading_space", "  hello  ", 10, "hello"},
-		{"very_long", "this is a very long string that should be truncated", 20, "this is a very long ..."},
+		{name: "short_string", input: "hello", maxLen: 10, expected: "hello"},
+		{name: "exact_length", input: "hello", maxLen: 5, expected: "hello"},
+		{name: "long_string", input: "hello world", maxLen: 5, expected: "hello..."},
+		{name: "empty_string", input: "", maxLen: 10, expected: ""},
+		{name: "whitespace_only", input: "   ", maxLen: 10, expected: ""},
+		{name: "with_leading_space", input: "  hello  ", maxLen: 10, expected: "hello"},
+		{name: "very_long", input: "this is a very long string that should be truncated", maxLen: 20, expected: "this is a very long ..."},
 	}
 
 	for _, tt := range tests {

internal/search/manager.go

@@ -35,40 +35,41 @@ func NewManager(
 
 // SearchParams contains parameters for unified search.
 type SearchParams struct {
-	Query         string
-	Type          string // "observations", "sessions", "prompts", or empty for all
-	Project       string
-	ObsType       string // Observation type filter
-	Concepts      string
-	Files         string
-	DateStart     int64
-	DateEnd       int64
-	OrderBy       string // "relevance", "date_desc", "date_asc"
-	Limit         int
-	Offset        int
-	Format        string // "index" or "full"
-	Scope         string // "project", "global", or empty for project+global
-	IncludeGlobal bool   // If true, include global observations along with project-scoped
+	Format            string
+	Type              string
+	Project           string
+	ObsType           string
+	Concepts          string
+	Files             string
+	Query             string
+	Scope             string
+	OrderBy           string
+	DateStart         int64
+	Offset            int
+	Limit             int
+	DateEnd           int64
+	IncludeGlobal     bool
+	ExcludeSuperseded bool
 }
 
 // SearchResult represents a unified search result.
 type SearchResult struct {
-	Type      string                 `json:"type"` // "observation", "session", "prompt"
-	ID        int64                  `json:"id"`
+	Metadata  map[string]interface{} `json:"metadata,omitempty"`
+	Type      string                 `json:"type"`
 	Title     string                 `json:"title,omitempty"`
 	Content   string                 `json:"content,omitempty"`
 	Project   string                 `json:"project"`
-	Scope     string                 `json:"scope,omitempty"` // "project" or "global"
+	Scope     string                 `json:"scope,omitempty"`
+	ID        int64                  `json:"id"`
 	CreatedAt int64                  `json:"created_at_epoch"`
 	Score     float64                `json:"score,omitempty"`
-	Metadata  map[string]interface{} `json:"metadata,omitempty"`
 }
 
 // UnifiedSearchResult contains the combined search results.
 type UnifiedSearchResult struct {
+	Query      string         `json:"query,omitempty"`
 	Results    []SearchResult `json:"results"`
 	TotalCount int            `json:"total_count"`
-	Query      string         `json:"query,omitempty"`
 }
 
 // UnifiedSearch performs a unified search across all document types.
@@ -126,6 +127,10 @@ func (m *Manager) vectorSearch(ctx context.Context, params SearchParams) (*Unifi
 		obs, err := m.observationStore.GetObservationsByIDs(ctx, obsIDs, params.OrderBy, 0)
 		if err == nil {
 			for _, o := range obs {
+				// Skip superseded observations when requested
+				if params.ExcludeSuperseded && o.IsSuperseded {
+					continue
+				}
 				results = append(results, m.observationToResult(o, params.Format))
 			}
 		}
@@ -167,7 +172,16 @@ func (m *Manager) filterSearch(ctx context.Context, params SearchParams) (*Unifi
 
 	// Search observations
 	if params.Type == "" || params.Type == "observations" {
-		obs, err := m.observationStore.GetRecentObservations(ctx, params.Project, params.Limit)
+		var obs []*models.Observation
+		var err error
+
+		// Use active observations (excluding superseded) when requested
+		if params.ExcludeSuperseded {
+			obs, err = m.observationStore.GetActiveObservations(ctx, params.Project, params.Limit)
+		} else {
+			obs, err = m.observationStore.GetRecentObservations(ctx, params.Project, params.Limit)
+		}
+
 		if err == nil {
 			for _, o := range obs {
 				results = append(results, m.observationToResult(o, params.Format))

internal/search/manager_test.go

@@ -4,7 +4,6 @@ package search
 import (
 	"database/sql"
 	"testing"
-	"time"
 
 	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
 	"github.com/stretchr/testify/assert"
@@ -94,8 +93,8 @@ func TestTruncate(t *testing.T) {
 	tests := []struct {
 		name     string
 		input    string
-		maxLen   int
 		expected string
+		maxLen   int
 	}{
 		{
 			name:     "short string no truncation",
@@ -147,11 +146,11 @@ func TestTruncate(t *testing.T) {
 func TestObservationToResult(t *testing.T) {
 	m := NewManager(nil, nil, nil, nil)
 
-	tests := []struct {
-		name     string
+	tests := []struct { //nolint:govet
 		obs      *models.Observation
-		format   string
 		expected SearchResult
+		format   string
+		name     string
 	}{
 		{
 			name: "full format with all fields",
@@ -239,11 +238,11 @@ func TestObservationToResult(t *testing.T) {
 func TestSummaryToResult(t *testing.T) {
 	m := NewManager(nil, nil, nil, nil)
 
-	tests := []struct {
-		name     string
+	tests := []struct { //nolint:govet
 		summary  *models.SessionSummary
-		format   string
 		expected SearchResult
+		format   string
+		name     string
 	}{
 		{
 			name: "full format with all fields",
@@ -321,11 +320,11 @@ func TestSummaryToResult(t *testing.T) {
 func TestPromptToResult(t *testing.T) {
 	m := NewManager(nil, nil, nil, nil)
 
-	tests := []struct {
-		name     string
+	tests := []struct { //nolint:govet
 		prompt   *models.UserPromptWithSession
-		format   string
 		expected SearchResult
+		format   string
+		name     string
 	}{
 		{
 			name: "full format with content",
@@ -406,9 +405,9 @@ func TestPromptToResult(t *testing.T) {
 func TestSearchParamsValidation(t *testing.T) {
 	tests := []struct {
 		name          string
+		expectedOrder string
 		params        SearchParams
 		expectedLimit int
-		expectedOrder string
 	}{
 		{
 			name: "default limit applied",
@@ -731,23 +730,21 @@ func TestPromptToResultFormats(t *testing.T) {
 func TestSearchParamsDefaults(t *testing.T) {
 	tests := []struct {
 		name          string
-		initialLimit  int
 		initialOrder  string
-		expectedLimit int
 		expectedOrder string
+		initialLimit  int
+		expectedLimit int
 	}{
-		{"zero_limit", 0, "", 20, "date_desc"},
-		{"negative_limit", -5, "", 20, "date_desc"},
-		{"over_100_limit", 150, "", 100, "date_desc"},
-		{"valid_limit_50", 50, "relevance", 50, "relevance"},
-		{"custom_order", 30, "date_asc", 30, "date_asc"},
+		{name: "zero_limit", initialLimit: 0, initialOrder: "", expectedLimit: 20, expectedOrder: "date_desc"},
+		{name: "negative_limit", initialLimit: -5, initialOrder: "", expectedLimit: 20, expectedOrder: "date_desc"},
+		{name: "over_100_limit", initialLimit: 150, initialOrder: "", expectedLimit: 100, expectedOrder: "date_desc"},
+		{name: "valid_limit_50", initialLimit: 50, initialOrder: "relevance", expectedLimit: 50, expectedOrder: "relevance"},
+		{name: "custom_order", initialLimit: 30, initialOrder: "date_asc", expectedLimit: 30, expectedOrder: "date_asc"},
 	}
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			params := SearchParams{
-				Query:   "test",
-				Project: "project",
 				Limit:   tt.initialLimit,
 				OrderBy: tt.initialOrder,
 			}
@@ -774,18 +771,18 @@ func TestTruncateEdgeCases(t *testing.T) {
 	tests := []struct {
 		name     string
 		input    string
-		maxLen   int
 		expected string
+		maxLen   int
 	}{
 		// Unicode strings - uses byte length so ensure maxLen accommodates full string
-		{"unicode_string_no_truncate", "日本語テスト", 20, "日本語テスト"},
-		{"mixed_unicode_no_truncate", "Hello世界", 15, "Hello世界"},
+		{name: "unicode_string_no_truncate", input: "日本語テスト", maxLen: 20, expected: "日本語テスト"},
+		{name: "mixed_unicode_no_truncate", input: "Hello世界", maxLen: 15, expected: "Hello世界"},
 		// ASCII truncation
-		{"ascii_truncate", "Hello World", 5, "Hello..."},
-		{"only_whitespace", "   ", 10, ""},
-		{"tabs_and_newlines", "\t\n  \t", 10, ""},
-		{"newlines_with_content", "\n\nhello\n\n", 10, "hello"},
-		{"zero_max_len", "hello", 0, "..."},
+		{name: "ascii_truncate", input: "Hello World", maxLen: 5, expected: "Hello..."},
+		{name: "only_whitespace", input: "   ", maxLen: 10, expected: ""},
+		{name: "tabs_and_newlines", input: "\t\n  \t", maxLen: 10, expected: ""},
+		{name: "newlines_with_content", input: "\n\nhello\n\n", maxLen: 10, expected: "hello"},
+		{name: "zero_max_len", input: "hello", maxLen: 0, expected: "..."},
 	}
 
 	for _, tt := range tests {
@@ -812,8 +809,6 @@ func TestUnifiedSearchResultEmpty(t *testing.T) {
 // TestSearchResultMetadata tests SearchResult metadata handling.
 func TestSearchResultMetadata(t *testing.T) {
 	result := SearchResult{
-		Type: "observation",
-		ID:   1,
 		Metadata: map[string]interface{}{
 			"obs_type": "discovery",
 			"scope":    "project",
@@ -835,10 +830,7 @@ func TestSearchResultTypes(t *testing.T) {
 	for _, typ := range types {
 		t.Run(typ, func(t *testing.T) {
 			result := SearchResult{
-				Type:      typ,
-				ID:        1,
-				Project:   "test",
-				CreatedAt: time.Now().UnixMilli(),
+				Type: typ,
 			}
 			assert.Equal(t, typ, result.Type)
 		})
@@ -952,8 +944,8 @@ func TestSearchParams_OrderByValues(t *testing.T) {
 	for _, order := range validOrders {
 		t.Run("order_"+order, func(t *testing.T) {
 			params := SearchParams{
-				Query:   "test",
-				Project: "test",
+				Query:   "test", //nolint:govet
+				Project: "test", //nolint:govet
 				OrderBy: order,
 			}
 			assert.Equal(t, order, params.OrderBy)
@@ -968,9 +960,7 @@ func TestSearchParams_TypeValues(t *testing.T) {
 	for _, typ := range validTypes {
 		t.Run("type_"+typ, func(t *testing.T) {
 			params := SearchParams{
-				Query:   "test",
-				Project: "test",
-				Type:    typ,
+				Type: typ,
 			}
 			assert.Equal(t, typ, params.Type)
 		})
@@ -984,8 +974,8 @@ func TestSearchParams_ScopeValues(t *testing.T) {
 	for _, scope := range validScopes {
 		t.Run("scope_"+scope, func(t *testing.T) {
 			params := SearchParams{
-				Query:   "test",
-				Project: "test",
+				Query:   "test", //nolint:govet
+				Project: "test", //nolint:govet
 				Scope:   scope,
 			}
 			assert.Equal(t, scope, params.Scope)
@@ -1000,8 +990,8 @@ func TestSearchParams_FormatValues(t *testing.T) {
 	for _, format := range validFormats {
 		t.Run("format_"+format, func(t *testing.T) {
 			params := SearchParams{
-				Query:   "test",
-				Project: "test",
+				Query:   "test", //nolint:govet
+				Project: "test", //nolint:govet
 				Format:  format,
 			}
 			assert.Equal(t, format, params.Format)
@@ -1020,7 +1010,7 @@ func TestUnifiedSearchResult_MultipleResults(t *testing.T) {
 	result := UnifiedSearchResult{
 		Results:    results,
 		TotalCount: 3,
-		Query:      "test query",
+		Query:      "test query", //nolint:govet
 	}
 
 	assert.Len(t, result.Results, 3)
@@ -1040,8 +1030,8 @@ func TestSearchResult_Metadata(t *testing.T) {
 	}
 
 	result := SearchResult{
-		Type:     "observation",
-		ID:       1,
+		Type:     "observation", //nolint:govet
+		ID:       1,             //nolint:govet
 		Metadata: metadata,
 	}
 
@@ -1066,8 +1056,6 @@ func TestSearchResult_Scores(t *testing.T) {
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			result := SearchResult{
-				Type:  "observation",
-				ID:    1,
 				Score: tt.score,
 			}
 			assert.Equal(t, tt.score, result.Score)

internal/update/update.go

@@ -33,11 +33,11 @@ const (
 
 // Release represents a GitHub release.
 type Release struct {
+	PublishedAt time.Time `json:"published_at"`
 	TagName     string    `json:"tag_name"`
 	Name        string    `json:"name"`
-	PublishedAt time.Time `json:"published_at"`
-	Assets      []Asset   `json:"assets"`
 	Body        string    `json:"body"`
+	Assets      []Asset   `json:"assets"`
 }
 
 // Asset represents a release asset.
@@ -49,15 +49,15 @@ type Asset struct {
 
 // UpdateInfo contains information about an available update.
 type UpdateInfo struct {
-	Available           bool      `json:"available"`
+	PublishedAt         time.Time `json:"published_at,omitempty"`
 	CurrentVersion      string    `json:"current_version"`
 	LatestVersion       string    `json:"latest_version"`
 	ReleaseNotes        string    `json:"release_notes,omitempty"`
-	PublishedAt         time.Time `json:"published_at,omitempty"`
 	DownloadURL         string    `json:"download_url,omitempty"`
 	ChecksumsURL        string    `json:"checksums_url,omitempty"`
-	BundleURL           string    `json:"bundle_url,omitempty"` // Sigstore bundle (.sigstore.json)
+	BundleURL           string    `json:"bundle_url,omitempty"`
 	ManualUpdateCommand string    `json:"manual_update_command,omitempty"`
+	Available           bool      `json:"available"`
 }
 
 // InstallScriptURL is the URL to the remote installation script.
@@ -74,23 +74,22 @@ func GetManualUpdateCommand(version string) string {
 
 // UpdateStatus represents the current update status.
 type UpdateStatus struct {
-	State               string  `json:"state"` // "idle", "checking", "downloading", "verifying", "applying", "done", "error"
-	Progress            float64 `json:"progress"`
+	State               string  `json:"state"`
 	Message             string  `json:"message"`
 	Error               string  `json:"error,omitempty"`
-	ManualUpdateCommand string  `json:"manual_update_command,omitempty"` // Shown when update fails
+	ManualUpdateCommand string  `json:"manual_update_command,omitempty"`
+	Progress            float64 `json:"progress"`
 }
 
 // Updater handles self-updates.
-type Updater struct {
+type Updater struct { //nolint:govet
+	httpClient     *http.Client
+	cachedUpdate   *UpdateInfo
+	lastCheck      time.Time
+	status         UpdateStatus
 	currentVersion string
 	installDir     string
-	httpClient     *http.Client
-
-	mu           sync.RWMutex
-	status       UpdateStatus
-	lastCheck    time.Time
-	cachedUpdate *UpdateInfo
+	mu             sync.RWMutex
 }
 
 // New creates a new Updater.

internal/vector/hybrid/autotuner.go

@@ -0,0 +1,309 @@
+package hybrid
+
+import (
+	"context"
+	"sync"
+	"time"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/vector/sqlitevec"
+	"github.com/rs/zerolog/log"
+)
+
+// AutoTuner dynamically adjusts hub threshold based on query performance
+type AutoTuner struct {
+	ctx           context.Context
+	client        *Client
+	cancel        context.CancelFunc
+	latencies     []time.Duration
+	wg            sync.WaitGroup
+	queries       int64
+	targetLatency time.Duration
+	adjustPeriod  time.Duration
+	minThreshold  int
+	maxThreshold  int
+	adjustments   int
+	latenciesMu   sync.Mutex
+}
+
+// AutoTunerConfig configures the auto-tuner
+type AutoTunerConfig struct {
+	TargetLatency time.Duration // Target p95 latency (default: 50ms)
+	MinThreshold  int           // Min hub threshold (default: 2)
+	MaxThreshold  int           // Max hub threshold (default: 20)
+	AdjustPeriod  time.Duration // Adjustment frequency (default: 5min)
+}
+
+// DefaultAutoTunerConfig returns sensible defaults
+func DefaultAutoTunerConfig() AutoTunerConfig {
+	return AutoTunerConfig{
+		TargetLatency: 50 * time.Millisecond,
+		MinThreshold:  2,
+		MaxThreshold:  20,
+		AdjustPeriod:  5 * time.Minute,
+	}
+}
+
+// NewAutoTuner creates a new auto-tuner for the hybrid client
+func NewAutoTuner(client *Client, cfg AutoTunerConfig) *AutoTuner {
+	ctx, cancel := context.WithCancel(context.Background())
+
+	tuner := &AutoTuner{
+		client:        client,
+		targetLatency: cfg.TargetLatency,
+		minThreshold:  cfg.MinThreshold,
+		maxThreshold:  cfg.MaxThreshold,
+		adjustPeriod:  cfg.AdjustPeriod,
+		latencies:     make([]time.Duration, 0, 1000),
+		ctx:           ctx,
+		cancel:        cancel,
+	}
+
+	return tuner
+}
+
+// Start begins auto-tuning in the background
+func (a *AutoTuner) Start() {
+	a.wg.Add(1)
+	go a.tuningLoop()
+
+	log.Info().
+		Dur("target_latency", a.targetLatency).
+		Int("min_threshold", a.minThreshold).
+		Int("max_threshold", a.maxThreshold).
+		Dur("adjust_period", a.adjustPeriod).
+		Msg("Auto-tuner started")
+}
+
+// Stop stops the auto-tuner
+func (a *AutoTuner) Stop() {
+	a.cancel()
+	a.wg.Wait()
+	log.Info().Msg("Auto-tuner stopped")
+}
+
+// RecordQuery records a query latency for analysis
+func (a *AutoTuner) RecordQuery(latency time.Duration) {
+	a.latenciesMu.Lock()
+	defer a.latenciesMu.Unlock()
+
+	a.queries++
+	a.latencies = append(a.latencies, latency)
+
+	// Keep only recent queries (last 1000)
+	if len(a.latencies) > 1000 {
+		a.latencies = a.latencies[len(a.latencies)-1000:]
+	}
+}
+
+// tuningLoop periodically adjusts hub threshold
+func (a *AutoTuner) tuningLoop() {
+	defer a.wg.Done()
+
+	ticker := time.NewTicker(a.adjustPeriod)
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-a.ctx.Done():
+			return
+
+		case <-ticker.C:
+			a.adjustThreshold()
+		}
+	}
+}
+
+// adjustThreshold analyzes recent queries and adjusts hub threshold
+func (a *AutoTuner) adjustThreshold() {
+	a.latenciesMu.Lock()
+	defer a.latenciesMu.Unlock()
+
+	if len(a.latencies) < 10 {
+		// Not enough data yet
+		return
+	}
+
+	// Calculate p95 latency
+	p95 := calculateP95(a.latencies)
+
+	currentThreshold := a.client.hubThreshold
+
+	log.Debug().
+		Dur("p95_latency", p95).
+		Dur("target_latency", a.targetLatency).
+		Int("current_threshold", currentThreshold).
+		Int("queries", len(a.latencies)).
+		Msg("Auto-tuner evaluating performance")
+
+	// Determine adjustment direction
+	var newThreshold int
+
+	if p95 > a.targetLatency {
+		// Too slow - lower threshold (more hubs = faster queries)
+		adjustment := calculateAdjustment(p95, a.targetLatency)
+		newThreshold = currentThreshold - adjustment
+
+		if newThreshold < a.minThreshold {
+			newThreshold = a.minThreshold
+		}
+
+		log.Info().
+			Dur("p95", p95).
+			Int("old_threshold", currentThreshold).
+			Int("new_threshold", newThreshold).
+			Msg("Auto-tuner: Lowering hub threshold (too slow)")
+
+	} else if p95 < a.targetLatency*8/10 {
+		// Too fast - raise threshold (fewer hubs = more savings)
+		// Only adjust if significantly faster (20% margin)
+		adjustment := calculateAdjustment(a.targetLatency, p95)
+		newThreshold = currentThreshold + adjustment
+
+		if newThreshold > a.maxThreshold {
+			newThreshold = a.maxThreshold
+		}
+
+		log.Info().
+			Dur("p95", p95).
+			Int("old_threshold", currentThreshold).
+			Int("new_threshold", newThreshold).
+			Msg("Auto-tuner: Raising hub threshold (room for savings)")
+
+	} else {
+		// Within acceptable range, no adjustment needed
+		log.Debug().
+			Dur("p95", p95).
+			Int("threshold", currentThreshold).
+			Msg("Auto-tuner: Performance acceptable, no adjustment")
+		return
+	}
+
+	// Apply adjustment
+	if newThreshold != currentThreshold {
+		a.client.hubThreshold = newThreshold
+		a.adjustments++
+
+		// Clear latency history after adjustment
+		a.latencies = make([]time.Duration, 0, 1000)
+
+		log.Info().
+			Int("threshold", newThreshold).
+			Int("total_adjustments", a.adjustments).
+			Msg("Hub threshold adjusted by auto-tuner")
+	}
+}
+
+// calculateP95 computes the 95th percentile latency
+func calculateP95(latencies []time.Duration) time.Duration {
+	if len(latencies) == 0 {
+		return 0
+	}
+
+	// Sort latencies
+	sorted := make([]time.Duration, len(latencies))
+	copy(sorted, latencies)
+
+	// Simple bubble sort (small dataset)
+	n := len(sorted)
+	for i := 0; i < n-1; i++ {
+		for j := 0; j < n-i-1; j++ {
+			if sorted[j] > sorted[j+1] {
+				sorted[j], sorted[j+1] = sorted[j+1], sorted[j]
+			}
+		}
+	}
+
+	// Return 95th percentile
+	idx := int(float64(len(sorted)) * 0.95)
+	if idx >= len(sorted) {
+		idx = len(sorted) - 1
+	}
+
+	return sorted[idx]
+}
+
+// calculateAdjustment determines how much to adjust threshold
+func calculateAdjustment(actual, target time.Duration) int {
+	// Calculate percentage difference
+	diff := float64(actual-target) / float64(target)
+
+	// Adjust more aggressively for larger differences
+	if diff > 0.5 || diff < -0.5 {
+		return 3 // Large adjustment
+	} else if diff > 0.2 || diff < -0.2 {
+		return 2 // Medium adjustment
+	}
+
+	return 1 // Small adjustment
+}
+
+// GetStats returns auto-tuner statistics
+func (a *AutoTuner) GetStats() AutoTunerStats {
+	a.latenciesMu.Lock()
+	defer a.latenciesMu.Unlock()
+
+	stats := AutoTunerStats{
+		CurrentThreshold: a.client.hubThreshold,
+		TargetLatency:    a.targetLatency,
+		TotalQueries:     a.queries,
+		TotalAdjustments: a.adjustments,
+		RecentQueries:    len(a.latencies),
+	}
+
+	if len(a.latencies) > 0 {
+		stats.P95Latency = calculateP95(a.latencies)
+
+		// Calculate average
+		var total time.Duration
+		for _, lat := range a.latencies {
+			total += lat
+		}
+		stats.AvgLatency = total / time.Duration(len(a.latencies))
+	}
+
+	return stats
+}
+
+// AutoTunerStats contains auto-tuner statistics
+type AutoTunerStats struct {
+	CurrentThreshold int
+	TargetLatency    time.Duration
+	P95Latency       time.Duration
+	AvgLatency       time.Duration
+	TotalQueries     int64
+	TotalAdjustments int
+	RecentQueries    int
+}
+
+// AutoTunedClient wraps Client with automatic performance tuning
+type AutoTunedClient struct {
+	*Client
+	tuner *AutoTuner
+}
+
+// Query wraps the underlying Query call with latency tracking
+func (a *AutoTunedClient) Query(ctx context.Context, query string, limit int, where map[string]any) ([]sqlitevec.QueryResult, error) {
+	start := time.Now()
+	results, err := a.Client.Query(ctx, query, limit, where)
+	latency := time.Since(start)
+
+	a.tuner.RecordQuery(latency)
+
+	return results, err
+}
+
+// WithAutoTuning wraps a hybrid client with auto-tuning enabled
+func WithAutoTuning(client *Client, cfg AutoTunerConfig) *AutoTunedClient {
+	tuner := NewAutoTuner(client, cfg)
+	tuner.Start()
+
+	return &AutoTunedClient{
+		Client: client,
+		tuner:  tuner,
+	}
+}
+
+// Stop stops the auto-tuner
+func (a *AutoTunedClient) StopTuning() {
+	a.tuner.Stop()
+}

internal/vector/hybrid/client.go

@@ -0,0 +1,515 @@
+// Package hybrid provides LEANN-inspired selective vector storage for claude-mnemonic.
+//
+// This package implements a hybrid storage strategy where frequently-accessed
+// observations ("hubs") have their embeddings stored, while infrequently-accessed
+// observations have their embeddings recomputed on-demand during search.
+//
+// This approach reduces storage by 60-80% with minimal impact on search latency (<50ms).
+package hybrid
+
+import (
+	"context"
+	"database/sql"
+	"fmt"
+	"math"
+	"sync"
+	"time"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/embedding"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/vector/sqlitevec"
+	"github.com/rs/zerolog/log"
+)
+
+// VectorStorageStrategy defines how embeddings are stored/computed
+type VectorStorageStrategy int
+
+const (
+	// StorageAlways stores all embeddings (current behavior, backwards compatible)
+	StorageAlways VectorStorageStrategy = iota
+	// StorageHub stores only frequently-accessed "hub" embeddings (recommended)
+	StorageHub
+	// StorageOnDemand recomputes all embeddings during search (maximum savings)
+	StorageOnDemand
+)
+
+// Client wraps sqlitevec.Client with selective storage logic
+type Client struct {
+	base         *sqlitevec.Client
+	db           *sql.DB
+	embedSvc     *embedding.Service
+	accessCount  map[string]int
+	lastAccess   map[string]time.Time
+	contentCache map[string]string
+	strategy     VectorStorageStrategy
+	hubThreshold int
+	mu           sync.RWMutex
+	cacheMu      sync.RWMutex
+}
+
+// Config for hybrid client
+type Config struct {
+	BaseClient   *sqlitevec.Client
+	DB           *sql.DB
+	EmbedSvc     *embedding.Service
+	Strategy     VectorStorageStrategy
+	HubThreshold int // Default: 5 accesses
+}
+
+// NewClient creates a new hybrid vector client
+func NewClient(cfg Config) *Client {
+	if cfg.HubThreshold <= 0 {
+		cfg.HubThreshold = 5
+	}
+
+	log.Info().
+		Str("strategy", strategyToString(cfg.Strategy)).
+		Int("hub_threshold", cfg.HubThreshold).
+		Msg("Initializing LEANN hybrid vector client")
+
+	return &Client{
+		base:         cfg.BaseClient,
+		db:           cfg.DB,
+		embedSvc:     cfg.EmbedSvc,
+		strategy:     cfg.Strategy,
+		hubThreshold: cfg.HubThreshold,
+		accessCount:  make(map[string]int),
+		lastAccess:   make(map[string]time.Time),
+		contentCache: make(map[string]string),
+	}
+}
+
+// AddDocuments implements selective storage based on strategy
+func (c *Client) AddDocuments(ctx context.Context, docs []sqlitevec.Document) error {
+	if len(docs) == 0 {
+		return nil
+	}
+
+	switch c.strategy {
+	case StorageAlways:
+		// Use existing implementation - store all embeddings
+		return c.base.AddDocuments(ctx, docs)
+
+	case StorageHub:
+		// Store only hub candidates
+		return c.addDocumentsSelective(ctx, docs)
+
+	case StorageOnDemand:
+		// Don't store embeddings, only cache content
+		return c.cacheDocuments(ctx, docs)
+
+	default:
+		return c.base.AddDocuments(ctx, docs)
+	}
+}
+
+// addDocumentsSelective stores embeddings only for hub-qualified documents
+func (c *Client) addDocumentsSelective(ctx context.Context, docs []sqlitevec.Document) error {
+	// Always cache content for potential recomputation
+	if err := c.cacheDocuments(ctx, docs); err != nil {
+		return err
+	}
+
+	// Filter to hub documents
+	hubDocs := make([]sqlitevec.Document, 0, len(docs))
+	for _, doc := range docs {
+		if c.isHub(doc.ID) {
+			hubDocs = append(hubDocs, doc)
+		}
+	}
+
+	// Store only hub embeddings
+	if len(hubDocs) > 0 {
+		log.Debug().
+			Int("total", len(docs)).
+			Int("hubs", len(hubDocs)).
+			Msg("Storing selective embeddings")
+		return c.base.AddDocuments(ctx, hubDocs)
+	}
+
+	log.Debug().Int("total", len(docs)).Msg("All documents cached, no hubs to store")
+	return nil
+}
+
+// cacheDocuments stores content for later recomputation
+func (c *Client) cacheDocuments(ctx context.Context, docs []sqlitevec.Document) error {
+	c.cacheMu.Lock()
+	defer c.cacheMu.Unlock()
+
+	for _, doc := range docs {
+		c.contentCache[doc.ID] = doc.Content
+	}
+
+	return nil
+}
+
+// DeleteDocuments removes documents by their IDs
+func (c *Client) DeleteDocuments(ctx context.Context, ids []string) error {
+	// Remove from base storage
+	if err := c.base.DeleteDocuments(ctx, ids); err != nil {
+		return err
+	}
+
+	// Clean up caches
+	c.mu.Lock()
+	for _, id := range ids {
+		delete(c.accessCount, id)
+		delete(c.lastAccess, id)
+	}
+	c.mu.Unlock()
+
+	c.cacheMu.Lock()
+	for _, id := range ids {
+		delete(c.contentCache, id)
+	}
+	c.cacheMu.Unlock()
+
+	return nil
+}
+
+// Query performs search with dynamic recomputation
+func (c *Client) Query(ctx context.Context, query string, limit int, where map[string]any) ([]sqlitevec.QueryResult, error) {
+	switch c.strategy {
+	case StorageAlways:
+		// Use existing implementation
+		return c.queryAndTrack(ctx, query, limit, where)
+
+	case StorageHub:
+		// Search hubs, then expand with recomputation
+		return c.queryHybrid(ctx, query, limit, where)
+
+	case StorageOnDemand:
+		// Fully dynamic search
+		return c.queryDynamic(ctx, query, limit, where)
+
+	default:
+		return c.queryAndTrack(ctx, query, limit, where)
+	}
+}
+
+// queryAndTrack wraps base Query with access tracking
+func (c *Client) queryAndTrack(ctx context.Context, query string, limit int, where map[string]any) ([]sqlitevec.QueryResult, error) {
+	results, err := c.base.Query(ctx, query, limit, where)
+	if err != nil {
+		return nil, err
+	}
+
+	// Track access for hub detection
+	c.trackAccess(results)
+
+	return results, nil
+}
+
+// queryHybrid searches stored hubs and recomputes non-hubs
+func (c *Client) queryHybrid(ctx context.Context, query string, limit int, where map[string]any) ([]sqlitevec.QueryResult, error) {
+	startTime := time.Now()
+
+	// 1. Query stored hub embeddings (limit * 2 for expansion)
+	hubResults, err := c.base.Query(ctx, query, limit*2, where)
+	if err != nil {
+		return nil, err
+	}
+
+	// 2. Track access
+	c.trackAccess(hubResults)
+
+	// 3. Get candidate non-hub IDs (from content cache)
+	candidates := c.getCandidateNonHubs(where, limit*2)
+
+	// 4. Recompute embeddings for candidates if we have any
+	var recomputedResults []sqlitevec.QueryResult
+	if len(candidates) > 0 {
+		recomputedResults, err = c.recomputeAndScore(ctx, query, candidates)
+		if err != nil {
+			// Log but don't fail - use hub results only
+			log.Warn().Err(err).Msg("Failed to recompute embeddings, using hub results only")
+			recomputedResults = nil
+		}
+	}
+
+	// 5. Merge and rank
+	allResults := append(hubResults, recomputedResults...)
+	sortBySimilarity(allResults)
+
+	// 6. Return top K
+	if len(allResults) > limit {
+		allResults = allResults[:limit]
+	}
+
+	duration := time.Since(startTime)
+	log.Debug().
+		Dur("duration_ms", duration).
+		Int("hubs", len(hubResults)).
+		Int("recomputed", len(recomputedResults)).
+		Int("results", len(allResults)).
+		Msg("Hybrid search completed")
+
+	return allResults, nil
+}
+
+// queryDynamic recomputes all embeddings on-the-fly
+func (c *Client) queryDynamic(ctx context.Context, query string, limit int, where map[string]any) ([]sqlitevec.QueryResult, error) {
+	startTime := time.Now()
+
+	// Get all candidate IDs from content cache
+	candidates := c.getCandidateNonHubs(where, limit*5)
+
+	// Recompute and score all
+	results, err := c.recomputeAndScore(ctx, query, candidates)
+	if err != nil {
+		return nil, err
+	}
+
+	// Track access
+	c.trackAccess(results)
+
+	// Return top K
+	if len(results) > limit {
+		results = results[:limit]
+	}
+
+	duration := time.Since(startTime)
+	log.Debug().
+		Dur("duration_ms", duration).
+		Int("recomputed", len(candidates)).
+		Int("results", len(results)).
+		Msg("Dynamic search completed")
+
+	return results, nil
+}
+
+// recomputeAndScore generates embeddings and computes similarities
+func (c *Client) recomputeAndScore(ctx context.Context, query string, candidateIDs []string) ([]sqlitevec.QueryResult, error) {
+	if len(candidateIDs) == 0 {
+		return nil, nil
+	}
+
+	// Generate query embedding
+	queryEmb, err := c.embedSvc.Embed(query)
+	if err != nil {
+		return nil, fmt.Errorf("embed query: %w", err)
+	}
+
+	// Get content for candidates
+	c.cacheMu.RLock()
+	texts := make([]string, 0, len(candidateIDs))
+	validIDs := make([]string, 0, len(candidateIDs))
+	for _, id := range candidateIDs {
+		if content, ok := c.contentCache[id]; ok && content != "" {
+			texts = append(texts, content)
+			validIDs = append(validIDs, id)
+		}
+	}
+	c.cacheMu.RUnlock()
+
+	if len(texts) == 0 {
+		return nil, nil
+	}
+
+	// Batch generate embeddings
+	embeddings, err := c.embedSvc.EmbedBatch(texts)
+	if err != nil {
+		return nil, fmt.Errorf("batch embed: %w", err)
+	}
+
+	// Compute similarities
+	results := make([]sqlitevec.QueryResult, len(embeddings))
+	for i, emb := range embeddings {
+		similarity := cosineSimilarity(queryEmb, emb)
+		distance := 1.0 - similarity // Convert to distance
+
+		results[i] = sqlitevec.QueryResult{
+			ID:         validIDs[i],
+			Distance:   float64(distance),
+			Similarity: float64(similarity),
+			Metadata:   make(map[string]any),
+		}
+	}
+
+	return results, nil
+}
+
+// trackAccess records document access for hub detection
+func (c *Client) trackAccess(results []sqlitevec.QueryResult) {
+	if len(results) == 0 {
+		return
+	}
+
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	now := time.Now()
+	for _, r := range results {
+		c.accessCount[r.ID]++
+		c.lastAccess[r.ID] = now
+	}
+}
+
+// isHub checks if a document qualifies as a hub
+func (c *Client) isHub(docID string) bool {
+	c.mu.RLock()
+	defer c.mu.RUnlock()
+
+	count := c.accessCount[docID]
+	return count >= c.hubThreshold
+}
+
+// getCandidateNonHubs returns IDs of non-hub documents matching filter
+func (c *Client) getCandidateNonHubs(where map[string]any, limit int) []string {
+	c.cacheMu.RLock()
+	defer c.cacheMu.RUnlock()
+
+	candidates := make([]string, 0, limit)
+	for id := range c.contentCache {
+		if !c.isHub(id) {
+			candidates = append(candidates, id)
+			if len(candidates) >= limit {
+				break
+			}
+		}
+	}
+
+	return candidates
+}
+
+// IsConnected always returns true (wraps base client)
+func (c *Client) IsConnected() bool {
+	return c.base.IsConnected()
+}
+
+// Close releases resources
+func (c *Client) Close() error {
+	return c.base.Close()
+}
+
+// Count returns the total number of vectors in the store
+func (c *Client) Count(ctx context.Context) (int64, error) {
+	return c.base.Count(ctx)
+}
+
+// ModelVersion returns the current embedding model version
+func (c *Client) ModelVersion() string {
+	return c.base.ModelVersion()
+}
+
+// NeedsRebuild checks if vectors need to be rebuilt due to model version change
+func (c *Client) NeedsRebuild(ctx context.Context) (bool, string) {
+	return c.base.NeedsRebuild(ctx)
+}
+
+// GetStaleVectors returns doc_ids of vectors with mismatched or null model versions
+func (c *Client) GetStaleVectors(ctx context.Context) ([]sqlitevec.StaleVectorInfo, error) {
+	return c.base.GetStaleVectors(ctx)
+}
+
+// DeleteVectorsByDocIDs removes vectors by their doc_ids
+func (c *Client) DeleteVectorsByDocIDs(ctx context.Context, docIDs []string) error {
+	return c.base.DeleteVectorsByDocIDs(ctx, docIDs)
+}
+
+// GetStorageStats returns storage efficiency metrics
+func (c *Client) GetStorageStats(ctx context.Context) (StorageStats, error) {
+	c.mu.RLock()
+	c.cacheMu.RLock()
+	defer c.mu.RUnlock()
+	defer c.cacheMu.RUnlock()
+
+	totalDocs := len(c.contentCache)
+	hubCount := 0
+	for id := range c.contentCache {
+		if c.accessCount[id] >= c.hubThreshold {
+			hubCount++
+		}
+	}
+
+	storedCount := hubCount
+	if c.strategy == StorageAlways {
+		// Get actual count from database
+		if count, err := c.base.Count(ctx); err == nil {
+			storedCount = int(count)
+		}
+	} else if c.strategy == StorageOnDemand {
+		storedCount = 0
+	}
+
+	embeddingSize := 384 * 4 // 384 dims × 4 bytes (float32)
+	storedBytes := storedCount * embeddingSize
+	potentialBytes := totalDocs * embeddingSize
+
+	savingsPercent := 0.0
+	if potentialBytes > 0 {
+		savingsPercent = (1.0 - float64(storedBytes)/float64(potentialBytes)) * 100
+	}
+
+	return StorageStats{
+		TotalDocuments:   totalDocs,
+		HubDocuments:     hubCount,
+		StoredEmbeddings: storedCount,
+		StorageBytes:     storedBytes,
+		SavingsPercent:   savingsPercent,
+		Strategy:         c.strategy,
+	}, nil
+}
+
+// StorageStats contains storage efficiency metrics
+type StorageStats struct {
+	TotalDocuments   int
+	HubDocuments     int
+	StoredEmbeddings int
+	StorageBytes     int
+	SavingsPercent   float64
+	Strategy         VectorStorageStrategy
+}
+
+// Helper functions
+
+func cosineSimilarity(a, b []float32) float32 {
+	var dotProduct, normA, normB float32
+	for i := range a {
+		dotProduct += a[i] * b[i]
+		normA += a[i] * a[i]
+		normB += b[i] * b[i]
+	}
+	if normA == 0 || normB == 0 {
+		return 0
+	}
+	return dotProduct / float32(math.Sqrt(float64(normA))*math.Sqrt(float64(normB)))
+}
+
+func sortBySimilarity(results []sqlitevec.QueryResult) {
+	// Use a simple but efficient sorting algorithm
+	n := len(results)
+	for i := 0; i < n-1; i++ {
+		for j := 0; j < n-i-1; j++ {
+			if results[j].Similarity < results[j+1].Similarity {
+				results[j], results[j+1] = results[j+1], results[j]
+			}
+		}
+	}
+}
+
+func strategyToString(s VectorStorageStrategy) string {
+	switch s {
+	case StorageAlways:
+		return "always"
+	case StorageHub:
+		return "hub"
+	case StorageOnDemand:
+		return "on_demand"
+	default:
+		return "unknown"
+	}
+}
+
+// ParseStrategy converts a string to VectorStorageStrategy
+func ParseStrategy(s string) VectorStorageStrategy {
+	switch s {
+	case "hub":
+		return StorageHub
+	case "on_demand":
+		return StorageOnDemand
+	case "always":
+		return StorageAlways
+	default:
+		return StorageHub // Default to hub strategy
+	}
+}

internal/vector/hybrid/client_test.go

@@ -0,0 +1,186 @@
+package hybrid
+
+import (
+	"testing"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/vector/sqlitevec"
+	"github.com/stretchr/testify/assert"
+)
+
+func TestParseStrategy(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    string
+		expected VectorStorageStrategy
+	}{
+		{"hub_strategy", "hub", StorageHub},
+		{"on_demand_strategy", "on_demand", StorageOnDemand},
+		{"always_strategy", "always", StorageAlways},
+		{"invalid_defaults_to_hub", "invalid", StorageHub},
+		{"empty_defaults_to_hub", "", StorageHub},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := ParseStrategy(tt.input)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+func TestStrategyToString(t *testing.T) {
+	tests := []struct {
+		name     string
+		expected string
+		input    VectorStorageStrategy
+	}{
+		{"hub_to_string", "hub", StorageHub},
+		{"on_demand_to_string", "on_demand", StorageOnDemand},
+		{"always_to_string", "always", StorageAlways},
+		{"invalid_to_unknown", "unknown", VectorStorageStrategy(99)},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := strategyToString(tt.input)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
+func TestCosineSimilarity(t *testing.T) {
+	tests := []struct {
+		name     string
+		a        []float32
+		b        []float32
+		expected float32
+	}{
+		{
+			name:     "identical_vectors",
+			a:        []float32{1, 0, 0},
+			b:        []float32{1, 0, 0},
+			expected: 1.0,
+		},
+		{
+			name:     "orthogonal_vectors",
+			a:        []float32{1, 0, 0},
+			b:        []float32{0, 1, 0},
+			expected: 0.0,
+		},
+		{
+			name:     "opposite_vectors",
+			a:        []float32{1, 0, 0},
+			b:        []float32{-1, 0, 0},
+			expected: -1.0,
+		},
+		{
+			name:     "zero_vector",
+			a:        []float32{0, 0, 0},
+			b:        []float32{1, 1, 1},
+			expected: 0.0,
+		},
+		{
+			name:     "parallel_vectors",
+			a:        []float32{2, 0, 0},
+			b:        []float32{4, 0, 0},
+			expected: 1.0,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := cosineSimilarity(tt.a, tt.b)
+			assert.InDelta(t, tt.expected, result, 0.001)
+		})
+	}
+}
+
+func TestSortBySimilarity(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    []sqlitevec.QueryResult
+		expected []string // Expected order of IDs
+	}{
+		{
+			name: "already_sorted",
+			input: []sqlitevec.QueryResult{
+				{ID: "doc1", Similarity: 0.9},
+				{ID: "doc2", Similarity: 0.7},
+				{ID: "doc3", Similarity: 0.5},
+			},
+			expected: []string{"doc1", "doc2", "doc3"},
+		},
+		{
+			name: "reverse_sorted",
+			input: []sqlitevec.QueryResult{
+				{ID: "doc1", Similarity: 0.3},
+				{ID: "doc2", Similarity: 0.7},
+				{ID: "doc3", Similarity: 0.9},
+			},
+			expected: []string{"doc3", "doc2", "doc1"},
+		},
+		{
+			name: "random_order",
+			input: []sqlitevec.QueryResult{
+				{ID: "doc1", Similarity: 0.5},
+				{ID: "doc2", Similarity: 0.9},
+				{ID: "doc3", Similarity: 0.3},
+				{ID: "doc4", Similarity: 0.7},
+			},
+			expected: []string{"doc2", "doc4", "doc1", "doc3"},
+		},
+		{
+			name: "identical_similarities",
+			input: []sqlitevec.QueryResult{
+				{ID: "doc1", Similarity: 0.5},
+				{ID: "doc2", Similarity: 0.5},
+				{ID: "doc3", Similarity: 0.5},
+			},
+			expected: []string{"doc1", "doc2", "doc3"},
+		},
+		{
+			name:     "empty_list",
+			input:    []sqlitevec.QueryResult{},
+			expected: []string{},
+		},
+		{
+			name: "single_element",
+			input: []sqlitevec.QueryResult{
+				{ID: "doc1", Similarity: 0.5},
+			},
+			expected: []string{"doc1"},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			sortBySimilarity(tt.input)
+
+			actual := make([]string, len(tt.input))
+			for i, r := range tt.input {
+				actual[i] = r.ID
+			}
+
+			assert.Equal(t, tt.expected, actual)
+		})
+	}
+}
+
+func TestSortBySimilarity_PreserveOtherFields(t *testing.T) {
+	input := []sqlitevec.QueryResult{
+		{ID: "doc1", Similarity: 0.3, Distance: 0.7, Metadata: map[string]any{"key": "val1"}},
+		{ID: "doc2", Similarity: 0.9, Distance: 0.1, Metadata: map[string]any{"key": "val2"}},
+	}
+
+	sortBySimilarity(input)
+
+	assert.Equal(t, "doc2", input[0].ID)
+	assert.InDelta(t, 0.9, input[0].Similarity, 0.001)
+	assert.InDelta(t, 0.1, input[0].Distance, 0.001)
+	assert.Equal(t, "val2", input[0].Metadata["key"])
+
+	assert.Equal(t, "doc1", input[1].ID)
+	assert.InDelta(t, 0.3, input[1].Similarity, 0.001)
+	assert.InDelta(t, 0.7, input[1].Distance, 0.001)
+	assert.Equal(t, "val1", input[1].Metadata["key"])
+}

internal/vector/hybrid/config.go

@@ -0,0 +1,62 @@
+package hybrid
+
+import (
+	"os"
+	"strconv"
+
+	"github.com/rs/zerolog/log"
+)
+
+// GetStrategyFromEnv reads CLAUDE_MNEMONIC_VECTOR_STRATEGY from environment
+func GetStrategyFromEnv() VectorStorageStrategy {
+	strategyStr := os.Getenv("CLAUDE_MNEMONIC_VECTOR_STRATEGY")
+	if strategyStr == "" {
+		// Default to hub strategy for optimal balance
+		return StorageHub
+	}
+
+	strategy := ParseStrategy(strategyStr)
+	log.Info().
+		Str("env_value", strategyStr).
+		Str("strategy", strategyToString(strategy)).
+		Msg("Vector storage strategy from environment")
+
+	return strategy
+}
+
+// GetHubThresholdFromEnv reads CLAUDE_MNEMONIC_HUB_THRESHOLD from environment
+func GetHubThresholdFromEnv() int {
+	thresholdStr := os.Getenv("CLAUDE_MNEMONIC_HUB_THRESHOLD")
+	if thresholdStr == "" {
+		return 5 // Default threshold
+	}
+
+	threshold, err := strconv.Atoi(thresholdStr)
+	if err != nil {
+		log.Warn().
+			Err(err).
+			Str("env_value", thresholdStr).
+			Msg("Invalid hub threshold in environment, using default")
+		return 5
+	}
+
+	if threshold < 1 {
+		log.Warn().
+			Int("env_value", threshold).
+			Msg("Hub threshold too low, using minimum of 1")
+		return 1
+	}
+
+	log.Info().
+		Int("threshold", threshold).
+		Msg("Hub threshold from environment")
+
+	return threshold
+}
+
+// IsHybridEnabled checks if hybrid storage should be used
+// Returns false if CLAUDE_MNEMONIC_VECTOR_STRATEGY=always (backwards compat)
+func IsHybridEnabled() bool {
+	strategy := GetStrategyFromEnv()
+	return strategy != StorageAlways
+}

internal/vector/hybrid/graph_search.go

@@ -0,0 +1,308 @@
+package hybrid
+
+import (
+	"context"
+	"fmt"
+	"sort"
+	"time"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/graph"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/vector/sqlitevec"
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+	"github.com/rs/zerolog/log"
+)
+
+// GraphConfig configures graph-aware search
+type GraphConfig struct {
+	Enabled      bool
+	MaxHops      int     // Maximum graph traversal depth (default: 2)
+	BranchFactor int     // Number of neighbors to expand per node (default: 5)
+	EdgeWeight   float64 // Minimum edge weight to follow (default: 0.3)
+}
+
+// DefaultGraphConfig returns sensible defaults for graph search
+func DefaultGraphConfig() GraphConfig {
+	return GraphConfig{
+		Enabled:      true,
+		MaxHops:      2,
+		BranchFactor: 5,
+		EdgeWeight:   0.3,
+	}
+}
+
+// GraphSearchClient wraps hybrid.Client with graph-aware search
+type GraphSearchClient struct {
+	*Client
+	graph       *graph.ObservationGraph
+	graphConfig GraphConfig
+}
+
+// NewGraphSearchClient creates a graph-enhanced hybrid client
+func NewGraphSearchClient(baseClient *Client, observationGraph *graph.ObservationGraph, cfg GraphConfig) *GraphSearchClient {
+	return &GraphSearchClient{
+		Client:      baseClient,
+		graph:       observationGraph,
+		graphConfig: cfg,
+	}
+}
+
+// Query performs graph-aware vector search with two-level traversal
+func (g *GraphSearchClient) Query(ctx context.Context, query string, limit int, where map[string]any) ([]sqlitevec.QueryResult, error) {
+	if !g.graphConfig.Enabled || g.graph == nil {
+		// Fall back to standard hybrid search
+		return g.Client.Query(ctx, query, limit, where)
+	}
+
+	startTime := time.Now()
+
+	// 1. Generate query embedding
+	queryEmb, err := g.embedSvc.Embed(query)
+	if err != nil {
+		return nil, fmt.Errorf("embed query: %w", err)
+	}
+
+	// 2. Search hub nodes (stored embeddings)
+	hubResults, err := g.base.Query(ctx, query, limit*2, where)
+	if err != nil {
+		// Fall back to standard search on error
+		log.Warn().Err(err).Msg("Hub search failed, falling back to hybrid search")
+		return g.Client.Query(ctx, query, limit, where)
+	}
+
+	// 3. Track hub access
+	g.trackAccess(hubResults)
+
+	// 4. Expand via graph traversal
+	expandedIDs := g.expandFromHubs(hubResults, limit*4)
+
+	// 5. Filter to non-hubs that need recomputation
+	nonHubIDs := make([]string, 0)
+	for _, id := range expandedIDs {
+		if !g.isHub(id) {
+			nonHubIDs = append(nonHubIDs, id)
+		}
+	}
+
+	// 6. Batch recompute non-hub embeddings
+	recomputedResults, err := g.recomputeAndScore(ctx, query, nonHubIDs)
+	if err != nil {
+		log.Warn().Err(err).Msg("Recomputation failed, using hub results only")
+		recomputedResults = nil
+	}
+
+	// 7. Apply graph-based ranking boost
+	allResults := g.mergeAndRankWithGraph(hubResults, recomputedResults, queryEmb)
+
+	// 8. Return top K
+	if len(allResults) > limit {
+		allResults = allResults[:limit]
+	}
+
+	duration := time.Since(startTime)
+	log.Debug().
+		Dur("duration_ms", duration).
+		Int("hubs", len(hubResults)).
+		Int("expanded", len(expandedIDs)).
+		Int("recomputed", len(recomputedResults)).
+		Int("results", len(allResults)).
+		Msg("Graph search completed")
+
+	return allResults, nil
+}
+
+// expandFromHubs traverses graph from hub nodes to find promising candidates
+func (g *GraphSearchClient) expandFromHubs(hubResults []sqlitevec.QueryResult, maxCandidates int) []string {
+	if g.graph == nil {
+		return nil
+	}
+
+	expanded := make(map[string]float64) // doc_id -> relevance score
+	visited := make(map[int64]bool)
+
+	// Start from top hub results
+	for i, result := range hubResults {
+		if i >= g.graphConfig.BranchFactor*2 {
+			break // Limit starting points
+		}
+
+		// Parse observation ID from doc_id
+		obsID := parseObservationID(result.ID)
+		if obsID == 0 {
+			continue
+		}
+
+		// Mark as visited with high relevance (direct match)
+		visited[obsID] = true
+		expanded[result.ID] = result.Similarity
+
+		// Traverse graph from this hub
+		g.traverseGraph(obsID, result.Similarity, 0, expanded, visited)
+	}
+
+	// Convert to sorted list
+	type candidate struct {
+		ID        string
+		Relevance float64
+	}
+
+	candidates := make([]candidate, 0, len(expanded))
+	for id, rel := range expanded {
+		candidates = append(candidates, candidate{ID: id, Relevance: rel})
+	}
+
+	// Sort by relevance descending
+	sort.Slice(candidates, func(i, j int) bool {
+		return candidates[i].Relevance > candidates[j].Relevance
+	})
+
+	// Return top candidates
+	if len(candidates) > maxCandidates {
+		candidates = candidates[:maxCandidates]
+	}
+
+	result := make([]string, len(candidates))
+	for i, c := range candidates {
+		result[i] = c.ID
+	}
+
+	return result
+}
+
+// traverseGraph performs depth-limited graph traversal
+func (g *GraphSearchClient) traverseGraph(nodeID int64, baseRelevance float64, depth int, expanded map[string]float64, visited map[int64]bool) {
+	if depth >= g.graphConfig.MaxHops {
+		return // Max depth reached
+	}
+
+	// Get neighbors from graph
+	neighbors, weights, err := g.graph.GetNeighbors(nodeID)
+	if err != nil {
+		return // No neighbors or error
+	}
+
+	// Traverse top neighbors by weight
+	type neighborWeight struct {
+		ID     int64
+		Weight float32
+	}
+
+	neighborList := make([]neighborWeight, len(neighbors))
+	for i := range neighbors {
+		neighborList[i] = neighborWeight{
+			ID:     neighbors[i],
+			Weight: weights[i],
+		}
+	}
+
+	// Sort by weight descending
+	sort.Slice(neighborList, func(i, j int) bool {
+		return neighborList[i].Weight > neighborList[j].Weight
+	})
+
+	// Expand top branch_factor neighbors
+	expanded_count := 0
+	for _, nw := range neighborList {
+		if expanded_count >= g.graphConfig.BranchFactor {
+			break
+		}
+
+		// Skip if edge weight too low
+		if float64(nw.Weight) < g.graphConfig.EdgeWeight {
+			continue
+		}
+
+		// Skip if already visited
+		if visited[nw.ID] {
+			continue
+		}
+		visited[nw.ID] = true
+
+		// Calculate propagated relevance (decays with distance)
+		decay := 0.7 // 30% decay per hop
+		propagatedRelevance := baseRelevance * float64(nw.Weight) * decay
+
+		// Add to expanded set
+		docID := formatObservationDocID(nw.ID)
+		if existing, ok := expanded[docID]; !ok || propagatedRelevance > existing {
+			expanded[docID] = propagatedRelevance
+		}
+
+		// Recursively traverse
+		g.traverseGraph(nw.ID, propagatedRelevance, depth+1, expanded, visited)
+		expanded_count++
+	}
+}
+
+// mergeAndRankWithGraph combines hub and recomputed results with graph-based ranking
+func (g *GraphSearchClient) mergeAndRankWithGraph(hubResults, recomputedResults []sqlitevec.QueryResult, queryEmb []float32) []sqlitevec.QueryResult {
+	// Merge results
+	allResults := append(hubResults, recomputedResults...)
+
+	// Apply graph-based re-ranking
+	if g.graph != nil {
+		for i := range allResults {
+			obsID := parseObservationID(allResults[i].ID)
+			if obsID == 0 {
+				continue
+			}
+
+			// Boost score based on node degree (hubs are more important)
+			node, err := g.graph.GetNode(obsID)
+			if err == nil && node.Degree > 0 {
+				// Degree boost: up to 10% increase for high-degree nodes
+				degreeBoost := 1.0 + (0.1 * float64(node.Degree) / 20.0)
+				if degreeBoost > 1.1 {
+					degreeBoost = 1.1
+				}
+				allResults[i].Similarity *= degreeBoost
+			}
+		}
+	}
+
+	// Sort by adjusted similarity
+	sortBySimilarity(allResults)
+
+	return allResults
+}
+
+// parseObservationID extracts observation ID from doc_id
+// Format: "obs-{id}-{field}"
+func parseObservationID(docID string) int64 {
+	var obsID int64
+	// Ignore error - returns 0 on parse failure, which callers handle
+	_, _ = fmt.Sscanf(docID, "obs-%d-", &obsID)
+	return obsID
+}
+
+// formatObservationDocID creates a doc_id for an observation
+func formatObservationDocID(obsID int64) string {
+	return fmt.Sprintf("obs-%d-combined", obsID)
+}
+
+// GetGraphStats returns statistics about the observation graph
+func (g *GraphSearchClient) GetGraphStats() graph.GraphStats {
+	if g.graph == nil {
+		return graph.GraphStats{}
+	}
+	return g.graph.Stats()
+}
+
+// RebuildGraph rebuilds the observation graph from current observations
+// This should be called periodically or when observations change significantly
+func (g *GraphSearchClient) RebuildGraph(ctx context.Context, observations []*models.Observation) error {
+	log.Info().Int("observations", len(observations)).Msg("Rebuilding observation graph")
+
+	newGraph, err := graph.BuildFromObservations(ctx, observations)
+	if err != nil {
+		return fmt.Errorf("build graph: %w", err)
+	}
+
+	g.graph = newGraph
+
+	log.Info().
+		Int("nodes", newGraph.Stats().NodeCount).
+		Int("edges", newGraph.Stats().EdgeCount).
+		Msg("Graph rebuilt successfully")
+
+	return nil
+}

internal/vector/hybrid/interface_test.go

@@ -0,0 +1,16 @@
+package hybrid
+
+import (
+	"testing"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/vector"
+)
+
+// TestInterfaceImplementation verifies that hybrid clients implement vector.Client interface
+func TestInterfaceImplementation(t *testing.T) {
+	// Compile-time check that Client implements vector.Client
+	var _ vector.Client = (*Client)(nil)
+
+	// Compile-time check that GraphSearchClient implements vector.Client
+	var _ vector.Client = (*GraphSearchClient)(nil)
+}

internal/vector/hybrid/metrics.go

@@ -0,0 +1,272 @@
+package hybrid
+
+import (
+	"fmt"
+	"sync"
+	"sync/atomic"
+	"time"
+)
+
+// Metrics tracks performance and usage statistics for hybrid vector storage
+type Metrics struct {
+	startTime         time.Time
+	recentLatencies   []time.Duration
+	latenciesMu       sync.Mutex
+	totalQueries      atomic.Int64
+	hubOnlyQueries    atomic.Int64
+	hybridQueries     atomic.Int64
+	onDemandQueries   atomic.Int64
+	graphQueries      atomic.Int64
+	totalLatency      atomic.Int64 // Sum in microseconds
+	hubLatency        atomic.Int64
+	recomputeLatency  atomic.Int64
+	totalDocuments    atomic.Int64
+	hubDocuments      atomic.Int64
+	storedEmbeddings  atomic.Int64
+	recomputedCount   atomic.Int64
+	cacheHits         atomic.Int64
+	cacheMisses       atomic.Int64
+	graphTraversals   atomic.Int64
+	avgTraversalDepth atomic.Int64
+}
+
+// NewMetrics creates a new metrics tracker
+func NewMetrics() *Metrics {
+	return &Metrics{
+		recentLatencies: make([]time.Duration, 0, 1000),
+		startTime:       time.Now(),
+	}
+}
+
+// RecordQuery records a query execution
+func (m *Metrics) RecordQuery(queryType string, latency time.Duration, recomputed int) {
+	m.totalQueries.Add(1)
+	m.totalLatency.Add(latency.Microseconds())
+
+	switch queryType {
+	case "hub_only":
+		m.hubOnlyQueries.Add(1)
+	case "hybrid":
+		m.hybridQueries.Add(1)
+	case "on_demand":
+		m.onDemandQueries.Add(1)
+	case "graph":
+		m.graphQueries.Add(1)
+	}
+
+	if recomputed > 0 {
+		m.recomputedCount.Add(int64(recomputed))
+	}
+
+	// Track recent latencies
+	m.latenciesMu.Lock()
+	m.recentLatencies = append(m.recentLatencies, latency)
+	if len(m.recentLatencies) > 1000 {
+		m.recentLatencies = m.recentLatencies[len(m.recentLatencies)-1000:]
+	}
+	m.latenciesMu.Unlock()
+}
+
+// RecordHubLatency records time spent in hub search
+func (m *Metrics) RecordHubLatency(latency time.Duration) {
+	m.hubLatency.Add(latency.Microseconds())
+}
+
+// RecordRecomputeLatency records time spent recomputing embeddings
+func (m *Metrics) RecordRecomputeLatency(latency time.Duration) {
+	m.recomputeLatency.Add(latency.Microseconds())
+}
+
+// RecordCacheHit records a content cache hit
+func (m *Metrics) RecordCacheHit() {
+	m.cacheHits.Add(1)
+}
+
+// RecordCacheMiss records a content cache miss
+func (m *Metrics) RecordCacheMiss() {
+	m.cacheMisses.Add(1)
+}
+
+// RecordGraphTraversal records a graph traversal operation
+func (m *Metrics) RecordGraphTraversal(depth int) {
+	m.graphTraversals.Add(1)
+	m.avgTraversalDepth.Add(int64(depth))
+}
+
+// UpdateStorageStats updates current storage statistics
+func (m *Metrics) UpdateStorageStats(total, hubs, stored int) {
+	m.totalDocuments.Store(int64(total))
+	m.hubDocuments.Store(int64(hubs))
+	m.storedEmbeddings.Store(int64(stored))
+}
+
+// GetSnapshot returns current metrics snapshot
+func (m *Metrics) GetSnapshot() MetricsSnapshot {
+	m.latenciesMu.Lock()
+	defer m.latenciesMu.Unlock()
+
+	totalQueries := m.totalQueries.Load()
+
+	snapshot := MetricsSnapshot{
+		// Query counts
+		TotalQueries:    totalQueries,
+		HubOnlyQueries:  m.hubOnlyQueries.Load(),
+		HybridQueries:   m.hybridQueries.Load(),
+		OnDemandQueries: m.onDemandQueries.Load(),
+		GraphQueries:    m.graphQueries.Load(),
+
+		// Storage
+		TotalDocuments:   int(m.totalDocuments.Load()),
+		HubDocuments:     int(m.hubDocuments.Load()),
+		StoredEmbeddings: int(m.storedEmbeddings.Load()),
+		RecomputedTotal:  m.recomputedCount.Load(),
+
+		// Cache
+		CacheHits:   m.cacheHits.Load(),
+		CacheMisses: m.cacheMisses.Load(),
+
+		// Graph
+		GraphTraversals: m.graphTraversals.Load(),
+
+		// Runtime
+		Uptime: time.Since(m.startTime),
+	}
+
+	// Calculate latencies
+	if totalQueries > 0 {
+		snapshot.AvgLatency = time.Duration(m.totalLatency.Load()/totalQueries) * time.Microsecond
+		snapshot.AvgHubLatency = time.Duration(m.hubLatency.Load()/totalQueries) * time.Microsecond
+	}
+
+	if m.recomputedCount.Load() > 0 {
+		snapshot.AvgRecomputeLatency = time.Duration(m.recomputeLatency.Load()/m.recomputedCount.Load()) * time.Microsecond
+	}
+
+	// Calculate percentiles
+	if len(m.recentLatencies) > 0 {
+		sorted := make([]time.Duration, len(m.recentLatencies))
+		copy(sorted, m.recentLatencies)
+		sortDurations(sorted)
+
+		snapshot.P50Latency = percentile(sorted, 0.50)
+		snapshot.P95Latency = percentile(sorted, 0.95)
+		snapshot.P99Latency = percentile(sorted, 0.99)
+	}
+
+	// Calculate cache hit rate
+	totalCacheOps := snapshot.CacheHits + snapshot.CacheMisses
+	if totalCacheOps > 0 {
+		snapshot.CacheHitRate = float64(snapshot.CacheHits) / float64(totalCacheOps)
+	}
+
+	// Calculate storage savings
+	if snapshot.TotalDocuments > 0 {
+		embeddingSize := 384 * 4 // 384 dims × 4 bytes
+		fullStorage := snapshot.TotalDocuments * embeddingSize
+		actualStorage := snapshot.StoredEmbeddings * embeddingSize
+
+		if fullStorage > 0 {
+			snapshot.StorageSavingsPercent = (1.0 - float64(actualStorage)/float64(fullStorage)) * 100
+		}
+	}
+
+	// Calculate avg traversal depth
+	if snapshot.GraphTraversals > 0 {
+		snapshot.AvgTraversalDepth = float64(m.avgTraversalDepth.Load()) / float64(snapshot.GraphTraversals)
+	}
+
+	return snapshot
+}
+
+// MetricsSnapshot represents a point-in-time metrics snapshot
+type MetricsSnapshot struct {
+	// Query metrics
+	TotalQueries    int64
+	HubOnlyQueries  int64
+	HybridQueries   int64
+	OnDemandQueries int64
+	GraphQueries    int64
+
+	// Latency metrics
+	AvgLatency          time.Duration
+	P50Latency          time.Duration
+	P95Latency          time.Duration
+	P99Latency          time.Duration
+	AvgHubLatency       time.Duration
+	AvgRecomputeLatency time.Duration
+
+	// Storage metrics
+	TotalDocuments        int
+	HubDocuments          int
+	StoredEmbeddings      int
+	StorageSavingsPercent float64
+	RecomputedTotal       int64
+
+	// Cache metrics
+	CacheHits    int64
+	CacheMisses  int64
+	CacheHitRate float64
+
+	// Graph metrics
+	GraphTraversals   int64
+	AvgTraversalDepth float64
+
+	// Runtime
+	Uptime time.Duration
+}
+
+// sortDurations sorts a slice of durations in ascending order
+func sortDurations(durations []time.Duration) {
+	n := len(durations)
+	for i := 0; i < n-1; i++ {
+		for j := 0; j < n-i-1; j++ {
+			if durations[j] > durations[j+1] {
+				durations[j], durations[j+1] = durations[j+1], durations[j]
+			}
+		}
+	}
+}
+
+// percentile calculates the Nth percentile from a sorted slice
+func percentile(sorted []time.Duration, p float64) time.Duration {
+	if len(sorted) == 0 {
+		return 0
+	}
+
+	idx := int(float64(len(sorted)) * p)
+	if idx >= len(sorted) {
+		idx = len(sorted) - 1
+	}
+
+	return sorted[idx]
+}
+
+// String returns a human-readable representation of metrics
+func (s MetricsSnapshot) String() string {
+	return fmt.Sprintf(`Hybrid Vector Storage Metrics:
+  Queries:
+    Total: %d (Hub: %d, Hybrid: %d, OnDemand: %d, Graph: %d)
+    Avg Latency: %v (p50: %v, p95: %v, p99: %v)
+    Hub Latency: %v, Recompute Latency: %v
+  Storage:
+    Documents: %d (Hubs: %d, %.1f%%)
+    Stored Embeddings: %d
+    Savings: %.1f%%
+    Total Recomputed: %d
+  Cache:
+    Hits: %d, Misses: %d (Hit Rate: %.1f%%)
+  Graph:
+    Traversals: %d (Avg Depth: %.2f)
+  Runtime: %v`,
+		s.TotalQueries, s.HubOnlyQueries, s.HybridQueries, s.OnDemandQueries, s.GraphQueries,
+		s.AvgLatency, s.P50Latency, s.P95Latency, s.P99Latency,
+		s.AvgHubLatency, s.AvgRecomputeLatency,
+		s.TotalDocuments, s.HubDocuments, float64(s.HubDocuments)/float64(s.TotalDocuments)*100,
+		s.StoredEmbeddings,
+		s.StorageSavingsPercent,
+		s.RecomputedTotal,
+		s.CacheHits, s.CacheMisses, s.CacheHitRate*100,
+		s.GraphTraversals, s.AvgTraversalDepth,
+		s.Uptime,
+	)
+}

internal/vector/interface.go

@@ -0,0 +1,42 @@
+// Package vector provides common interfaces for vector storage implementations
+package vector
+
+import (
+	"context"
+
+	"github.com/lukaszraczylo/claude-mnemonic/internal/vector/sqlitevec"
+)
+
+// Client defines the interface for vector storage operations.
+// Both sqlitevec.Client and hybrid.Client implement this interface.
+type Client interface {
+	// AddDocuments adds documents with their embeddings to the vector store
+	AddDocuments(ctx context.Context, docs []sqlitevec.Document) error
+
+	// DeleteDocuments removes documents by their IDs
+	DeleteDocuments(ctx context.Context, ids []string) error
+
+	// Query performs a vector similarity search
+	Query(ctx context.Context, query string, limit int, where map[string]any) ([]sqlitevec.QueryResult, error)
+
+	// IsConnected checks if the vector store is available
+	IsConnected() bool
+
+	// Close releases resources
+	Close() error
+
+	// Count returns the total number of vectors in the store
+	Count(ctx context.Context) (int64, error)
+
+	// ModelVersion returns the current embedding model version
+	ModelVersion() string
+
+	// NeedsRebuild checks if vectors need to be rebuilt due to model version change
+	NeedsRebuild(ctx context.Context) (bool, string)
+
+	// GetStaleVectors returns doc_ids of vectors with mismatched or null model versions
+	GetStaleVectors(ctx context.Context) ([]sqlitevec.StaleVectorInfo, error)
+
+	// DeleteVectorsByDocIDs removes vectors by their doc_ids
+	DeleteVectorsByDocIDs(ctx context.Context, docIDs []string) error
+}

internal/vector/sqlitevec/client.go

@@ -87,9 +87,9 @@ func (c *Client) AddDocuments(ctx context.Context, docs []Document) error {
 
 	for i, doc := range docs {
 		// Serialize embedding to blob format
-		embBlob, err := sqlite_vec.SerializeFloat32(embeddings[i])
-		if err != nil {
-			return fmt.Errorf("serialize embedding for %s: %w", doc.ID, err)
+		embBlob, serErr := sqlite_vec.SerializeFloat32(embeddings[i])
+		if serErr != nil {
+			return fmt.Errorf("serialize embedding for %s: %w", doc.ID, serErr)
 		}
 
 		// Extract metadata
@@ -212,10 +212,11 @@ func (c *Client) Query(ctx context.Context, query string, limit int, where map[s
 		var sqliteID int64
 		var docType, fieldType, project, scope sql.NullString
 
-		if err := rows.Scan(&r.ID, &r.Distance, &sqliteID, &docType, &fieldType, &project, &scope); err != nil {
-			return nil, fmt.Errorf("scan row: %w", err)
+		if scanErr := rows.Scan(&r.ID, &r.Distance, &sqliteID, &docType, &fieldType, &project, &scope); scanErr != nil {
+			return nil, fmt.Errorf("scan row: %w", scanErr)
 		}
 
+		r.Similarity = DistanceToSimilarity(r.Distance)
 		r.Metadata = map[string]any{
 			"sqlite_id":  float64(sqliteID), // Keep as float64 for compatibility
 			"doc_type":   docType.String,
@@ -318,11 +319,11 @@ func (c *Client) NeedsRebuild(ctx context.Context) (bool, string) {
 // StaleVectorInfo contains information about a vector that needs rebuilding.
 type StaleVectorInfo struct {
 	DocID     string
-	SQLiteID  int64
 	DocType   string
 	FieldType string
 	Project   string
 	Scope     string
+	SQLiteID  int64
 }
 
 // GetStaleVectors returns doc_ids of vectors with mismatched or null model versions.
@@ -351,8 +352,8 @@ func (c *Client) GetStaleVectors(ctx context.Context) ([]StaleVectorInfo, error)
 		var sqliteID sql.NullInt64
 		var docType, fieldType, project, scope sql.NullString
 
-		if err := rows.Scan(&info.DocID, &sqliteID, &docType, &fieldType, &project, &scope); err != nil {
-			return nil, fmt.Errorf("scan row: %w", err)
+		if scanErr := rows.Scan(&info.DocID, &sqliteID, &docType, &fieldType, &project, &scope); scanErr != nil {
+			return nil, fmt.Errorf("scan row: %w", scanErr)
 		}
 
 		info.SQLiteID = sqliteID.Int64

internal/vector/sqlitevec/helpers.go

@@ -8,20 +8,44 @@ const (
 	DocTypeObservation    DocType = "observation"
 	DocTypeSessionSummary DocType = "session_summary"
 	DocTypeUserPrompt     DocType = "user_prompt"
+	DocTypeCodeChunk      DocType = "code_chunk"
 )
 
 // Document represents a document to store with vector embedding.
 type Document struct {
+	Metadata map[string]any
 	ID       string
 	Content  string
-	Metadata map[string]any
 }
 
 // QueryResult represents a search result from vector search.
 type QueryResult struct {
-	ID       string
-	Distance float64
-	Metadata map[string]any
+	Metadata   map[string]any
+	ID         string
+	Distance   float64
+	Similarity float64
+}
+
+// DistanceToSimilarity converts sqlite-vec cosine distance to similarity score.
+// Cosine distance: 0 = identical, 2 = opposite
+// Similarity: 1.0 = identical, 0.0 = opposite
+func DistanceToSimilarity(distance float64) float64 {
+	return 1.0 - (distance / 2.0)
+}
+
+// FilterByThreshold filters results to only include those above the similarity threshold.
+// If maxResults > 0, also caps the number of results.
+func FilterByThreshold(results []QueryResult, threshold float64, maxResults int) []QueryResult {
+	var filtered []QueryResult
+	for _, r := range results {
+		if r.Similarity >= threshold {
+			filtered = append(filtered, r)
+			if maxResults > 0 && len(filtered) >= maxResults {
+				break
+			}
+		}
+	}
+	return filtered
 }
 
 // ExtractedIDs contains SQLite IDs extracted from query results, grouped by document type.

internal/vector/sqlitevec/helpers_test.go

@@ -42,10 +42,10 @@ func TestQueryResult_Fields(t *testing.T) {
 
 func TestBuildWhereFilter(t *testing.T) {
 	tests := []struct {
+		expected map[string]interface{}
 		name     string
 		docType  DocType
 		project  string
-		expected map[string]interface{}
 	}{
 		{
 			name:     "empty_filters",
@@ -474,9 +474,9 @@ func TestCopyMetadataMulti(t *testing.T) {
 func TestJoinStrings(t *testing.T) {
 	tests := []struct {
 		name     string
-		strs     []string
 		sep      string
 		expected string
+		strs     []string
 	}{
 		{
 			name:     "empty_slice",
@@ -522,8 +522,8 @@ func TestTruncateString(t *testing.T) {
 	tests := []struct {
 		name     string
 		input    string
-		maxLen   int
 		expected string
+		maxLen   int
 	}{
 		{
 			name:     "shorter_than_max",

internal/vector/sqlitevec/sync.go

@@ -5,13 +5,15 @@ import (
 	"context"
 	"fmt"
 
+	"github.com/lukaszraczylo/claude-mnemonic/internal/chunking"
 	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
 	"github.com/rs/zerolog/log"
 )
 
 // Sync provides synchronization between SQLite data and vector embeddings.
 type Sync struct {
-	client *Client
+	client          *Client
+	chunkingManager *chunking.Manager
 }
 
 // NewSync creates a new sync service.
@@ -19,9 +21,23 @@ func NewSync(client *Client) *Sync {
 	return &Sync{client: client}
 }
 
+// SetChunkingManager sets the code chunking manager (optional).
+// If set, observations will include code chunks from tracked files.
+func (s *Sync) SetChunkingManager(manager *chunking.Manager) {
+	s.chunkingManager = manager
+}
+
 // SyncObservation syncs a single observation to the vector store.
+// If a chunking manager is configured, also chunks tracked code files.
 func (s *Sync) SyncObservation(ctx context.Context, obs *models.Observation) error {
 	docs := s.formatObservationDocs(obs)
+
+	// Add code chunks from tracked files if chunking manager is available
+	if s.chunkingManager != nil {
+		codeChunkDocs := s.formatCodeChunkDocs(ctx, obs)
+		docs = append(docs, codeChunkDocs...)
+	}
+
 	if len(docs) == 0 {
 		return nil
 	}
@@ -99,6 +115,98 @@ func (s *Sync) formatObservationDocs(obs *models.Observation) []Document {
 	return docs
 }
 
+// formatCodeChunkDocs formats code chunks from tracked files into vector documents.
+// Uses AST-aware chunking to extract semantic code units (functions, classes, methods).
+func (s *Sync) formatCodeChunkDocs(ctx context.Context, obs *models.Observation) []Document {
+	if s.chunkingManager == nil {
+		return nil
+	}
+
+	// Determine scope for metadata
+	scope := string(obs.Scope)
+	if scope == "" {
+		scope = "project"
+	}
+
+	// Collect all tracked files (read + modified)
+	allFiles := make([]string, 0, len(obs.FilesRead)+len(obs.FilesModified))
+	allFiles = append(allFiles, obs.FilesRead...)
+	allFiles = append(allFiles, obs.FilesModified...)
+
+	// Filter to only files supported by chunking manager
+	var supportedFiles []string
+	for _, file := range allFiles {
+		if s.chunkingManager.SupportsFile(file) {
+			supportedFiles = append(supportedFiles, file)
+		}
+	}
+
+	if len(supportedFiles) == 0 {
+		return nil
+	}
+
+	// Chunk all supported files
+	results, errs := s.chunkingManager.ChunkFiles(ctx, supportedFiles)
+	if len(errs) > 0 {
+		// Log errors but don't fail the entire sync
+		for _, err := range errs {
+			log.Warn().Err(err).Msg("Failed to chunk file")
+		}
+	}
+
+	// Convert chunks to vector documents
+	docs := make([]Document, 0)
+	chunkIndex := 0
+
+	for filePath, chunks := range results {
+		for _, chunk := range chunks {
+			doc := Document{
+				ID:      fmt.Sprintf("obs_%d_chunk_%d", obs.ID, chunkIndex),
+				Content: chunk.SearchableContent(),
+				Metadata: map[string]any{
+					"sqlite_id":        obs.ID,
+					"doc_type":         string(DocTypeCodeChunk),
+					"field_type":       "code_chunk",
+					"sdk_session_id":   obs.SDKSessionID,
+					"project":          obs.Project,
+					"scope":            scope,
+					"created_at_epoch": obs.CreatedAtEpoch,
+					// Code chunk specific metadata
+					"file_path":   filePath,
+					"language":    string(chunk.Language),
+					"chunk_type":  string(chunk.Type),
+					"symbol_name": chunk.Name,
+					"start_line":  chunk.StartLine,
+					"end_line":    chunk.EndLine,
+				},
+			}
+
+			// Add parent name if this is a method
+			if chunk.ParentName != "" {
+				doc.Metadata["parent_name"] = chunk.ParentName
+			}
+
+			// Add signature if available
+			if chunk.Signature != "" {
+				doc.Metadata["signature"] = chunk.Signature
+			}
+
+			docs = append(docs, doc)
+			chunkIndex++
+		}
+	}
+
+	if len(docs) > 0 {
+		log.Debug().
+			Int64("observationId", obs.ID).
+			Int("codeChunks", len(docs)).
+			Int("files", len(results)).
+			Msg("Generated code chunk documents")
+	}
+
+	return docs
+}
+
 // SyncSummary syncs a single session summary to the vector store.
 func (s *Sync) SyncSummary(ctx context.Context, summary *models.SessionSummary) error {
 	docs := s.formatSummaryDocs(summary)
@@ -191,21 +299,27 @@ func (s *Sync) SyncUserPrompt(ctx context.Context, prompt *models.UserPromptWith
 }
 
 // DeleteObservations removes observation documents from the vector store.
+// Includes both observation fields (narrative, facts) and code chunks.
 func (s *Sync) DeleteObservations(ctx context.Context, observationIDs []int64) error {
 	if len(observationIDs) == 0 {
 		return nil
 	}
 
 	// Generate all possible document IDs for these observations
-	// Pattern: obs_{id}_narrative, obs_{id}_fact_{0..n}
+	// Pattern: obs_{id}_narrative, obs_{id}_fact_{0..n}, obs_{id}_chunk_{0..n}
 	const maxFactsPerObs = 20
-	ids := make([]string, 0, len(observationIDs)*(maxFactsPerObs+1))
+	const maxChunksPerObs = 100 // Reasonable upper bound for code chunks
+	ids := make([]string, 0, len(observationIDs)*(maxFactsPerObs+maxChunksPerObs+1))
 
 	for _, obsID := range observationIDs {
 		ids = append(ids, fmt.Sprintf("obs_%d_narrative", obsID))
 		for i := 0; i < maxFactsPerObs; i++ {
 			ids = append(ids, fmt.Sprintf("obs_%d_fact_%d", obsID, i))
 		}
+		// Include code chunk IDs
+		for i := 0; i < maxChunksPerObs; i++ {
+			ids = append(ids, fmt.Sprintf("obs_%d_chunk_%d", obsID, i))
+		}
 	}
 
 	if err := s.client.DeleteDocuments(ctx, ids); err != nil {
@@ -240,3 +354,101 @@ func (s *Sync) DeleteUserPrompts(ctx context.Context, promptIDs []int64) error {
 
 	return nil
 }
+
+// SyncPattern syncs a single pattern to the vector store.
+func (s *Sync) SyncPattern(ctx context.Context, pattern *models.Pattern) error {
+	docs := s.formatPatternDocs(pattern)
+	if len(docs) == 0 {
+		return nil
+	}
+
+	if err := s.client.AddDocuments(ctx, docs); err != nil {
+		return fmt.Errorf("add pattern docs: %w", err)
+	}
+
+	log.Debug().
+		Int64("patternId", pattern.ID).
+		Int("docCount", len(docs)).
+		Msg("Synced pattern to sqlite-vec")
+
+	return nil
+}
+
+// formatPatternDocs formats a pattern into vector documents.
+func (s *Sync) formatPatternDocs(pattern *models.Pattern) []Document {
+	docs := make([]Document, 0, 3)
+
+	baseMetadata := map[string]any{
+		"sqlite_id":        pattern.ID,
+		"doc_type":         "pattern",
+		"pattern_type":     string(pattern.Type),
+		"status":           string(pattern.Status),
+		"scope":            "global", // Patterns are always global
+		"frequency":        pattern.Frequency,
+		"confidence":       pattern.Confidence,
+		"created_at_epoch": pattern.CreatedAtEpoch,
+	}
+
+	if len(pattern.Signature) > 0 {
+		baseMetadata["signature"] = joinStrings(pattern.Signature, ",")
+	}
+	if len(pattern.Projects) > 0 {
+		baseMetadata["projects"] = joinStrings(pattern.Projects, ",")
+	}
+
+	// Pattern name as document
+	if pattern.Name != "" {
+		docs = append(docs, Document{
+			ID:       fmt.Sprintf("pattern_%d_name", pattern.ID),
+			Content:  pattern.Name,
+			Metadata: copyMetadata(baseMetadata, "field_type", "name"),
+		})
+	}
+
+	// Pattern description as document
+	if pattern.Description.Valid && pattern.Description.String != "" {
+		docs = append(docs, Document{
+			ID:       fmt.Sprintf("pattern_%d_description", pattern.ID),
+			Content:  pattern.Description.String,
+			Metadata: copyMetadata(baseMetadata, "field_type", "description"),
+		})
+	}
+
+	// Pattern recommendation as document
+	if pattern.Recommendation.Valid && pattern.Recommendation.String != "" {
+		docs = append(docs, Document{
+			ID:       fmt.Sprintf("pattern_%d_recommendation", pattern.ID),
+			Content:  pattern.Recommendation.String,
+			Metadata: copyMetadata(baseMetadata, "field_type", "recommendation"),
+		})
+	}
+
+	return docs
+}
+
+// DeletePatterns removes pattern documents from the vector store.
+func (s *Sync) DeletePatterns(ctx context.Context, patternIDs []int64) error {
+	if len(patternIDs) == 0 {
+		return nil
+	}
+
+	// Generate all possible document IDs for these patterns
+	// Pattern: pattern_{id}_name, pattern_{id}_description, pattern_{id}_recommendation
+	ids := make([]string, 0, len(patternIDs)*3)
+
+	for _, patternID := range patternIDs {
+		ids = append(ids, fmt.Sprintf("pattern_%d_name", patternID))
+		ids = append(ids, fmt.Sprintf("pattern_%d_description", patternID))
+		ids = append(ids, fmt.Sprintf("pattern_%d_recommendation", patternID))
+	}
+
+	if err := s.client.DeleteDocuments(ctx, ids); err != nil {
+		return fmt.Errorf("delete pattern docs: %w", err)
+	}
+
+	log.Debug().
+		Int("patternCount", len(patternIDs)).
+		Msg("Deleted patterns from sqlite-vec")
+
+	return nil
+}

internal/watcher/watcher.go

@@ -16,15 +16,15 @@ import (
 // Watcher monitors a file or directory for deletion and calls onDelete when removed.
 // It watches the parent directory since fsnotify cannot watch non-existent files.
 type Watcher struct {
-	targetPath string // The file/directory to watch for deletion
-	parentPath string // Parent directory (what we actually watch)
-	onDelete   func() // Callback when target is deleted
-	watcher    *fsnotify.Watcher
 	ctx        context.Context
+	onDelete   func()
+	watcher    *fsnotify.Watcher
 	cancel     context.CancelFunc
+	targetPath string
+	parentPath string
+	debounce   time.Duration
 	mu         sync.Mutex
 	running    bool
-	debounce   time.Duration
 }
 
 // New creates a new Watcher for the given target path.

internal/worker/handlers.go

@@ -4,7 +4,9 @@ package worker
 import (
 	"context"
 	"encoding/json"
+	"fmt"
 	"net/http"
+	"sort"
 	"strconv"
 	"time"
 
@@ -12,6 +14,8 @@ import (
 	"github.com/lukaszraczylo/claude-mnemonic/internal/db/sqlite"
 	"github.com/lukaszraczylo/claude-mnemonic/internal/embedding"
 	"github.com/lukaszraczylo/claude-mnemonic/internal/privacy"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/reranking"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/search/expansion"
 	"github.com/lukaszraczylo/claude-mnemonic/internal/vector/sqlitevec"
 	"github.com/lukaszraczylo/claude-mnemonic/internal/worker/sdk"
 	"github.com/lukaszraczylo/claude-mnemonic/internal/worker/session"
@@ -154,10 +158,10 @@ type SessionInitRequest struct {
 
 // SessionInitResponse is the response for session initialization.
 type SessionInitResponse struct {
+	Reason       string `json:"reason,omitempty"`
 	SessionDBID  int64  `json:"sessionDbId"`
 	PromptNumber int    `json:"promptNumber"`
 	Skipped      bool   `json:"skipped,omitempty"`
-	Reason       string `json:"reason,omitempty"`
 }
 
 // DuplicatePromptWindowSeconds is the time window for detecting duplicate prompt submissions.
@@ -292,8 +296,8 @@ func (s *Service) handleSessionStart(w http.ResponseWriter, r *http.Request) {
 	}
 
 	var req SessionStartRequest
-	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
-		http.Error(w, err.Error(), http.StatusBadRequest)
+	if decodeErr := json.NewDecoder(r.Body).Decode(&req); decodeErr != nil {
+		http.Error(w, decodeErr.Error(), http.StatusBadRequest)
 		return
 	}
 
@@ -744,15 +748,69 @@ func (s *Service) handleSearchByPrompt(w http.ResponseWriter, r *http.Request) {
 	var observations []*models.Observation
 	var err error
 	var usedVector bool
+	similarityScores := make(map[int64]float64) // Track similarity per observation
+
+	// Get threshold settings from config
+	threshold := s.config.ContextRelevanceThreshold
+	maxResults := s.config.ContextMaxPromptResults
+
+	// Generate expanded queries if query expander is available
+	var expandedQueries []expansion.ExpandedQuery
+	var detectedIntent string
+	if s.queryExpander != nil {
+		cfg := expansion.DefaultConfig()
+		cfg.EnableVocabularyExpansion = false // Vocabulary expansion is optional
+		expandedQueries = s.queryExpander.Expand(r.Context(), query, cfg)
+		if len(expandedQueries) > 0 {
+			detectedIntent = string(expandedQueries[0].Intent)
+		}
+	}
+	if len(expandedQueries) == 0 {
+		// Fallback to just the original query
+		expandedQueries = []expansion.ExpandedQuery{
+			{Query: query, Weight: 1.0, Source: "original"},
+		}
+	}
 
 	// Try vector search first if available
 	if s.vectorClient != nil && s.vectorClient.IsConnected() {
 		where := sqlitevec.BuildWhereFilter(sqlitevec.DocTypeObservation, "")
 
-		vectorResults, vecErr := s.vectorClient.Query(r.Context(), query, limit*2, where)
-		if vecErr == nil && len(vectorResults) > 0 {
+		// Search with each expanded query and merge results
+		allVectorResults := make([]sqlitevec.QueryResult, 0)
+		queryWeights := make(map[string]float64) // Track weights for score merging
+
+		for _, eq := range expandedQueries {
+			vectorResults, vecErr := s.vectorClient.Query(r.Context(), eq.Query, limit*2, where)
+			if vecErr == nil && len(vectorResults) > 0 {
+				// Apply weight to similarity scores before merging
+				for i := range vectorResults {
+					vectorResults[i].Similarity *= eq.Weight
+				}
+				allVectorResults = append(allVectorResults, vectorResults...)
+				queryWeights[eq.Query] = eq.Weight
+			}
+		}
+
+		if len(allVectorResults) > 0 {
+			// Filter by relevance threshold before extracting IDs
+			// Use a slightly lower threshold for expanded queries
+			effectiveThreshold := threshold * 0.9 // Allow slightly lower scores for expanded queries
+			filteredResults := sqlitevec.FilterByThreshold(allVectorResults, effectiveThreshold, 0)
+
+			// Build similarity map for filtered results (keeping highest weighted score per observation)
+			for _, vr := range filteredResults {
+				if sqliteID, ok := vr.Metadata["sqlite_id"].(float64); ok {
+					id := int64(sqliteID)
+					// Keep the highest score for each observation
+					if existing, exists := similarityScores[id]; !exists || vr.Similarity > existing {
+						similarityScores[id] = vr.Similarity
+					}
+				}
+			}
+
 			// Extract observation IDs with project/scope filtering using shared helper
-			obsIDs := sqlitevec.ExtractObservationIDs(vectorResults, project)
+			obsIDs := sqlitevec.ExtractObservationIDs(filteredResults, project)
 
 			if len(obsIDs) > 0 {
 				// Fetch full observations from SQLite
@@ -802,23 +860,132 @@ func (s *Service) handleSearchByPrompt(w http.ResponseWriter, r *http.Request) {
 		freshObservations = append(freshObservations, obs)
 	}
 
+	// Apply cross-encoder reranking if available
+	var reranked bool
+	if s.reranker != nil && len(freshObservations) > 0 && usedVector {
+		// Build candidates from observations with their bi-encoder scores
+		candidates := make([]reranking.Candidate, len(freshObservations))
+		for i, obs := range freshObservations {
+			content := obs.Title.String
+			if obs.Narrative.Valid && obs.Narrative.String != "" {
+				content = content + " " + obs.Narrative.String
+			}
+			candidates[i] = reranking.Candidate{
+				ID:       fmt.Sprintf("%d", obs.ID),
+				Content:  content,
+				Score:    similarityScores[obs.ID],
+				Metadata: map[string]any{"obs_idx": i},
+			}
+		}
+
+		// Rerank using cross-encoder - use pure mode or combined scores
+		var rerankResults []reranking.RerankResult
+		var rerankErr error
+		if s.config.RerankingPureMode {
+			rerankResults, rerankErr = s.reranker.RerankByScore(query, candidates, s.config.RerankingResults)
+		} else {
+			rerankResults, rerankErr = s.reranker.Rerank(query, candidates, s.config.RerankingResults)
+		}
+		if rerankErr != nil {
+			log.Warn().Err(rerankErr).Msg("Cross-encoder reranking failed, using original order")
+		} else if len(rerankResults) > 0 {
+			// Update similarity scores with reranked scores
+			for _, rr := range rerankResults {
+				if id, err := strconv.ParseInt(rr.ID, 10, 64); err == nil {
+					similarityScores[id] = rr.CombinedScore
+				}
+			}
+
+			// Reorder observations based on rerank results
+			reorderedObs := make([]*models.Observation, 0, len(rerankResults))
+			obsMap := make(map[int64]*models.Observation)
+			for _, obs := range freshObservations {
+				obsMap[obs.ID] = obs
+			}
+			for _, rr := range rerankResults {
+				if id, err := strconv.ParseInt(rr.ID, 10, 64); err == nil {
+					if obs, ok := obsMap[id]; ok {
+						reorderedObs = append(reorderedObs, obs)
+					}
+				}
+			}
+			freshObservations = reorderedObs
+			reranked = true
+
+			log.Debug().
+				Int("candidates", len(candidates)).
+				Int("returned", len(rerankResults)).
+				Msg("Cross-encoder reranking complete")
+		}
+	}
+
 	// Cluster similar observations to remove duplicates
 	clusteredObservations := clusterObservations(freshObservations, 0.4)
 
+	// Sort by similarity score (highest first) if we have scores and didn't rerank
+	if len(similarityScores) > 0 && len(clusteredObservations) > 0 && !reranked {
+		sort.Slice(clusteredObservations, func(i, j int) bool {
+			scoreI := similarityScores[clusteredObservations[i].ID]
+			scoreJ := similarityScores[clusteredObservations[j].ID]
+			return scoreI > scoreJ
+		})
+	}
+
+	// Apply max results cap if configured
+	if maxResults > 0 && len(clusteredObservations) > maxResults {
+		clusteredObservations = clusteredObservations[:maxResults]
+	}
+
 	// Record retrieval stats (no verification done, so verified=0, deleted=0)
 	s.recordRetrievalStats(project, int64(len(clusteredObservations)), 0, 0, true)
 
+	// Increment retrieval counts for scoring (async, non-blocking)
+	if len(clusteredObservations) > 0 {
+		ids := make([]int64, len(clusteredObservations))
+		for i, obs := range clusteredObservations {
+			ids[i] = obs.ID
+		}
+		s.incrementRetrievalCounts(ids)
+	}
+
 	log.Info().
 		Str("project", project).
 		Str("query", query).
+		Str("intent", detectedIntent).
+		Int("expansions", len(expandedQueries)).
 		Int("found", len(clusteredObservations)).
 		Int("stale_excluded", staleCount).
+		Float64("threshold", threshold).
 		Msg("Prompt-based observation search")
 
+	// Build response with similarity scores
+	obsWithScores := make([]map[string]interface{}, len(clusteredObservations))
+	for i, obs := range clusteredObservations {
+		obsMap := obs.ToMap()
+		if score, ok := similarityScores[obs.ID]; ok {
+			obsMap["similarity"] = score
+		}
+		obsWithScores[i] = obsMap
+	}
+
+	// Build expansion info for response
+	expansionInfo := make([]map[string]interface{}, len(expandedQueries))
+	for i, eq := range expandedQueries {
+		expansionInfo[i] = map[string]interface{}{
+			"query":  eq.Query,
+			"weight": eq.Weight,
+			"source": eq.Source,
+		}
+	}
+
 	writeJSON(w, map[string]interface{}{
 		"project":      project,
 		"query":        query,
-		"observations": clusteredObservations,
+		"intent":       detectedIntent,
+		"expansions":   expansionInfo,
+		"observations": obsWithScores,
+		"threshold":    threshold,
+		"max_results":  maxResults,
 	})
 }
 
@@ -886,6 +1053,15 @@ func (s *Service) handleContextInject(w http.ResponseWriter, r *http.Request) {
 	// Record retrieval stats (no verification done)
 	s.recordRetrievalStats(project, int64(len(clusteredObservations)), 0, 0, false)
 
+	// Increment retrieval counts for scoring (async, non-blocking)
+	if len(clusteredObservations) > 0 {
+		ids := make([]int64, len(clusteredObservations))
+		for i, obs := range clusteredObservations {
+			ids[i] = obs.ID
+		}
+		s.incrementRetrievalCounts(ids)
+	}
+
 	log.Info().
 		Str("project", project).
 		Int("total", len(observations)).
@@ -1044,6 +1220,35 @@ func (s *Service) handleSelfCheck(w http.ResponseWriter, r *http.Request) {
 	}
 	components = append(components, sseStatus)
 
+	// Check Cross-Encoder Reranker
+	rerankerStatus := ComponentHealth{Name: "Cross-Encoder Reranker", Status: "healthy"}
+	if !s.config.RerankingEnabled {
+		rerankerStatus.Status = "degraded"
+		rerankerStatus.Message = "Disabled in config"
+		if overall == "healthy" {
+			overall = "degraded"
+		}
+	} else if s.reranker == nil {
+		rerankerStatus.Status = "degraded"
+		rerankerStatus.Message = "Not initialized"
+		if overall == "healthy" {
+			overall = "degraded"
+		}
+	} else {
+		// Verify reranker is functional using Score
+		_, normalizedScore, err := s.reranker.Score("test query", "test document")
+		if err != nil {
+			rerankerStatus.Status = "unhealthy"
+			rerankerStatus.Message = fmt.Sprintf("Score check failed: %v", err)
+			if overall == "healthy" {
+				overall = "degraded"
+			}
+		} else {
+			rerankerStatus.Message = fmt.Sprintf("Score check passed (%.4f)", normalizedScore)
+		}
+	}
+	components = append(components, rerankerStatus)
+
 	// Calculate uptime
 	uptime := time.Since(s.startTime).Round(time.Second).String()
 
@@ -1107,3 +1312,85 @@ func (s *Service) handleRestart(w http.ResponseWriter, r *http.Request) {
 		}
 	}()
 }
+
+// handleGetGraphStats returns observation graph statistics.
+func (s *Service) handleGetGraphStats(w http.ResponseWriter, r *http.Request) {
+	if s.graphSearchClient == nil {
+		writeJSON(w, map[string]interface{}{
+			"enabled": false,
+			"message": "Graph search not enabled",
+		})
+		return
+	}
+
+	stats := s.graphSearchClient.GetGraphStats()
+
+	response := map[string]interface{}{
+		"enabled":      s.config.GraphEnabled,
+		"nodeCount":    stats.NodeCount,
+		"edgeCount":    stats.EdgeCount,
+		"avgDegree":    stats.AvgDegree,
+		"maxDegree":    stats.MaxDegree,
+		"minDegree":    stats.MinDegree,
+		"medianDegree": stats.MedianDegree,
+		"edgeTypes":    stats.EdgeTypes,
+		"config": map[string]interface{}{
+			"maxHops":            s.config.GraphMaxHops,
+			"branchFactor":       s.config.GraphBranchFactor,
+			"edgeWeight":         s.config.GraphEdgeWeight,
+			"rebuildIntervalMin": s.config.GraphRebuildIntervalMin,
+		},
+	}
+
+	writeJSON(w, response)
+}
+
+// handleGetVectorMetrics returns hybrid vector storage metrics.
+func (s *Service) handleGetVectorMetrics(w http.ResponseWriter, r *http.Request) {
+	if s.hybridMetrics == nil {
+		writeJSON(w, map[string]interface{}{
+			"enabled": false,
+			"message": "Vector metrics not available",
+		})
+		return
+	}
+
+	snapshot := s.hybridMetrics.GetSnapshot()
+
+	response := map[string]interface{}{
+		"queries": map[string]interface{}{
+			"total":    snapshot.TotalQueries,
+			"hubOnly":  snapshot.HubOnlyQueries,
+			"hybrid":   snapshot.HybridQueries,
+			"onDemand": snapshot.OnDemandQueries,
+			"graph":    snapshot.GraphQueries,
+		},
+		"latency": map[string]interface{}{
+			"avg":          snapshot.AvgLatency.String(),
+			"p50":          snapshot.P50Latency.String(),
+			"p95":          snapshot.P95Latency.String(),
+			"p99":          snapshot.P99Latency.String(),
+			"avgHub":       snapshot.AvgHubLatency.String(),
+			"avgRecompute": snapshot.AvgRecomputeLatency.String(),
+		},
+		"storage": map[string]interface{}{
+			"totalDocuments":   snapshot.TotalDocuments,
+			"hubDocuments":     snapshot.HubDocuments,
+			"storedEmbeddings": snapshot.StoredEmbeddings,
+			"savingsPercent":   snapshot.StorageSavingsPercent,
+			"recomputedTotal":  snapshot.RecomputedTotal,
+		},
+		"cache": map[string]interface{}{
+			"hits":    snapshot.CacheHits,
+			"misses":  snapshot.CacheMisses,
+			"hitRate": snapshot.CacheHitRate,
+		},
+		"graph": map[string]interface{}{
+			"traversals": snapshot.GraphTraversals,
+			"avgDepth":   snapshot.AvgTraversalDepth,
+		},
+		"uptime": snapshot.Uptime.String(),
+	}
+
+	writeJSON(w, response)
+}

internal/worker/handlers_patterns.go

@@ -0,0 +1,292 @@
+// Package worker provides the main worker service for claude-mnemonic.
+package worker
+
+import (
+	"encoding/json"
+	"net/http"
+	"strconv"
+
+	"github.com/go-chi/chi/v5"
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+)
+
+// DefaultPatternsLimit is the default number of patterns to return.
+const DefaultPatternsLimit = 100
+
+// handleGetPatterns returns all active patterns, optionally filtered by type or project.
+func (s *Service) handleGetPatterns(w http.ResponseWriter, r *http.Request) {
+	s.initMu.RLock()
+	store := s.patternStore
+	s.initMu.RUnlock()
+
+	if store == nil {
+		http.Error(w, "pattern store not initialized", http.StatusServiceUnavailable)
+		return
+	}
+
+	// Parse query parameters
+	limit := DefaultPatternsLimit
+	if l := r.URL.Query().Get("limit"); l != "" {
+		if parsed, err := strconv.Atoi(l); err == nil && parsed > 0 {
+			limit = parsed
+		}
+	}
+
+	patternType := r.URL.Query().Get("type")
+	project := r.URL.Query().Get("project")
+
+	var patterns []*models.Pattern
+	var err error
+
+	if patternType != "" {
+		// Filter by type
+		patterns, err = store.GetPatternsByType(r.Context(), models.PatternType(patternType), limit)
+	} else if project != "" {
+		// Filter by project
+		patterns, err = store.GetPatternsByProject(r.Context(), project, limit)
+	} else {
+		// Get all active patterns
+		patterns, err = store.GetActivePatterns(r.Context(), limit)
+	}
+
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	writeJSON(w, patterns)
+}
+
+// handleGetPatternStats returns aggregate statistics about patterns.
+func (s *Service) handleGetPatternStats(w http.ResponseWriter, r *http.Request) {
+	s.initMu.RLock()
+	store := s.patternStore
+	s.initMu.RUnlock()
+
+	if store == nil {
+		http.Error(w, "pattern store not initialized", http.StatusServiceUnavailable)
+		return
+	}
+
+	stats, err := store.GetPatternStats(r.Context())
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	writeJSON(w, stats)
+}
+
+// handleGetPatternByID returns a single pattern by ID.
+func (s *Service) handleGetPatternByID(w http.ResponseWriter, r *http.Request) {
+	s.initMu.RLock()
+	store := s.patternStore
+	s.initMu.RUnlock()
+
+	if store == nil {
+		http.Error(w, "pattern store not initialized", http.StatusServiceUnavailable)
+		return
+	}
+
+	idStr := chi.URLParam(r, "id")
+	id, err := strconv.ParseInt(idStr, 10, 64)
+	if err != nil {
+		http.Error(w, "invalid pattern ID", http.StatusBadRequest)
+		return
+	}
+
+	pattern, err := store.GetPatternByID(r.Context(), id)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+	if pattern == nil {
+		http.Error(w, "pattern not found", http.StatusNotFound)
+		return
+	}
+
+	writeJSON(w, pattern)
+}
+
+// handleGetPatternInsight returns a formatted insight string for a pattern.
+func (s *Service) handleGetPatternInsight(w http.ResponseWriter, r *http.Request) {
+	s.initMu.RLock()
+	detector := s.patternDetector
+	s.initMu.RUnlock()
+
+	if detector == nil {
+		http.Error(w, "pattern detector not initialized", http.StatusServiceUnavailable)
+		return
+	}
+
+	idStr := chi.URLParam(r, "id")
+	id, err := strconv.ParseInt(idStr, 10, 64)
+	if err != nil {
+		http.Error(w, "invalid pattern ID", http.StatusBadRequest)
+		return
+	}
+
+	insight, err := detector.GetPatternInsight(r.Context(), id)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	writeJSON(w, map[string]string{"insight": insight})
+}
+
+// handleDeletePattern deletes a pattern by ID.
+func (s *Service) handleDeletePattern(w http.ResponseWriter, r *http.Request) {
+	s.initMu.RLock()
+	store := s.patternStore
+	s.initMu.RUnlock()
+
+	if store == nil {
+		http.Error(w, "pattern store not initialized", http.StatusServiceUnavailable)
+		return
+	}
+
+	idStr := chi.URLParam(r, "id")
+	id, err := strconv.ParseInt(idStr, 10, 64)
+	if err != nil {
+		http.Error(w, "invalid pattern ID", http.StatusBadRequest)
+		return
+	}
+
+	if err := store.DeletePattern(r.Context(), id); err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	writeJSON(w, map[string]string{"status": "deleted"})
+}
+
+// handleDeprecatePattern marks a pattern as deprecated.
+func (s *Service) handleDeprecatePattern(w http.ResponseWriter, r *http.Request) {
+	s.initMu.RLock()
+	store := s.patternStore
+	s.initMu.RUnlock()
+
+	if store == nil {
+		http.Error(w, "pattern store not initialized", http.StatusServiceUnavailable)
+		return
+	}
+
+	idStr := chi.URLParam(r, "id")
+	id, err := strconv.ParseInt(idStr, 10, 64)
+	if err != nil {
+		http.Error(w, "invalid pattern ID", http.StatusBadRequest)
+		return
+	}
+
+	if err := store.MarkPatternDeprecated(r.Context(), id); err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	writeJSON(w, map[string]string{"status": "deprecated"})
+}
+
+// MergePatternsRequest is the request body for merging patterns.
+type MergePatternsRequest struct {
+	SourceID int64 `json:"source_id"`
+	TargetID int64 `json:"target_id"`
+}
+
+// handleSearchPatterns performs full-text search on patterns.
+func (s *Service) handleSearchPatterns(w http.ResponseWriter, r *http.Request) {
+	s.initMu.RLock()
+	store := s.patternStore
+	s.initMu.RUnlock()
+
+	if store == nil {
+		http.Error(w, "pattern store not initialized", http.StatusServiceUnavailable)
+		return
+	}
+
+	query := r.URL.Query().Get("q")
+	if query == "" {
+		http.Error(w, "query parameter 'q' is required", http.StatusBadRequest)
+		return
+	}
+
+	limit := DefaultPatternsLimit
+	if l := r.URL.Query().Get("limit"); l != "" {
+		if parsed, err := strconv.Atoi(l); err == nil && parsed > 0 {
+			limit = parsed
+		}
+	}
+
+	patterns, err := store.SearchPatternsFTS(r.Context(), query, limit)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	writeJSON(w, patterns)
+}
+
+// handleGetPatternByName returns a pattern by its name.
+func (s *Service) handleGetPatternByName(w http.ResponseWriter, r *http.Request) {
+	s.initMu.RLock()
+	store := s.patternStore
+	s.initMu.RUnlock()
+
+	if store == nil {
+		http.Error(w, "pattern store not initialized", http.StatusServiceUnavailable)
+		return
+	}
+
+	name := r.URL.Query().Get("name")
+	if name == "" {
+		http.Error(w, "query parameter 'name' is required", http.StatusBadRequest)
+		return
+	}
+
+	pattern, err := store.GetPatternByName(r.Context(), name)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+	if pattern == nil {
+		http.Error(w, "pattern not found", http.StatusNotFound)
+		return
+	}
+
+	writeJSON(w, pattern)
+}
+
+// handleMergePatterns merges a source pattern into a target pattern.
+func (s *Service) handleMergePatterns(w http.ResponseWriter, r *http.Request) {
+	s.initMu.RLock()
+	store := s.patternStore
+	s.initMu.RUnlock()
+
+	if store == nil {
+		http.Error(w, "pattern store not initialized", http.StatusServiceUnavailable)
+		return
+	}
+
+	var req MergePatternsRequest
+	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+		http.Error(w, err.Error(), http.StatusBadRequest)
+		return
+	}
+
+	if req.SourceID == 0 || req.TargetID == 0 {
+		http.Error(w, "source_id and target_id are required", http.StatusBadRequest)
+		return
+	}
+
+	if req.SourceID == req.TargetID {
+		http.Error(w, "source_id and target_id cannot be the same", http.StatusBadRequest)
+		return
+	}
+
+	if err := store.MergePatterns(r.Context(), req.SourceID, req.TargetID); err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	writeJSON(w, map[string]string{"status": "merged"})
+}

internal/worker/handlers_relations.go

@@ -0,0 +1,174 @@
+// Package worker provides the main worker service for claude-mnemonic.
+package worker
+
+import (
+	"net/http"
+	"strconv"
+
+	"github.com/go-chi/chi/v5"
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+)
+
+// DefaultRelationsLimit is the default number of relations to return.
+const DefaultRelationsLimit = 50
+
+// handleGetRelations returns relations for an observation.
+func (s *Service) handleGetRelations(w http.ResponseWriter, r *http.Request) {
+	idStr := chi.URLParam(r, "id")
+	id, err := strconv.ParseInt(idStr, 10, 64)
+	if err != nil {
+		http.Error(w, "invalid observation id", http.StatusBadRequest)
+		return
+	}
+
+	relations, err := s.relationStore.GetRelationsWithDetails(r.Context(), id)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	if relations == nil {
+		relations = []*models.RelationWithDetails{}
+	}
+
+	writeJSON(w, relations)
+}
+
+// handleGetRelationGraph returns the relation graph for an observation.
+func (s *Service) handleGetRelationGraph(w http.ResponseWriter, r *http.Request) {
+	idStr := chi.URLParam(r, "id")
+	id, err := strconv.ParseInt(idStr, 10, 64)
+	if err != nil {
+		http.Error(w, "invalid observation id", http.StatusBadRequest)
+		return
+	}
+
+	// Get depth parameter (default 2)
+	depth := 2
+	if depthStr := r.URL.Query().Get("depth"); depthStr != "" {
+		if d, atoiErr := strconv.Atoi(depthStr); atoiErr == nil && d > 0 && d <= 5 {
+			depth = d
+		}
+	}
+
+	graph, err := s.relationStore.GetRelationGraph(r.Context(), id, depth)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	writeJSON(w, graph)
+}
+
+// handleGetRelatedObservations returns observations related to a given one.
+func (s *Service) handleGetRelatedObservations(w http.ResponseWriter, r *http.Request) {
+	idStr := chi.URLParam(r, "id")
+	id, err := strconv.ParseInt(idStr, 10, 64)
+	if err != nil {
+		http.Error(w, "invalid observation id", http.StatusBadRequest)
+		return
+	}
+
+	// Get minimum confidence parameter (default 0.4)
+	minConfidence := 0.4
+	if confStr := r.URL.Query().Get("min_confidence"); confStr != "" {
+		if c, parseErr := strconv.ParseFloat(confStr, 64); parseErr == nil && c >= 0 && c <= 1 {
+			minConfidence = c
+		}
+	}
+
+	// Get related observation IDs
+	relatedIDs, err := s.relationStore.GetRelatedObservationIDs(r.Context(), id, minConfidence)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	if len(relatedIDs) == 0 {
+		writeJSON(w, []*models.Observation{})
+		return
+	}
+
+	// Fetch full observations
+	observations, err := s.observationStore.GetObservationsByIDs(r.Context(), relatedIDs, "importance", 50)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	if observations == nil {
+		observations = []*models.Observation{}
+	}
+
+	writeJSON(w, observations)
+}
+
+// handleGetRelationsByType returns all relations of a specific type.
+func (s *Service) handleGetRelationsByType(w http.ResponseWriter, r *http.Request) {
+	relType := chi.URLParam(r, "type")
+
+	// Validate relation type
+	validType := false
+	for _, t := range models.AllRelationTypes {
+		if string(t) == relType {
+			validType = true
+			break
+		}
+	}
+	if !validType {
+		http.Error(w, "invalid relation type", http.StatusBadRequest)
+		return
+	}
+
+	limit := DefaultRelationsLimit
+	if limitStr := r.URL.Query().Get("limit"); limitStr != "" {
+		if l, err := strconv.Atoi(limitStr); err == nil && l > 0 && l <= 100 {
+			limit = l
+		}
+	}
+
+	relations, err := s.relationStore.GetRelationsByType(r.Context(), models.RelationType(relType), limit)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	if relations == nil {
+		relations = []*models.ObservationRelation{}
+	}
+
+	writeJSON(w, relations)
+}
+
+// handleGetRelationStats returns statistics about relations.
+func (s *Service) handleGetRelationStats(w http.ResponseWriter, r *http.Request) {
+	// Get total relation count
+	totalCount, err := s.relationStore.GetTotalRelationCount(r.Context())
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	// Get high confidence relations count
+	highConfRelations, err := s.relationStore.GetHighConfidenceRelations(r.Context(), 0.7, 1000)
+	if err != nil {
+		http.Error(w, err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	// Count by relation type
+	typeCounts := make(map[string]int)
+	for _, t := range models.AllRelationTypes {
+		relations, err := s.relationStore.GetRelationsByType(r.Context(), t, 1000)
+		if err == nil {
+			typeCounts[string(t)] = len(relations)
+		}
+	}
+
+	writeJSON(w, map[string]interface{}{
+		"total_count":         totalCount,
+		"high_confidence":     len(highConfRelations),
+		"by_type":             typeCounts,
+		"min_confidence_used": 0.4,
+	})
+}

internal/worker/handlers_scoring.go

@@ -0,0 +1,359 @@
+// Package worker provides the main worker service for claude-mnemonic.
+package worker
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"strconv"
+	"time"
+
+	"github.com/go-chi/chi/v5"
+	"github.com/lukaszraczylo/claude-mnemonic/pkg/models"
+	"github.com/rs/zerolog/log"
+)
+
+// FeedbackRequest represents a user feedback submission.
+type FeedbackRequest struct {
+	Feedback int `json:"feedback"` // -1 (thumbs down), 0 (neutral), 1 (thumbs up)
+}
+
+// handleObservationFeedback handles user feedback (thumbs up/down) for an observation.
+// POST /api/observations/{id}/feedback
+func (s *Service) handleObservationFeedback(w http.ResponseWriter, r *http.Request) {
+	// Parse observation ID
+	idStr := chi.URLParam(r, "id")
+	id, err := strconv.ParseInt(idStr, 10, 64)
+	if err != nil {
+		http.Error(w, "invalid observation id", http.StatusBadRequest)
+		return
+	}
+
+	// Parse request body
+	var req FeedbackRequest
+	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+		http.Error(w, "invalid request body", http.StatusBadRequest)
+		return
+	}
+
+	// Validate feedback value
+	if req.Feedback < -1 || req.Feedback > 1 {
+		http.Error(w, "feedback must be -1, 0, or 1", http.StatusBadRequest)
+		return
+	}
+
+	// Get required components
+	s.initMu.RLock()
+	observationStore := s.observationStore
+	scoreCalculator := s.scoreCalculator
+	s.initMu.RUnlock()
+
+	if observationStore == nil {
+		http.Error(w, "service not ready", http.StatusServiceUnavailable)
+		return
+	}
+
+	// Update feedback in database
+	if err := observationStore.UpdateObservationFeedback(r.Context(), id, req.Feedback); err != nil {
+		http.Error(w, "failed to update feedback", http.StatusInternalServerError)
+		return
+	}
+
+	// Recalculate score immediately if calculator is available
+	var newScore float64
+	if scoreCalculator != nil {
+		obs, err := observationStore.GetObservationByID(r.Context(), id)
+		if err == nil && obs != nil {
+			obs.UserFeedback = req.Feedback // Apply the new feedback
+			newScore = scoreCalculator.Calculate(obs, time.Now())
+			if scoreErr := observationStore.UpdateImportanceScore(r.Context(), id, newScore); scoreErr != nil {
+				// Log but don't fail - feedback was recorded
+				// Score will be updated on next recalculation cycle
+				log.Warn().Err(scoreErr).Int64("id", id).Msg("Failed to update importance score after feedback")
+			}
+		}
+	}
+
+	// Broadcast update via SSE
+	s.sseBroadcaster.Broadcast(map[string]interface{}{
+		"type":     "observation_feedback",
+		"id":       id,
+		"feedback": req.Feedback,
+		"score":    newScore,
+	})
+
+	writeJSON(w, map[string]interface{}{
+		"status":   "ok",
+		"id":       id,
+		"feedback": req.Feedback,
+		"score":    newScore,
+	})
+}
+
+// handleGetScoringStats returns scoring statistics and configuration.
+// GET /api/scoring/stats
+func (s *Service) handleGetScoringStats(w http.ResponseWriter, r *http.Request) {
+	project := r.URL.Query().Get("project")
+
+	s.initMu.RLock()
+	observationStore := s.observationStore
+	recalculator := s.recalculator
+	s.initMu.RUnlock()
+
+	if observationStore == nil {
+		http.Error(w, "service not ready", http.StatusServiceUnavailable)
+		return
+	}
+
+	// Get feedback statistics
+	feedbackStats, err := observationStore.GetObservationFeedbackStats(r.Context(), project)
+	if err != nil {
+		http.Error(w, "failed to get feedback stats", http.StatusInternalServerError)
+		return
+	}
+
+	response := map[string]interface{}{
+		"feedback": feedbackStats,
+	}
+
+	// Add recalculator stats if available
+	if recalculator != nil {
+		response["recalculator"] = recalculator.GetStats()
+	}
+
+	writeJSON(w, response)
+}
+
+// handleGetTopObservations returns the highest-scoring observations.
+// GET /api/observations/top
+func (s *Service) handleGetTopObservations(w http.ResponseWriter, r *http.Request) {
+	limit := parseIntParam(r, "limit", 10)
+	project := r.URL.Query().Get("project")
+
+	s.initMu.RLock()
+	observationStore := s.observationStore
+	s.initMu.RUnlock()
+
+	if observationStore == nil {
+		http.Error(w, "service not ready", http.StatusServiceUnavailable)
+		return
+	}
+
+	observations, err := observationStore.GetTopScoringObservations(r.Context(), project, limit)
+	if err != nil {
+		http.Error(w, "failed to get top observations", http.StatusInternalServerError)
+		return
+	}
+
+	if observations == nil {
+		observations = []*models.Observation{}
+	}
+
+	writeJSON(w, observations)
+}
+
+// handleGetMostRetrieved returns the most frequently retrieved observations.
+// GET /api/observations/most-retrieved
+func (s *Service) handleGetMostRetrieved(w http.ResponseWriter, r *http.Request) {
+	limit := parseIntParam(r, "limit", 10)
+	project := r.URL.Query().Get("project")
+
+	s.initMu.RLock()
+	observationStore := s.observationStore
+	s.initMu.RUnlock()
+
+	if observationStore == nil {
+		http.Error(w, "service not ready", http.StatusServiceUnavailable)
+		return
+	}
+
+	observations, err := observationStore.GetMostRetrievedObservations(r.Context(), project, limit)
+	if err != nil {
+		http.Error(w, "failed to get most retrieved observations", http.StatusInternalServerError)
+		return
+	}
+
+	if observations == nil {
+		observations = []*models.Observation{}
+	}
+
+	writeJSON(w, observations)
+}
+
+// handleExplainScore returns a breakdown of how an observation's score was calculated.
+// GET /api/observations/{id}/score
+func (s *Service) handleExplainScore(w http.ResponseWriter, r *http.Request) {
+	// Parse observation ID
+	idStr := chi.URLParam(r, "id")
+	id, err := strconv.ParseInt(idStr, 10, 64)
+	if err != nil {
+		http.Error(w, "invalid observation id", http.StatusBadRequest)
+		return
+	}
+
+	s.initMu.RLock()
+	observationStore := s.observationStore
+	scoreCalculator := s.scoreCalculator
+	s.initMu.RUnlock()
+
+	if observationStore == nil || scoreCalculator == nil {
+		http.Error(w, "service not ready", http.StatusServiceUnavailable)
+		return
+	}
+
+	// Get observation
+	obs, err := observationStore.GetObservationByID(r.Context(), id)
+	if err != nil {
+		http.Error(w, "failed to get observation", http.StatusInternalServerError)
+		return
+	}
+	if obs == nil {
+		http.Error(w, "observation not found", http.StatusNotFound)
+		return
+	}
+
+	// Calculate score components
+	components := scoreCalculator.CalculateComponents(obs, time.Now())
+
+	writeJSON(w, map[string]interface{}{
+		"id":         id,
+		"components": components,
+		"config":     scoreCalculator.GetConfig(),
+	})
+}
+
+// handleUpdateConceptWeight updates a concept weight.
+// PUT /api/scoring/concepts/{concept}
+func (s *Service) handleUpdateConceptWeight(w http.ResponseWriter, r *http.Request) {
+	concept := chi.URLParam(r, "concept")
+	if concept == "" {
+		http.Error(w, "concept is required", http.StatusBadRequest)
+		return
+	}
+
+	var req struct {
+		Weight float64 `json:"weight"`
+	}
+	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+		http.Error(w, "invalid request body", http.StatusBadRequest)
+		return
+	}
+
+	// Validate weight
+	if req.Weight < 0 || req.Weight > 1 {
+		http.Error(w, "weight must be between 0 and 1", http.StatusBadRequest)
+		return
+	}
+
+	s.initMu.RLock()
+	observationStore := s.observationStore
+	recalculator := s.recalculator
+	s.initMu.RUnlock()
+
+	if observationStore == nil {
+		http.Error(w, "service not ready", http.StatusServiceUnavailable)
+		return
+	}
+
+	// Update in database
+	if err := observationStore.UpdateConceptWeight(r.Context(), concept, req.Weight); err != nil {
+		http.Error(w, "failed to update concept weight", http.StatusInternalServerError)
+		return
+	}
+
+	// Refresh concept weights in recalculator
+	if recalculator != nil {
+		if refreshErr := recalculator.RefreshConceptWeights(r.Context()); refreshErr != nil {
+			// Log but don't fail - weight was saved
+			log.Warn().Err(refreshErr).Str("concept", concept).Msg("Failed to refresh concept weights in recalculator")
+		}
+	}
+
+	writeJSON(w, map[string]interface{}{
+		"status":  "ok",
+		"concept": concept,
+		"weight":  req.Weight,
+	})
+}
+
+// handleGetConceptWeights returns all concept weights.
+// GET /api/scoring/concepts
+func (s *Service) handleGetConceptWeights(w http.ResponseWriter, r *http.Request) {
+	s.initMu.RLock()
+	observationStore := s.observationStore
+	s.initMu.RUnlock()
+
+	if observationStore == nil {
+		http.Error(w, "service not ready", http.StatusServiceUnavailable)
+		return
+	}
+
+	weights, err := observationStore.GetConceptWeights(r.Context())
+	if err != nil {
+		http.Error(w, "failed to get concept weights", http.StatusInternalServerError)
+		return
+	}
+
+	writeJSON(w, weights)
+}
+
+// handleTriggerRecalculation triggers an immediate score recalculation.
+// POST /api/scoring/recalculate
+func (s *Service) handleTriggerRecalculation(w http.ResponseWriter, r *http.Request) {
+	s.initMu.RLock()
+	recalculator := s.recalculator
+	s.initMu.RUnlock()
+
+	if recalculator == nil {
+		http.Error(w, "recalculator not available", http.StatusServiceUnavailable)
+		return
+	}
+
+	// Run recalculation in background
+	go func() {
+		if recalcErr := recalculator.RecalculateNow(r.Context()); recalcErr != nil {
+			// Log error but don't block response
+			log.Warn().Err(recalcErr).Msg("Failed to trigger score recalculation")
+		}
+	}()
+
+	writeJSON(w, map[string]string{"status": "recalculation triggered"})
+}
+
+// parseIntParam parses an integer query parameter with a default value.
+func parseIntParam(r *http.Request, name string, defaultVal int) int {
+	if val := r.URL.Query().Get(name); val != "" {
+		if parsed, err := strconv.Atoi(val); err == nil && parsed > 0 {
+			return parsed
+		}
+	}
+	return defaultVal
+}
+
+// incrementRetrievalCounts increments retrieval counts for observations.
+// Called after search results are returned to track popularity.
+func (s *Service) incrementRetrievalCounts(ids []int64) {
+	if len(ids) == 0 {
+		return
+	}
+
+	s.initMu.RLock()
+	store := s.observationStore
+	s.initMu.RUnlock()
+
+	if store == nil {
+		return
+	}
+
+	// Increment in background to not block response
+	go func() {
+		// Create a new context with timeout for the background operation
+		ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
+		defer cancel()
+
+		if incrErr := store.IncrementRetrievalCount(ctx, ids); incrErr != nil {
+			// Log but don't fail - this is a background operation
+			log.Warn().Err(incrErr).Msg("Failed to increment retrieval count in background")
+		}
+	}()
+}

internal/worker/handlers_test.go

@@ -1276,8 +1276,6 @@ func TestObservationRequest_Fields(t *testing.T) {
 		ClaudeSessionID: "session-abc",
 		Project:         "my-project",
 		ToolName:        "Read",
-		ToolInput:       map[string]string{"path": "/file.go"},
-		ToolResponse:    "file contents",
 		CWD:             "/home/user/project",
 	}
 

internal/worker/sdk/parser_test.go

@@ -77,10 +77,10 @@ func TestParseObservations_TableDriven(t *testing.T) {
 	tests := []struct {
 		name          string
 		input         string
-		expectedCount int
 		expectedType  models.ObservationType
 		expectedTitle string
 		checkConcepts []string
+		expectedCount int
 	}{
 		{
 			name: "valid_bugfix_observation",
@@ -300,9 +300,9 @@ func TestParseSummary_TableDriven(t *testing.T) {
 	tests := []struct {
 		name            string
 		input           string
+		expectedRequest string
 		sessionID       int64
 		expectNil       bool
-		expectedRequest string
 	}{
 		{
 			name:      "empty_input",

internal/worker/sdk/processor.go

@@ -31,15 +31,14 @@ type SyncSummaryFunc func(summary *models.SessionSummary)
 
 // Processor handles SDK agent processing of observations and summaries using Claude Code CLI.
 type Processor struct {
-	claudePath          string
-	model               string
 	observationStore    *sqlite.ObservationStore
 	summaryStore        *sqlite.SummaryStore
 	broadcastFunc       BroadcastFunc
 	syncObservationFunc SyncObservationFunc
 	syncSummaryFunc     SyncSummaryFunc
-	// Semaphore to limit concurrent Claude CLI calls (prevents API overload)
-	sem chan struct{}
+	sem                 chan struct{}
+	claudePath          string
+	model               string
 }
 
 // SetBroadcastFunc sets the broadcast callback for SSE events.

internal/worker/sdk/processor_test.go

@@ -11,8 +11,8 @@ import (
 
 func TestIsSelfReferentialSummary(t *testing.T) {
 	tests := []struct {
-		name     string
 		summary  *models.ParsedSummary
+		name     string
 		expected bool
 	}{
 		{
@@ -281,8 +281,8 @@ func TestTruncateForLog(t *testing.T) {
 	tests := []struct {
 		name     string
 		input    string
-		maxLen   int
 		expected string
+		maxLen   int
 	}{
 		{
 			name:     "shorter_than_max",
@@ -371,11 +371,11 @@ func TestCaptureFileMtimes(t *testing.T) {
 	file1 := filepath.Join(tmpDir, "file1.txt")
 	file2 := filepath.Join(tmpDir, "file2.txt")
 
-	err = os.WriteFile(file1, []byte("content1"), 0644)
+	err = os.WriteFile(file1, []byte("content1"), 0600)
 	if err != nil {
 		t.Fatal(err)
 	}
-	err = os.WriteFile(file2, []byte("content2"), 0644)
+	err = os.WriteFile(file2, []byte("content2"), 0600)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -415,7 +415,7 @@ func TestGetFileMtimes(t *testing.T) {
 	defer os.RemoveAll(tmpDir)
 
 	testFile := filepath.Join(tmpDir, "test.txt")
-	err = os.WriteFile(testFile, []byte("content"), 0644)
+	err = os.WriteFile(testFile, []byte("content"), 0600)
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -437,7 +437,7 @@ func TestGetFileContent(t *testing.T) {
 	t.Run("reads_existing_file", func(t *testing.T) {
 		testFile := filepath.Join(tmpDir, "test.txt")
 		content := "test content"
-		err := os.WriteFile(testFile, []byte(content), 0644)
+		err := os.WriteFile(testFile, []byte(content), 0600)
 		if err != nil {
 			t.Fatal(err)
 		}
@@ -459,7 +459,7 @@ func TestGetFileContent(t *testing.T) {
 		for i := 0; i < 3000; i++ {
 			longContent += "x"
 		}
-		err := os.WriteFile(testFile, []byte(longContent), 0644)
+		err := os.WriteFile(testFile, []byte(longContent), 0600)
 		if err != nil {
 			t.Fatal(err)
 		}
@@ -473,7 +473,7 @@ func TestGetFileContent(t *testing.T) {
 	t.Run("resolves_relative_path_with_cwd", func(t *testing.T) {
 		testFile := filepath.Join(tmpDir, "relative.txt")
 		content := "relative content"
-		err := os.WriteFile(testFile, []byte(content), 0644)
+		err := os.WriteFile(testFile, []byte(content), 0600)
 		if err != nil {
 			t.Fatal(err)
 		}
@@ -719,8 +719,8 @@ func TestShouldSkipTrivialOperation_EdgeCases(t *testing.T) {
 // TestIsSelfReferentialSummary_MoreCases tests additional self-referential detection cases.
 func TestIsSelfReferentialSummary_MoreCases(t *testing.T) {
 	tests := []struct {
-		name     string
 		summary  *models.ParsedSummary
+		name     string
 		expected bool
 	}{
 		{
@@ -932,7 +932,7 @@ func TestCaptureFileMtimes_DuplicatePaths(t *testing.T) {
 	defer os.RemoveAll(tmpDir)
 
 	testFile := filepath.Join(tmpDir, "shared.txt")
-	err = os.WriteFile(testFile, []byte("content"), 0644)
+	err = os.WriteFile(testFile, []byte("content"), 0600)
 	if err != nil {
 		t.Fatal(err)
 	}

internal/worker/sdk/prompts.go

@@ -24,12 +24,12 @@ var ObservationConcepts = []string{
 
 // ToolExecution represents a tool execution for observation.
 type ToolExecution struct {
-	ID             int64
 	ToolName       string
 	ToolInput      string
 	ToolOutput     string
-	CreatedAtEpoch int64
 	CWD            string
+	ID             int64
+	CreatedAtEpoch int64
 }
 
 // BuildObservationPrompt builds a prompt for processing a tool observation.
@@ -67,12 +67,12 @@ func BuildObservationPrompt(exec ToolExecution) string {
 
 // SummaryRequest contains data for building a summary prompt.
 type SummaryRequest struct {
-	SessionDBID          int64
 	SDKSessionID         string
 	Project              string
 	UserPrompt           string
 	LastUserMessage      string
 	LastAssistantMessage string
+	SessionDBID          int64
 }
 
 // BuildSummaryPrompt builds a prompt requesting a session summary.

internal/worker/sdk/prompts_test.go

@@ -12,8 +12,8 @@ func TestTruncate(t *testing.T) {
 	tests := []struct {
 		name     string
 		input    string
-		maxLen   int
 		expected string
+		maxLen   int
 	}{
 		{
 			name:     "shorter_than_max",
@@ -58,10 +58,10 @@ func TestTruncate(t *testing.T) {
 func TestBuildObservationPrompt(t *testing.T) {
 	now := time.Now().UnixMilli()
 
-	tests := []struct {
-		name     string
+	tests := []struct { //nolint:govet
 		exec     ToolExecution
 		contains []string
+		name     string
 	}{
 		{
 			name: "basic_read_tool",

internal/worker/service.go

@@ -12,10 +12,20 @@ import (
 
 	"github.com/go-chi/chi/v5"
 	"github.com/go-chi/chi/v5/middleware"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/chunking"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/chunking/golang"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/chunking/python"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/chunking/typescript"
 	"github.com/lukaszraczylo/claude-mnemonic/internal/config"
 	"github.com/lukaszraczylo/claude-mnemonic/internal/db/sqlite"
 	"github.com/lukaszraczylo/claude-mnemonic/internal/embedding"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/pattern"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/reranking"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/scoring"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/search/expansion"
 	"github.com/lukaszraczylo/claude-mnemonic/internal/update"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/vector"
+	"github.com/lukaszraczylo/claude-mnemonic/internal/vector/hybrid"
 	"github.com/lukaszraczylo/claude-mnemonic/internal/vector/sqlitevec"
 	"github.com/lukaszraczylo/claude-mnemonic/internal/watcher"
 	"github.com/lukaszraczylo/claude-mnemonic/internal/worker/sdk"
@@ -52,64 +62,52 @@ type RetrievalStats struct {
 
 // Service is the main worker service orchestrator.
 type Service struct {
-	// Version of the worker binary
-	version string
-
-	// Configuration
-	config *config.Config
-
-	// Database
-	store            *sqlite.Store
-	sessionStore     *sqlite.SessionStore
-	observationStore *sqlite.ObservationStore
-	summaryStore     *sqlite.SummaryStore
-	promptStore      *sqlite.PromptStore
-
-	// Domain services
-	sessionManager *session.Manager
-	sseBroadcaster *sse.Broadcaster
-	processor      *sdk.Processor
-
-	// Vector database (sqlite-vec with local embeddings)
-	embedSvc     *embedding.Service
-	vectorClient *sqlitevec.Client
-	vectorSync   *sqlitevec.Sync
-
-	// HTTP server
-	router    *chi.Mux
-	server    *http.Server
-	startTime time.Time
-
-	// Retrieval statistics (per-project)
-	retrievalStats   map[string]*RetrievalStats
-	retrievalStatsMu sync.RWMutex
-
-	// Lifecycle
-	ctx    context.Context
-	cancel context.CancelFunc
-	wg     sync.WaitGroup
-
-	// Initialization state (for deferred init)
-	ready     atomic.Bool
-	initError error
-	initMu    sync.RWMutex
-
-	// Background verification queue for stale observations
-	staleQueue     chan staleVerifyRequest
-	staleQueueOnce sync.Once
-
-	// File watchers for auto-recreation on deletion
-	dbWatcher     *watcher.Watcher
-	configWatcher *watcher.Watcher
-
-	// Self-updater
-	updater *update.Updater
+	startTime          time.Time
+	initError          error
+	vectorClient       vector.Client
+	ctx                context.Context
+	sseBroadcaster     *sse.Broadcaster
+	server             *http.Server
+	graphRebuildTicker *time.Ticker
+	hybridMetrics      *hybrid.Metrics
+	graphSearchClient  *hybrid.GraphSearchClient
+	retrievalStats     map[string]*RetrievalStats
+	staleQueue         chan staleVerifyRequest
+	queryExpander      *expansion.Expander
+	recalculator       *scoring.Recalculator
+	summaryStore       *sqlite.SummaryStore
+	promptStore        *sqlite.PromptStore
+	conflictStore      *sqlite.ConflictStore
+	patternStore       *sqlite.PatternStore
+	relationStore      *sqlite.RelationStore
+	patternDetector    *pattern.Detector
+	sessionManager     *session.Manager
+	router             *chi.Mux
+	config             *config.Config
+	store              *sqlite.Store
+	vectorSync         *sqlitevec.Sync
+	reranker           *reranking.Service
+	updater            *update.Updater
+	observationStore   *sqlite.ObservationStore
+	scoreCalculator    *scoring.Calculator
+	processor          *sdk.Processor
+	dbWatcher          *watcher.Watcher
+	sessionStore       *sqlite.SessionStore
+	configWatcher      *watcher.Watcher
+	embedSvc           *embedding.Service
+	cancel             context.CancelFunc
+	version            string
+	wg                 sync.WaitGroup
+	initMu             sync.RWMutex
+	retrievalStatsMu   sync.RWMutex
+	staleQueueOnce     sync.Once
+	ready              atomic.Bool
 }
 
 // staleVerifyRequest represents a request to verify a stale observation in background
 type staleVerifyRequest struct {
-	observationID int64
 	cwd           string
+	observationID int64
 }
 
 // NewService creates a new worker service with deferred initialization.
@@ -177,33 +175,117 @@ func (s *Service) initializeAsync() {
 	observationStore := sqlite.NewObservationStore(store)
 	summaryStore := sqlite.NewSummaryStore(store)
 	promptStore := sqlite.NewPromptStore(store)
+	conflictStore := sqlite.NewConflictStore(store)
+	patternStore := sqlite.NewPatternStore(store)
+	relationStore := sqlite.NewRelationStore(store)
+
+	// Enable conflict detection by linking stores
+	observationStore.SetConflictStore(conflictStore)
+
+	// Enable relation detection by linking stores
+	observationStore.SetRelationStore(relationStore)
 
 	// Create session manager
 	sessionManager := session.NewManager(sessionStore)
 
 	// Create embedding service and sqlite-vec client for vector search (optional)
 	var embedSvc *embedding.Service
-	var vectorClient *sqlitevec.Client
+	var vectorClient vector.Client
 	var vectorSync *sqlitevec.Sync
 
+	var reranker *reranking.Service
+
 	emb, err := embedding.NewService()
 	if err != nil {
 		log.Warn().Err(err).Msg("Embedding service creation failed - vector search disabled")
 	} else {
 		embedSvc = emb
 		// Create sqlite-vec client using the same DB connection
-		client, err := sqlitevec.NewClient(sqlitevec.Config{
+		baseClient, clientErr := sqlitevec.NewClient(sqlitevec.Config{
 			DB: store.DB(),
 		}, embedSvc)
-		if err != nil {
-			log.Warn().Err(err).Msg("sqlite-vec client creation failed - vector search disabled")
+		if clientErr != nil {
+			log.Warn().Err(clientErr).Msg("sqlite-vec client creation failed - vector search disabled")
 		} else {
-			vectorClient = client
-			vectorSync = sqlitevec.NewSync(client)
+			// Wrap with LEANN hybrid storage client
+			strategy := hybrid.GetStrategyFromEnv()
+			hybridClient := hybrid.NewClient(hybrid.Config{
+				BaseClient:   baseClient,
+				DB:           store.DB(),
+				EmbedSvc:     embedSvc,
+				Strategy:     strategy,
+				HubThreshold: hybrid.GetHubThresholdFromEnv(),
+			})
+
+			// Wrap with graph-aware search client
+			graphConfig := hybrid.GraphConfig{
+				Enabled:      s.config.GraphEnabled,
+				MaxHops:      s.config.GraphMaxHops,
+				BranchFactor: s.config.GraphBranchFactor,
+				EdgeWeight:   s.config.GraphEdgeWeight,
+			}
+			graphClient := hybrid.NewGraphSearchClient(hybridClient, nil, graphConfig)
+			vectorClient = graphClient
+			s.graphSearchClient = graphClient
+			s.hybridMetrics = hybrid.NewMetrics()
+
+			vectorSync = sqlitevec.NewSync(baseClient)
+
+			// Initialize AST-aware code chunking
+			chunkOpts := chunking.DefaultChunkOptions()
+			chunkers := []chunking.Chunker{
+				golang.NewChunker(chunkOpts),
+				python.NewChunker(chunkOpts),
+				typescript.NewChunker(chunkOpts),
+			}
+			chunkingManager := chunking.NewManager(chunkers, chunkOpts)
+			vectorSync.SetChunkingManager(chunkingManager)
+
+			strategyName := "hub" // default
+			switch strategy {
+			case hybrid.StorageAlways:
+				strategyName = "always"
+			case hybrid.StorageOnDemand:
+				strategyName = "on_demand"
+			}
+
 			log.Info().
 				Str("model", embedSvc.Version()).
-				Msg("sqlite-vec vector search enabled")
+				Str("vector_strategy", strategyName).
+				Bool("graph_enabled", s.config.GraphEnabled).
+				Strs("chunkers", []string{"go", "python", "typescript"}).
+				Msg("sqlite-vec vector search with AST-aware code chunking enabled")
+
+			if s.config.GraphEnabled {
+				log.Info().
+					Int("max_hops", s.config.GraphMaxHops).
+					Int("branch_factor", s.config.GraphBranchFactor).
+					Float64("edge_weight", s.config.GraphEdgeWeight).
+					Msg("Graph-aware search configured (graph will be built after initialization)")
+			}
 		}
+
+		// Create cross-encoder reranking service if enabled
+		if s.config.RerankingEnabled {
+			rerankCfg := reranking.DefaultConfig()
+			if s.config.RerankingAlpha > 0 && s.config.RerankingAlpha <= 1 {
+				rerankCfg.Alpha = s.config.RerankingAlpha
+			}
+
+			ranker, rankerErr := reranking.NewService(rerankCfg)
+			if rankerErr != nil {
+				log.Warn().Err(rankerErr).Msg("Cross-encoder reranking service creation failed - reranking disabled")
+			} else {
+				reranker = ranker
+				log.Info().
+					Float64("alpha", rerankCfg.Alpha).
+					Msg("Cross-encoder reranking enabled")
+			}
+		}
+
+		// Create query expander for improved search recall
+		s.queryExpander = expansion.NewExpander(embedSvc)
+		log.Info().Msg("Query expansion enabled")
 	}
 
 	// Create SDK processor (optional - will be nil if Claude CLI not available)
@@ -227,11 +309,38 @@ func (s *Service) initializeAsync() {
 	s.observationStore = observationStore
 	s.summaryStore = summaryStore
 	s.promptStore = promptStore
+	s.conflictStore = conflictStore
+	s.patternStore = patternStore
+	s.relationStore = relationStore
 	s.sessionManager = sessionManager
 	s.processor = processor
 	s.embedSvc = embedSvc
 	s.vectorClient = vectorClient
 	s.vectorSync = vectorSync
+	s.reranker = reranker
+	s.initMu.Unlock()
+
+	// Initialize pattern detector
+	patternDetector := pattern.NewDetector(patternStore, observationStore, pattern.DefaultConfig())
+
+	// Set pattern sync callback if vector sync is available
+	if vectorSync != nil {
+		patternDetector.SetSyncFunc(func(p *models.Pattern) {
+			if err := vectorSync.SyncPattern(s.ctx, p); err != nil {
+				log.Warn().Err(err).Int64("id", p.ID).Msg("Failed to sync pattern to sqlite-vec")
+			}
+		})
+
+		// Set cleanup callback for pattern deletions
+		patternStore.SetCleanupFunc(func(ctx context.Context, deletedIDs []int64) {
+			if err := vectorSync.DeletePatterns(ctx, deletedIDs); err != nil {
+				log.Warn().Err(err).Ints64("ids", deletedIDs).Msg("Failed to delete patterns from sqlite-vec")
+			}
+		})
+	}
+
+	s.initMu.Lock()
+	s.patternDetector = patternDetector
 	s.initMu.Unlock()
 
 	// Set vector sync callbacks on processor if both are available
@@ -240,6 +349,22 @@ func (s *Service) initializeAsync() {
 			if err := vectorSync.SyncObservation(s.ctx, obs); err != nil {
 				log.Warn().Err(err).Int64("id", obs.ID).Msg("Failed to sync observation to sqlite-vec")
 			}
+			// Trigger pattern detection for the new observation
+			if patternDetector != nil {
+				go func(observation *models.Observation) {
+					detectCtx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+					defer cancel()
+					if result, err := patternDetector.AnalyzeObservation(detectCtx, observation); err != nil {
+						log.Warn().Err(err).Int64("obs_id", observation.ID).Msg("Pattern detection failed")
+					} else if result.MatchedPattern != nil {
+						log.Debug().
+							Int64("pattern_id", result.MatchedPattern.ID).
+							Str("pattern_name", result.MatchedPattern.Name).
+							Bool("is_new", result.IsNewPattern).
+							Msg("Pattern matched for observation")
+					}
+				}(obs)
+			}
 		})
 		processor.SetSyncSummaryFunc(func(summary *models.SessionSummary) {
 			if err := vectorSync.SyncSummary(s.ctx, summary); err != nil {
@@ -284,6 +409,37 @@ func (s *Service) initializeAsync() {
 		})
 	})
 
+	// Initialize importance scoring system
+	scoringConfig := models.DefaultScoringConfig()
+
+	// Load concept weights from database if available
+	if weights, err := observationStore.GetConceptWeights(s.ctx); err == nil && len(weights) > 0 {
+		scoringConfig.ConceptWeights = weights
+		log.Info().Int("count", len(weights)).Msg("Loaded concept weights from database")
+	}
+
+	scoreCalculator := scoring.NewCalculator(scoringConfig)
+	recalculator := scoring.NewRecalculator(observationStore, scoreCalculator, log.Logger)
+
+	s.initMu.Lock()
+	s.scoreCalculator = scoreCalculator
+	s.recalculator = recalculator
+	s.initMu.Unlock()
+
+	// Start background recalculator
+	s.wg.Add(1)
+	go func() {
+		defer s.wg.Done()
+		recalculator.Start(s.ctx)
+	}()
+	log.Info().Msg("Importance scoring system initialized")
+
+	// Start pattern detector background analysis
+	if patternDetector != nil {
+		patternDetector.Start()
+		log.Info().Msg("Pattern recognition engine started")
+	}
+
 	// Mark as ready
 	s.ready.Store(true)
 	log.Info().Msg("Async initialization complete - service ready")
@@ -297,6 +453,12 @@ func (s *Service) initializeAsync() {
 	// Start file watchers for auto-recreation on deletion
 	s.startWatchers()
 
+	// Build initial observation graph in background if graph search is enabled
+	if s.config.GraphEnabled && s.graphSearchClient != nil {
+		s.wg.Add(1)
+		go s.buildInitialGraph(observationStore)
+	}
+
 	// Check if vectors need rebuilding (empty or model version mismatch) and trigger background rebuild
 	if vectorClient != nil && vectorSync != nil {
 		needsRebuild, reason := vectorClient.NeedsRebuild(s.ctx)
@@ -330,8 +492,8 @@ func (s *Service) startWatchers() {
 		log.Warn().Err(err).Msg("Failed to create database watcher")
 	} else {
 		s.dbWatcher = dbWatcher
-		if err := dbWatcher.Start(); err != nil {
-			log.Warn().Err(err).Msg("Failed to start database watcher")
+		if startErr := dbWatcher.Start(); startErr != nil {
+			log.Warn().Err(startErr).Msg("Failed to start database watcher")
 		} else {
 			log.Info().Str("path", s.config.DBPath).Msg("Database file watcher started")
 		}
@@ -407,6 +569,15 @@ func (s *Service) reinitializeDatabase() {
 	observationStore := sqlite.NewObservationStore(store)
 	summaryStore := sqlite.NewSummaryStore(store)
 	promptStore := sqlite.NewPromptStore(store)
+	conflictStore := sqlite.NewConflictStore(store)
+	patternStore := sqlite.NewPatternStore(store)
+	relationStore := sqlite.NewRelationStore(store)
+
+	// Enable conflict detection by linking stores
+	observationStore.SetConflictStore(conflictStore)
+
+	// Enable relation detection by linking stores
+	observationStore.SetRelationStore(relationStore)
 
 	// Create new session manager
 	sessionManager := session.NewManager(sessionStore)
@@ -416,21 +587,62 @@ func (s *Service) reinitializeDatabase() {
 	var vectorClient *sqlitevec.Client
 	var vectorSync *sqlitevec.Sync
 
+	var reranker *reranking.Service
+
 	emb, err := embedding.NewService()
 	if err != nil {
 		log.Warn().Err(err).Msg("Embedding service creation failed after reinit")
 	} else {
 		embedSvc = emb
-		client, err := sqlitevec.NewClient(sqlitevec.Config{
+		client, clientErr := sqlitevec.NewClient(sqlitevec.Config{
 			DB: store.DB(),
 		}, embedSvc)
-		if err != nil {
-			log.Warn().Err(err).Msg("sqlite-vec client creation failed after reinit")
+		if clientErr != nil {
+			log.Warn().Err(clientErr).Msg("sqlite-vec client creation failed after reinit")
 		} else {
 			vectorClient = client
 			vectorSync = sqlitevec.NewSync(client)
-			log.Info().Msg("sqlite-vec reconnected after reinit")
+
+			// Initialize AST-aware code chunking
+			chunkOpts := chunking.DefaultChunkOptions()
+			chunkers := []chunking.Chunker{
+				golang.NewChunker(chunkOpts),
+				python.NewChunker(chunkOpts),
+				typescript.NewChunker(chunkOpts),
+			}
+			chunkingManager := chunking.NewManager(chunkers, chunkOpts)
+			vectorSync.SetChunkingManager(chunkingManager)
+
+			log.Info().Msg("sqlite-vec with code chunking reconnected after reinit")
 		}
+
+		// Recreate cross-encoder reranking service if enabled
+		if s.config.RerankingEnabled {
+			rerankCfg := reranking.DefaultConfig()
+			if s.config.RerankingAlpha > 0 && s.config.RerankingAlpha <= 1 {
+				rerankCfg.Alpha = s.config.RerankingAlpha
+			}
+
+			ranker, rankerErr := reranking.NewService(rerankCfg)
+			if rankerErr != nil {
+				log.Warn().Err(rankerErr).Msg("Cross-encoder reranking service creation failed after reinit")
+			} else {
+				reranker = ranker
+				log.Info().Msg("Cross-encoder reranking reconnected after reinit")
+			}
+		}
+
+		// Recreate query expander
+		s.queryExpander = expansion.NewExpander(embedSvc)
+		log.Info().Msg("Query expansion reconnected after reinit")
+	}
+
+	// Close old reranker if exists
+	s.initMu.RLock()
+	oldReranker := s.reranker
+	s.initMu.RUnlock()
+	if oldReranker != nil {
+		_ = oldReranker.Close()
 	}
 
 	// Recreate SDK processor with new stores
@@ -445,6 +657,30 @@ func (s *Service) reinitializeDatabase() {
 		})
 	}
 
+	// Stop old pattern detector if it exists
+	if s.patternDetector != nil {
+		s.patternDetector.Stop()
+	}
+
+	// Create new pattern detector
+	patternDetector := pattern.NewDetector(patternStore, observationStore, pattern.DefaultConfig())
+
+	// Set pattern sync callback if vector sync is available
+	if vectorSync != nil {
+		patternDetector.SetSyncFunc(func(p *models.Pattern) {
+			if err := vectorSync.SyncPattern(s.ctx, p); err != nil {
+				log.Warn().Err(err).Int64("id", p.ID).Msg("Failed to sync pattern to sqlite-vec")
+			}
+		})
+
+		// Set cleanup callback for pattern deletions
+		patternStore.SetCleanupFunc(func(ctx context.Context, deletedIDs []int64) {
+			if err := vectorSync.DeletePatterns(ctx, deletedIDs); err != nil {
+				log.Warn().Err(err).Ints64("ids", deletedIDs).Msg("Failed to delete patterns from sqlite-vec")
+			}
+		})
+	}
+
 	// Atomically swap all components
 	s.initMu.Lock()
 	s.store = store
@@ -452,20 +688,44 @@ func (s *Service) reinitializeDatabase() {
 	s.observationStore = observationStore
 	s.summaryStore = summaryStore
 	s.promptStore = promptStore
+	s.conflictStore = conflictStore
+	s.patternStore = patternStore
+	s.relationStore = relationStore
+	s.patternDetector = patternDetector
 	s.sessionManager = sessionManager
 	s.processor = processor
 	s.embedSvc = embedSvc
 	s.vectorClient = vectorClient
 	s.vectorSync = vectorSync
+	s.reranker = reranker
 	s.initError = nil
 	s.initMu.Unlock()
 
+	// Start pattern detector
+	patternDetector.Start()
+
 	// Set vector sync callbacks on processor if both are available
 	if processor != nil && vectorSync != nil {
 		processor.SetSyncObservationFunc(func(obs *models.Observation) {
 			if err := vectorSync.SyncObservation(s.ctx, obs); err != nil {
 				log.Warn().Err(err).Int64("id", obs.ID).Msg("Failed to sync observation to sqlite-vec")
 			}
+			// Trigger pattern detection for the new observation
+			if patternDetector != nil {
+				go func(observation *models.Observation) {
+					detectCtx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+					defer cancel()
+					if result, err := patternDetector.AnalyzeObservation(detectCtx, observation); err != nil {
+						log.Warn().Err(err).Int64("obs_id", observation.ID).Msg("Pattern detection failed")
+					} else if result.MatchedPattern != nil {
+						log.Debug().
+							Int64("pattern_id", result.MatchedPattern.ID).
+							Str("pattern_name", result.MatchedPattern.Name).
+							Bool("is_new", result.IsNewPattern).
+							Msg("Pattern matched for observation")
+					}
+				}(obs)
+			}
 		})
 		processor.SetSyncSummaryFunc(func(summary *models.SessionSummary) {
 			if err := vectorSync.SyncSummary(s.ctx, summary); err != nil {
@@ -610,8 +870,8 @@ func (s *Service) rebuildAllVectors(
 		log.Error().Err(err).Msg("Failed to fetch observations for vector rebuild")
 	} else {
 		for _, obs := range observations {
-			if err := vectorSync.SyncObservation(s.ctx, obs); err != nil {
-				log.Warn().Err(err).Int64("id", obs.ID).Msg("Failed to sync observation during rebuild")
+			if syncErr := vectorSync.SyncObservation(s.ctx, obs); syncErr != nil {
+				log.Warn().Err(syncErr).Int64("id", obs.ID).Msg("Failed to sync observation during rebuild")
 				syncErrors++
 			} else {
 				totalSynced++
@@ -626,8 +886,8 @@ func (s *Service) rebuildAllVectors(
 		log.Error().Err(err).Msg("Failed to fetch summaries for vector rebuild")
 	} else {
 		for _, summary := range summaries {
-			if err := vectorSync.SyncSummary(s.ctx, summary); err != nil {
-				log.Warn().Err(err).Int64("id", summary.ID).Msg("Failed to sync summary during rebuild")
+			if syncErr := vectorSync.SyncSummary(s.ctx, summary); syncErr != nil {
+				log.Warn().Err(syncErr).Int64("id", summary.ID).Msg("Failed to sync summary during rebuild")
 				syncErrors++
 			} else {
 				totalSynced++
@@ -666,7 +926,7 @@ func (s *Service) rebuildStaleVectors(
 	observationStore *sqlite.ObservationStore,
 	summaryStore *sqlite.SummaryStore,
 	promptStore *sqlite.PromptStore,
-	vectorClient *sqlitevec.Client,
+	vectorClient vector.Client,
 	vectorSync *sqlitevec.Sync,
 ) {
 	defer s.wg.Done()
@@ -831,6 +1091,113 @@ func (s *Service) verifyStaleObservation(req staleVerifyRequest) {
 	}
 }
 
+// buildInitialGraph builds the observation graph from all observations in background.
+func (s *Service) buildInitialGraph(observationStore *sqlite.ObservationStore) {
+	defer s.wg.Done()
+
+	log.Info().Msg("Building initial observation graph...")
+	start := time.Now()
+
+	// Fetch all observations
+	observations, err := observationStore.GetAllObservations(s.ctx)
+	if err != nil {
+		log.Error().Err(err).Msg("Failed to fetch observations for graph building")
+		return
+	}
+
+	if len(observations) == 0 {
+		log.Info().Msg("No observations to build graph from")
+		return
+	}
+
+	// Build graph using RebuildGraph method
+	if err := s.graphSearchClient.RebuildGraph(s.ctx, observations); err != nil {
+		log.Error().Err(err).Msg("Failed to build observation graph")
+		return
+	}
+
+	elapsed := time.Since(start)
+	stats := s.graphSearchClient.GetGraphStats()
+
+	log.Info().
+		Int("observations", len(observations)).
+		Int("nodes", stats.NodeCount).
+		Int("edges", stats.EdgeCount).
+		Float64("avg_degree", stats.AvgDegree).
+		Int("max_degree", stats.MaxDegree).
+		Dur("elapsed", elapsed).
+		Msg("Initial observation graph built successfully")
+
+	// Start periodic graph rebuild if configured
+	if s.config.GraphRebuildIntervalMin > 0 {
+		s.startGraphRebuildTimer(observationStore)
+	}
+}
+
+// startGraphRebuildTimer starts a periodic ticker to rebuild the observation graph.
+func (s *Service) startGraphRebuildTimer(observationStore *sqlite.ObservationStore) {
+	interval := time.Duration(s.config.GraphRebuildIntervalMin) * time.Minute
+	s.graphRebuildTicker = time.NewTicker(interval)
+
+	log.Info().
+		Dur("interval", interval).
+		Msg("Started periodic graph rebuild timer")
+
+	s.wg.Add(1)
+	go func() {
+		defer s.wg.Done()
+		defer s.graphRebuildTicker.Stop()
+
+		for {
+			select {
+			case <-s.ctx.Done():
+				return
+			case <-s.graphRebuildTicker.C:
+				log.Info().Msg("Periodic graph rebuild triggered")
+				s.rebuildGraph(observationStore)
+			}
+		}
+	}()
+}
+
+// rebuildGraph rebuilds the observation graph from current observations.
+func (s *Service) rebuildGraph(observationStore *sqlite.ObservationStore) {
+	if s.graphSearchClient == nil {
+		return
+	}
+
+	start := time.Now()
+
+	// Fetch all observations
+	observations, err := observationStore.GetAllObservations(s.ctx)
+	if err != nil {
+		log.Error().Err(err).Msg("Failed to fetch observations for graph rebuild")
+		return
+	}
+
+	if len(observations) == 0 {
+		log.Debug().Msg("No observations to rebuild graph from")
+		return
+	}
+
+	// Rebuild graph
+	if err := s.graphSearchClient.RebuildGraph(s.ctx, observations); err != nil {
+		log.Error().Err(err).Msg("Failed to rebuild observation graph")
+		return
+	}
+
+	elapsed := time.Since(start)
+	stats := s.graphSearchClient.GetGraphStats()
+
+	log.Info().
+		Int("observations", len(observations)).
+		Int("nodes", stats.NodeCount).
+		Int("edges", stats.EdgeCount).
+		Float64("avg_degree", stats.AvgDegree).
+		Dur("elapsed", elapsed).
+		Msg("Observation graph rebuilt successfully")
+}
+
 // setupMiddleware configures HTTP middleware.
 func (s *Service) setupMiddleware() {
 	s.router.Use(middleware.Logger)
@@ -896,10 +1263,44 @@ func (s *Service) setupRoutes() {
 		r.Get("/api/types", s.handleGetTypes)
 		r.Get("/api/models", s.handleGetModels)
 
+		// Graph and vector metrics routes
+		r.Get("/api/graph/stats", s.handleGetGraphStats)
+		r.Get("/api/vector/metrics", s.handleGetVectorMetrics)
+
+		// Observation scoring and feedback routes
+		r.Post("/api/observations/{id}/feedback", s.handleObservationFeedback)
+		r.Get("/api/observations/{id}/score", s.handleExplainScore)
+		r.Get("/api/observations/top", s.handleGetTopObservations)
+		r.Get("/api/observations/most-retrieved", s.handleGetMostRetrieved)
+
+		// Scoring configuration routes
+		r.Get("/api/scoring/stats", s.handleGetScoringStats)
+		r.Get("/api/scoring/concepts", s.handleGetConceptWeights)
+		r.Put("/api/scoring/concepts/{concept}", s.handleUpdateConceptWeight)
+		r.Post("/api/scoring/recalculate", s.handleTriggerRecalculation)
+
 		// Context injection
 		r.Get("/api/context/count", s.handleContextCount)
 		r.Get("/api/context/inject", s.handleContextInject)
 		r.Get("/api/context/search", s.handleSearchByPrompt)
+
+		// Pattern routes
+		r.Get("/api/patterns", s.handleGetPatterns)
+		r.Get("/api/patterns/stats", s.handleGetPatternStats)
+		r.Get("/api/patterns/search", s.handleSearchPatterns)
+		r.Get("/api/patterns/by-name", s.handleGetPatternByName)
+		r.Get("/api/patterns/{id}", s.handleGetPatternByID)
+		r.Get("/api/patterns/{id}/insight", s.handleGetPatternInsight)
+		r.Delete("/api/patterns/{id}", s.handleDeletePattern)
+		r.Post("/api/patterns/{id}/deprecate", s.handleDeprecatePattern)
+		r.Post("/api/patterns/merge", s.handleMergePatterns)
+
+		// Relation routes (knowledge graph)
+		r.Get("/api/relations/stats", s.handleGetRelationStats)
+		r.Get("/api/relations/type/{type}", s.handleGetRelationsByType)
+		r.Get("/api/observations/{id}/relations", s.handleGetRelations)
+		r.Get("/api/observations/{id}/graph", s.handleGetRelationGraph)
+		r.Get("/api/observations/{id}/related", s.handleGetRelatedObservations)
 	})
 }
 
@@ -1122,6 +1523,21 @@ func (s *Service) Shutdown(ctx context.Context) error {
 		_ = s.configWatcher.Stop()
 	}
 
+	// Stop background recalculator
+	if s.recalculator != nil {
+		s.recalculator.Stop()
+	}
+
+	// Stop pattern detector
+	if s.patternDetector != nil {
+		s.patternDetector.Stop()
+	}
+
+	// Stop graph rebuild ticker
+	if s.graphRebuildTicker != nil {
+		s.graphRebuildTicker.Stop()
+	}
+
 	// Shutdown all sessions
 	s.sessionManager.ShutdownAll(ctx)
 
@@ -1132,6 +1548,13 @@ func (s *Service) Shutdown(ctx context.Context) error {
 		}
 	}
 
+	// Close reranking service
+	if s.reranker != nil {
+		if err := s.reranker.Close(); err != nil {
+			log.Error().Err(err).Msg("Reranking service close error")
+		}
+	}
+
 	// Close embedding service
 	if s.embedSvc != nil {
 		if err := s.embedSvc.Close(); err != nil {

internal/worker/session/manager.go

@@ -21,11 +21,11 @@ const (
 
 // ObservationData contains data for a tool observation.
 type ObservationData struct {
-	ToolName     string
 	ToolInput    interface{}
 	ToolResponse interface{}
-	PromptNumber int
+	ToolName     string
 	CWD          string
+	PromptNumber int
 }
 
 // SummarizeData contains data for a summarize request.
@@ -36,30 +36,28 @@ type SummarizeData struct {
 
 // PendingMessage represents a message queued for SDK processing.
 type PendingMessage struct {
-	Type        MessageType
 	Observation *ObservationData
 	Summarize   *SummarizeData
+	Type        MessageType
 }
 
 // ActiveSession represents an in-memory active session being processed.
 type ActiveSession struct {
-	SessionDBID            int64
-	ClaudeSessionID        string
-	SDKSessionID           string
+	StartTime              time.Time
+	ctx                    context.Context
+	cancel                 context.CancelFunc
+	notify                 chan struct{}
 	Project                string
 	UserPrompt             string
+	SDKSessionID           string
+	ClaudeSessionID        string
+	pendingMessages        []PendingMessage
 	LastPromptNumber       int
-	StartTime              time.Time
 	CumulativeInputTokens  int64
 	CumulativeOutputTokens int64
-
-	// Concurrency control
-	pendingMessages []PendingMessage
-	messageMu       sync.Mutex
-	notify          chan struct{}
-	ctx             context.Context
-	cancel          context.CancelFunc
-	generatorActive atomic.Bool
+	SessionDBID            int64
+	messageMu              sync.Mutex
+	generatorActive        atomic.Bool
 }
 
 // SessionTimeout is how long an inactive session can exist before cleanup.
@@ -70,15 +68,14 @@ const CleanupInterval = 5 * time.Minute
 
 // Manager manages active session lifecycles.
 type Manager struct {
-	sessionStore *sqlite.SessionStore
-	sessions     map[int64]*ActiveSession
-	mu           sync.RWMutex
-	onCreated    func(int64)
-	onDeleted    func(int64)
-	ctx          context.Context
-	cancel       context.CancelFunc
-	// Global notification channel for immediate processing
+	ctx           context.Context
+	sessionStore  *sqlite.SessionStore
+	sessions      map[int64]*ActiveSession
+	onCreated     func(int64)
+	onDeleted     func(int64)
+	cancel        context.CancelFunc
 	ProcessNotify chan struct{}
+	mu            sync.RWMutex
 }
 
 // NewManager creates a new session manager.

internal/worker/session/manager_test.go

@@ -47,9 +47,9 @@ func (s *ManagerSuite) TestActiveSession() {
 		SDKSessionID:    "sdk-123",
 		Project:         "test-project",
 		UserPrompt:      "Hello",
-		StartTime:       time.Now(),
-		pendingMessages: make([]PendingMessage, 0),
-		notify:          make(chan struct{}, 1),
+		StartTime:       time.Now(),                //nolint:govet
+		pendingMessages: make([]PendingMessage, 0), //nolint:govet
+		notify:          make(chan struct{}, 1),    //nolint:govet
 	}
 
 	s.Equal(int64(1), session.SessionDBID)
@@ -134,7 +134,7 @@ func (s *ManagerSuite) TestDeleteSession() {
 	session := &ActiveSession{
 		SessionDBID:     1,
 		Project:         "test-project",
-		StartTime:       time.Now(),
+		StartTime:       time.Now(), //nolint:govet
 		pendingMessages: []PendingMessage{},
 		ctx:             ctx,
 		cancel:          cancel,
@@ -236,8 +236,8 @@ func TestTimeoutConstants(t *testing.T) {
 func TestObservationData(t *testing.T) {
 	data := ObservationData{
 		ToolName:     "Read",
-		ToolInput:    map[string]string{"path": "/test/file.go"},
-		ToolResponse: "file content",
+		ToolInput:    map[string]string{"path": "/test/file.go"}, //nolint:govet
+		ToolResponse: "file content",                             //nolint:govet
 		PromptNumber: 1,
 		CWD:          "/test",
 	}
@@ -333,7 +333,7 @@ func TestConcurrentSessionAccess(t *testing.T) {
 // TestProcessNotifyChannel tests the process notification channel.
 func TestProcessNotifyChannel(t *testing.T) {
 	manager := &Manager{
-		sessions:      make(map[int64]*ActiveSession),
+		sessions:      make(map[int64]*ActiveSession), //nolint:govet
 		ProcessNotify: make(chan struct{}, 1),
 	}
 
@@ -367,7 +367,7 @@ func TestActiveSessionContext(t *testing.T) {
 	ctx, cancel := context.WithCancel(context.Background())
 
 	session := &ActiveSession{
-		SessionDBID: 1,
+		SessionDBID: 1, //nolint:govet
 		ctx:         ctx,
 		cancel:      cancel,
 	}
@@ -438,7 +438,7 @@ func (s *ManagerSuite) TestShutdownAll() {
 		s.manager.sessions[i] = &ActiveSession{
 			SessionDBID:     i,
 			Project:         "test-project",
-			StartTime:       time.Now(),
+			StartTime:       time.Now(), //nolint:govet
 			pendingMessages: []PendingMessage{},
 			ctx:             ctx,
 			cancel:          cancel,
@@ -479,7 +479,7 @@ func (s *ManagerSuite) TestDeleteNonExistentSession() {
 // TestLastPromptNumber tests prompt number tracking.
 func TestLastPromptNumber(t *testing.T) {
 	session := &ActiveSession{
-		SessionDBID:      1,
+		SessionDBID:      1, //nolint:govet
 		LastPromptNumber: 0,
 	}
 
@@ -526,7 +526,7 @@ func TestActiveSessionNotifyChannel(t *testing.T) {
 // TestMessageMutex tests message mutex operations.
 func TestMessageMutex(t *testing.T) {
 	session := &ActiveSession{
-		pendingMessages: make([]PendingMessage, 0),
+		pendingMessages: make([]PendingMessage, 0), //nolint:govet
 	}
 
 	var wg sync.WaitGroup
@@ -559,7 +559,7 @@ func (s *ManagerSuite) TestQueueDepthMultipleSessions() {
 	}
 	s.manager.sessions[2] = &ActiveSession{
 		SessionDBID:     2,
-		pendingMessages: make([]PendingMessage, 0),
+		pendingMessages: make([]PendingMessage, 0), //nolint:govet
 	}
 	s.manager.sessions[3] = &ActiveSession{
 		SessionDBID:     3,
@@ -658,7 +658,7 @@ func TestActiveSessionCWD(t *testing.T) {
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			data := ObservationData{
-				ToolName: "Test",
+				ToolName: "Test", //nolint:govet
 				CWD:      tt.cwd,
 			}
 			assert.Equal(t, tt.cwd, data.CWD)
@@ -669,22 +669,22 @@ func TestActiveSessionCWD(t *testing.T) {
 // TestToolInputResponse tests various tool input/response types.
 func TestToolInputResponse(t *testing.T) {
 	tests := []struct {
-		name     string
 		input    interface{}
 		response interface{}
+		name     string
 	}{
-		{"nil_values", nil, nil},
-		{"string_values", "input string", "response string"},
-		{"map_values", map[string]string{"key": "value"}, map[string]interface{}{"result": true}},
-		{"slice_values", []string{"a", "b"}, []int{1, 2, 3}},
-		{"int_values", 42, 100},
-		{"bool_values", true, false},
+		{name: "nil_values", input: nil, response: nil},
+		{name: "string_values", input: "input string", response: "response string"},
+		{name: "map_values", input: map[string]string{"key": "value"}, response: map[string]interface{}{"result": true}},
+		{name: "slice_values", input: []string{"a", "b"}, response: []int{1, 2, 3}},
+		{name: "int_values", input: 42, response: 100},
+		{name: "bool_values", input: true, response: false},
 	}
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			data := ObservationData{
-				ToolName:     "TestTool",
+				ToolName:     "TestTool", //nolint:govet
 				ToolInput:    tt.input,
 				ToolResponse: tt.response,
 			}

internal/worker/sse/broadcaster.go

@@ -19,10 +19,10 @@ const (
 
 // Client represents a connected SSE client.
 type Client struct {
-	ID      string
 	Writer  http.ResponseWriter
 	Flusher http.Flusher
 	Done    chan struct{}
+	ID      string
 }
 
 // Broadcaster manages SSE client connections and message broadcasting.

internal/worker/sse/broadcaster_test.go

@@ -256,8 +256,8 @@ func TestHandleSSE(t *testing.T) {
 // TestBroadcastJSON tests broadcasting various JSON types.
 func TestBroadcastJSON(t *testing.T) {
 	tests := []struct {
-		name    string
 		data    interface{}
+		name    string
 		wantErr bool
 	}{
 		{

pkg/hooks/response.go

@@ -62,11 +62,11 @@ type BaseInput struct {
 // HookContext provides common context for hook handlers.
 type HookContext struct {
 	HookName  string
-	Port      int
 	Project   string
 	SessionID string
 	CWD       string
 	RawInput  []byte
+	Port      int
 }
 
 // HookHandler is a function that handles hook-specific logic.
@@ -92,8 +92,8 @@ func RunHook[T any](hookName string, handler HookHandler[T]) {
 
 	// Parse input
 	var input T
-	if err := json.Unmarshal(inputData, &input); err != nil {
-		WriteError(hookName, err)
+	if unmarshalErr := json.Unmarshal(inputData, &input); unmarshalErr != nil {
+		WriteError(hookName, unmarshalErr)
 		os.Exit(1)
 	}
 

pkg/hooks/worker_test.go

@@ -320,11 +320,11 @@ func TestExtractBaseVersion(t *testing.T) {
 // TestPOST tests the POST function with a mock server.
 func TestPOST(t *testing.T) {
 	tests := []struct {
-		name           string
-		serverHandler  func(w http.ResponseWriter, r *http.Request)
 		body           interface{}
-		expectError    bool
+		serverHandler  func(w http.ResponseWriter, r *http.Request)
 		expectedResult map[string]interface{}
+		name           string
+		expectError    bool
 	}{
 		{
 			name: "successful POST with JSON response",
@@ -393,10 +393,10 @@ func TestPOST(t *testing.T) {
 // TestGET tests the GET function with a mock server.
 func TestGET(t *testing.T) {
 	tests := []struct {
-		name           string
 		serverHandler  func(w http.ResponseWriter, r *http.Request)
-		expectError    bool
 		expectedResult map[string]interface{}
+		name           string
+		expectError    bool
 	}{
 		{
 			name: "successful GET with JSON response",
@@ -532,8 +532,8 @@ func TestExitCodes(t *testing.T) {
 func TestHookResponse(t *testing.T) {
 	tests := []struct {
 		name     string
-		response HookResponse
 		expected string
+		response HookResponse
 	}{
 		{
 			name:     "continue true",
@@ -597,8 +597,8 @@ func TestHookContext(t *testing.T) {
 // TestIsWorkerRunning_WithServer tests IsWorkerRunning with actual server.
 func TestIsWorkerRunning_WithServer(t *testing.T) {
 	tests := []struct {
-		name           string
 		serverHandler  func(w http.ResponseWriter, r *http.Request)
+		name           string
 		expectedResult bool
 	}{
 		{
@@ -828,8 +828,8 @@ func TestBaseInput_PartialFields(t *testing.T) {
 func TestHookResponse_Marshal(t *testing.T) {
 	tests := []struct {
 		name     string
-		response HookResponse
 		contains []string
+		response HookResponse
 	}{
 		{
 			name:     "continue true",
@@ -1177,7 +1177,7 @@ func TestHookContext_RawInput(t *testing.T) {
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			ctx := HookContext{
-				HookName: "test",
+				HookName: "test", //nolint:govet
 				RawInput: tt.rawInput,
 			}
 			assert.Equal(t, tt.rawInput, ctx.RawInput)

pkg/models/conflict.go

@@ -0,0 +1,258 @@
+// Package models contains domain models for claude-mnemonic.
+package models
+
+import (
+	"regexp"
+	"strings"
+	"time"
+)
+
+// ConflictType represents the type of conflict between observations.
+type ConflictType string
+
+const (
+	// ConflictSuperseded means newer observation supersedes older one (same topic, updated info).
+	ConflictSuperseded ConflictType = "superseded"
+	// ConflictContradicts means observations contain contradictory information.
+	ConflictContradicts ConflictType = "contradicts"
+	// ConflictOutdatedPattern means an outdated pattern/practice was identified.
+	ConflictOutdatedPattern ConflictType = "outdated_pattern"
+)
+
+// ConflictResolution indicates which observation to prefer.
+type ConflictResolution string
+
+const (
+	// ResolutionPreferNewer means prefer the newer observation.
+	ResolutionPreferNewer ConflictResolution = "prefer_newer"
+	// ResolutionPreferOlder means prefer the older observation (rare).
+	ResolutionPreferOlder ConflictResolution = "prefer_older"
+	// ResolutionManual means manual review is needed.
+	ResolutionManual ConflictResolution = "manual"
+)
+
+// ObservationConflict tracks conflicting observations.
+type ObservationConflict struct {
+	ResolvedAt      *string            `db:"resolved_at" json:"resolved_at,omitempty"`
+	ConflictType    ConflictType       `db:"conflict_type" json:"conflict_type"`
+	Resolution      ConflictResolution `db:"resolution" json:"resolution"`
+	Reason          string             `db:"reason" json:"reason"`
+	DetectedAt      string             `db:"detected_at" json:"detected_at"`
+	ID              int64              `db:"id" json:"id"`
+	NewerObsID      int64              `db:"newer_obs_id" json:"newer_obs_id"`
+	OlderObsID      int64              `db:"older_obs_id" json:"older_obs_id"`
+	DetectedAtEpoch int64              `db:"detected_at_epoch" json:"detected_at_epoch"`
+	Resolved        bool               `db:"resolved" json:"resolved"`
+}
+
+// ConflictDetectionResult contains the result of conflict detection.
+type ConflictDetectionResult struct {
+	Type        ConflictType
+	Resolution  ConflictResolution
+	Reason      string
+	OlderObsIDs []int64
+	HasConflict bool
+}
+
+// NewObservationConflict creates a new conflict record.
+func NewObservationConflict(newerID, olderID int64, conflictType ConflictType, resolution ConflictResolution, reason string) *ObservationConflict {
+	now := time.Now()
+	return &ObservationConflict{
+		NewerObsID:      newerID,
+		OlderObsID:      olderID,
+		ConflictType:    conflictType,
+		Resolution:      resolution,
+		Reason:          reason,
+		DetectedAt:      now.Format(time.RFC3339),
+		DetectedAtEpoch: now.UnixMilli(),
+		Resolved:        false,
+	}
+}
+
+// CorrectionPatterns contains regex patterns that indicate explicit corrections.
+var CorrectionPatterns = []*regexp.Regexp{
+	regexp.MustCompile(`(?i)\bactually[,\s]+that\s+was\s+wrong\b`),
+	regexp.MustCompile(`(?i)\bactually[,\s]+that's\s+(wrong|incorrect|not\s+right)\b`),
+	regexp.MustCompile(`(?i)\bpreviously\s+(said|mentioned|noted)\s+.*\s+but\b`),
+	regexp.MustCompile(`(?i)\bcorrection:\s*`),
+	regexp.MustCompile(`(?i)\bignore\s+(the\s+)?(previous|earlier)\b`),
+	regexp.MustCompile(`(?i)\bdisregard\s+(the\s+)?(previous|earlier)\b`),
+	regexp.MustCompile(`(?i)\bwas\s+(wrong|incorrect|mistaken)\b`),
+	regexp.MustCompile(`(?i)\bturns\s+out\s+.*(wrong|incorrect|not\s+the\s+case)\b`),
+	regexp.MustCompile(`(?i)\b(supersedes|replaces|overrides)\s+(the\s+)?(previous|earlier|old)\b`),
+	regexp.MustCompile(`(?i)\b(don't|do\s+not)\s+use\s+.*\s+anymore\b`),
+	regexp.MustCompile(`(?i)\bno\s+longer\s+(valid|applicable|correct|recommended)\b`),
+	regexp.MustCompile(`(?i)\bdeprecated\s+(approach|method|pattern|way)\b`),
+	regexp.MustCompile(`(?i)\bshould\s+have\s+(been|used)\b.*instead\b`),
+	regexp.MustCompile(`(?i)\bbetter\s+(approach|way|method|solution)\s+is\b`),
+}
+
+// OpposingChangePatterns detects add/remove conflicts.
+var OpposingChangePatterns = map[string]string{
+	"add":     "remove",
+	"added":   "removed",
+	"create":  "delete",
+	"created": "deleted",
+	"enable":  "disable",
+	"enabled": "disabled",
+	"include": "exclude",
+	"allow":   "deny",
+	"permit":  "block",
+}
+
+// DetectExplicitCorrection checks if text contains explicit correction language.
+func DetectExplicitCorrection(text string) (bool, string) {
+	for _, pattern := range CorrectionPatterns {
+		if match := pattern.FindString(text); match != "" {
+			return true, "Explicit correction detected: " + match
+		}
+	}
+	return false, ""
+}
+
+// DetectOpposingFileChanges checks if two observations have opposing changes on the same file.
+func DetectOpposingFileChanges(newer, older *Observation) (bool, string) {
+	// Check for overlapping modified files
+	newerFiles := make(map[string]bool)
+	for _, f := range newer.FilesModified {
+		newerFiles[f] = true
+	}
+
+	var overlappingFiles []string
+	for _, f := range older.FilesModified {
+		if newerFiles[f] {
+			overlappingFiles = append(overlappingFiles, f)
+		}
+	}
+
+	if len(overlappingFiles) == 0 {
+		return false, ""
+	}
+
+	// Check for opposing action words in titles/narratives
+	newerText := strings.ToLower(newer.Title.String + " " + newer.Narrative.String)
+	olderText := strings.ToLower(older.Title.String + " " + older.Narrative.String)
+
+	for action, opposite := range OpposingChangePatterns {
+		if (strings.Contains(newerText, action) && strings.Contains(olderText, opposite)) ||
+			(strings.Contains(newerText, opposite) && strings.Contains(olderText, action)) {
+			return true, "Opposing changes on files: " + strings.Join(overlappingFiles, ", ")
+		}
+	}
+
+	return false, ""
+}
+
+// DetectConceptTagMismatch checks if observations have same concepts but different recommendations.
+func DetectConceptTagMismatch(newer, older *Observation) (bool, string) {
+	// Find overlapping concepts
+	newerConcepts := make(map[string]bool)
+	for _, c := range newer.Concepts {
+		newerConcepts[c] = true
+	}
+
+	var overlapping []string
+	for _, c := range older.Concepts {
+		if newerConcepts[c] {
+			overlapping = append(overlapping, c)
+		}
+	}
+
+	if len(overlapping) == 0 {
+		return false, ""
+	}
+
+	// Check if same file was modified and concepts overlap
+	// This suggests the newer observation may update the approach
+	newerFiles := make(map[string]bool)
+	for _, f := range newer.FilesModified {
+		newerFiles[f] = true
+	}
+	for _, f := range older.FilesModified {
+		if newerFiles[f] {
+			// Same file modified with same concepts - likely an update
+			return true, "Same concepts (" + strings.Join(overlapping, ", ") + ") with overlapping file changes"
+		}
+	}
+
+	return false, ""
+}
+
+// DetectConflict performs comprehensive conflict detection between a new observation
+// and an existing one. Returns detection result.
+func DetectConflict(newer, older *Observation) *ConflictDetectionResult {
+	result := &ConflictDetectionResult{
+		HasConflict: false,
+	}
+
+	// 1. Check for explicit correction language in newer observation
+	if newer.Narrative.Valid {
+		if isCorrection, reason := DetectExplicitCorrection(newer.Narrative.String); isCorrection {
+			result.HasConflict = true
+			result.Type = ConflictContradicts
+			result.Resolution = ResolutionPreferNewer
+			result.Reason = reason
+			result.OlderObsIDs = append(result.OlderObsIDs, older.ID)
+			return result
+		}
+	}
+
+	// Check title as well
+	if newer.Title.Valid {
+		if isCorrection, reason := DetectExplicitCorrection(newer.Title.String); isCorrection {
+			result.HasConflict = true
+			result.Type = ConflictContradicts
+			result.Resolution = ResolutionPreferNewer
+			result.Reason = reason
+			result.OlderObsIDs = append(result.OlderObsIDs, older.ID)
+			return result
+		}
+	}
+
+	// 2. Check for opposing file changes
+	if isOpposing, reason := DetectOpposingFileChanges(newer, older); isOpposing {
+		result.HasConflict = true
+		result.Type = ConflictSuperseded
+		result.Resolution = ResolutionPreferNewer
+		result.Reason = reason
+		result.OlderObsIDs = append(result.OlderObsIDs, older.ID)
+		return result
+	}
+
+	// 3. Check for concept tag mismatches with same files
+	if isMismatch, reason := DetectConceptTagMismatch(newer, older); isMismatch {
+		result.HasConflict = true
+		result.Type = ConflictSuperseded
+		result.Resolution = ResolutionPreferNewer
+		result.Reason = reason
+		result.OlderObsIDs = append(result.OlderObsIDs, older.ID)
+		return result
+	}
+
+	return result
+}
+
+// DetectConflictsWithExisting checks a new observation against a list of existing observations.
+// Returns all detected conflicts.
+func DetectConflictsWithExisting(newer *Observation, existing []*Observation) []*ConflictDetectionResult {
+	var results []*ConflictDetectionResult
+
+	for _, older := range existing {
+		// Skip self-comparison
+		if older.ID == newer.ID {
+			continue
+		}
+
+		// Only compare within same project (or both global)
+		if newer.Project != older.Project && newer.Scope != ScopeGlobal && older.Scope != ScopeGlobal {
+			continue
+		}
+
+		result := DetectConflict(newer, older)
+		if result.HasConflict {
+			results = append(results, result)
+		}
+	}
+
+	return results
+}

pkg/models/conflict_test.go

@@ -0,0 +1,421 @@
+// Package models contains domain models for claude-mnemonic.
+package models
+
+import (
+	"database/sql"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/suite"
+)
+
+// ConflictSuite is a test suite for conflict detection operations.
+type ConflictSuite struct {
+	suite.Suite
+}
+
+func TestConflictSuite(t *testing.T) {
+	suite.Run(t, new(ConflictSuite))
+}
+
+// TestConflictTypeConstants tests conflict type constants.
+func (s *ConflictSuite) TestConflictTypeConstants() {
+	s.Equal(ConflictType("superseded"), ConflictSuperseded)
+	s.Equal(ConflictType("contradicts"), ConflictContradicts)
+	s.Equal(ConflictType("outdated_pattern"), ConflictOutdatedPattern)
+}
+
+// TestResolutionConstants tests resolution constants.
+func (s *ConflictSuite) TestResolutionConstants() {
+	s.Equal(ConflictResolution("prefer_newer"), ResolutionPreferNewer)
+	s.Equal(ConflictResolution("prefer_older"), ResolutionPreferOlder)
+	s.Equal(ConflictResolution("manual"), ResolutionManual)
+}
+
+// TestNewObservationConflict tests conflict creation.
+func (s *ConflictSuite) TestNewObservationConflict() {
+	conflict := NewObservationConflict(2, 1, ConflictSuperseded, ResolutionPreferNewer, "Test reason")
+
+	s.Equal(int64(2), conflict.NewerObsID)
+	s.Equal(int64(1), conflict.OlderObsID)
+	s.Equal(ConflictSuperseded, conflict.ConflictType)
+	s.Equal(ResolutionPreferNewer, conflict.Resolution)
+	s.Equal("Test reason", conflict.Reason)
+	s.False(conflict.Resolved)
+	s.NotEmpty(conflict.DetectedAt)
+	s.Greater(conflict.DetectedAtEpoch, int64(0))
+}
+
+// TestDetectExplicitCorrection_TableDriven tests explicit correction detection.
+func (s *ConflictSuite) TestDetectExplicitCorrection_TableDriven() {
+	tests := []struct {
+		name          string
+		text          string
+		expectPattern string
+		expectMatch   bool
+	}{
+		{
+			name:          "actually that was wrong",
+			text:          "Actually, that was wrong - we should use a different approach",
+			expectMatch:   true,
+			expectPattern: "actually, that was wrong",
+		},
+		{
+			name:          "correction prefix",
+			text:          "Correction: the previous implementation had a bug",
+			expectMatch:   true,
+			expectPattern: "correction:",
+		},
+		{
+			name:          "ignore previous",
+			text:          "Please ignore the previous recommendation",
+			expectMatch:   true,
+			expectPattern: "ignore",
+		},
+		{
+			name:          "disregard earlier",
+			text:          "Disregard the earlier suggestion, it was flawed",
+			expectMatch:   true,
+			expectPattern: "disregard",
+		},
+		{
+			name:          "was wrong",
+			text:          "The original approach was wrong",
+			expectMatch:   true,
+			expectPattern: "was wrong",
+		},
+		{
+			name:          "no longer valid",
+			text:          "This method is no longer valid after the refactor",
+			expectMatch:   true,
+			expectPattern: "no longer valid",
+		},
+		{
+			name:          "deprecated approach",
+			text:          "This is a deprecated approach that should not be used",
+			expectMatch:   true,
+			expectPattern: "deprecated approach",
+		},
+		{
+			name:          "better approach is",
+			text:          "A better approach is to use the new API",
+			expectMatch:   true,
+			expectPattern: "better approach is",
+		},
+		{
+			name:        "normal text - no correction",
+			text:        "This is a normal observation about the code",
+			expectMatch: false,
+		},
+		{
+			name:        "empty text",
+			text:        "",
+			expectMatch: false,
+		},
+	}
+
+	for _, tt := range tests {
+		s.Run(tt.name, func() {
+			found, reason := DetectExplicitCorrection(tt.text)
+			s.Equal(tt.expectMatch, found)
+			if tt.expectMatch {
+				s.Contains(reason, "Explicit correction detected")
+			}
+		})
+	}
+}
+
+// TestDetectOpposingFileChanges_TableDriven tests opposing file change detection.
+func (s *ConflictSuite) TestDetectOpposingFileChanges_TableDriven() {
+	tests := []struct {
+		newerObs       *Observation
+		olderObs       *Observation
+		name           string
+		expectConflict bool
+	}{
+		{
+			name: "add then remove - conflict",
+			newerObs: &Observation{
+				Title:         sql.NullString{String: "Remove authentication middleware", Valid: true},
+				Narrative:     sql.NullString{String: "Removed the auth middleware from handlers", Valid: true},
+				FilesModified: []string{"middleware.go", "handler.go"},
+			},
+			olderObs: &Observation{
+				Title:         sql.NullString{String: "Add authentication middleware", Valid: true},
+				Narrative:     sql.NullString{String: "Added auth middleware to secure endpoints", Valid: true},
+				FilesModified: []string{"middleware.go", "handler.go"},
+			},
+			expectConflict: true,
+		},
+		{
+			name: "enable then disable - conflict",
+			newerObs: &Observation{
+				Title:         sql.NullString{String: "Disable caching feature", Valid: true},
+				Narrative:     sql.NullString{String: "Disabled caching due to issues", Valid: true},
+				FilesModified: []string{"cache.go"},
+			},
+			olderObs: &Observation{
+				Title:         sql.NullString{String: "Enable caching feature", Valid: true},
+				Narrative:     sql.NullString{String: "Enabled caching for performance", Valid: true},
+				FilesModified: []string{"cache.go"},
+			},
+			expectConflict: true,
+		},
+		{
+			name: "different files - no conflict",
+			newerObs: &Observation{
+				Title:         sql.NullString{String: "Remove old code", Valid: true},
+				Narrative:     sql.NullString{String: "Removed deprecated functions", Valid: true},
+				FilesModified: []string{"old.go"},
+			},
+			olderObs: &Observation{
+				Title:         sql.NullString{String: "Add new feature", Valid: true},
+				Narrative:     sql.NullString{String: "Added new functions", Valid: true},
+				FilesModified: []string{"new.go"},
+			},
+			expectConflict: false,
+		},
+		{
+			name: "same files but no opposing keywords - no conflict",
+			newerObs: &Observation{
+				Title:         sql.NullString{String: "Update handler logic", Valid: true},
+				Narrative:     sql.NullString{String: "Updated the handler implementation", Valid: true},
+				FilesModified: []string{"handler.go"},
+			},
+			olderObs: &Observation{
+				Title:         sql.NullString{String: "Fix handler bug", Valid: true},
+				Narrative:     sql.NullString{String: "Fixed a bug in handler", Valid: true},
+				FilesModified: []string{"handler.go"},
+			},
+			expectConflict: false,
+		},
+	}
+
+	for _, tt := range tests {
+		s.Run(tt.name, func() {
+			found, _ := DetectOpposingFileChanges(tt.newerObs, tt.olderObs)
+			s.Equal(tt.expectConflict, found)
+		})
+	}
+}
+
+// TestDetectConceptTagMismatch_TableDriven tests concept tag mismatch detection.
+func (s *ConflictSuite) TestDetectConceptTagMismatch_TableDriven() {
+	tests := []struct {
+		newerObs       *Observation
+		olderObs       *Observation
+		name           string
+		expectConflict bool
+	}{
+		{
+			name: "same concepts and files - conflict",
+			newerObs: &Observation{
+				Concepts:      []string{"security", "authentication"},
+				FilesModified: []string{"auth.go"},
+			},
+			olderObs: &Observation{
+				Concepts:      []string{"security", "authentication"},
+				FilesModified: []string{"auth.go"},
+			},
+			expectConflict: true,
+		},
+		{
+			name: "overlapping concepts different files - no conflict",
+			newerObs: &Observation{
+				Concepts:      []string{"security"},
+				FilesModified: []string{"new_auth.go"},
+			},
+			olderObs: &Observation{
+				Concepts:      []string{"security"},
+				FilesModified: []string{"old_auth.go"},
+			},
+			expectConflict: false,
+		},
+		{
+			name: "different concepts same files - no conflict",
+			newerObs: &Observation{
+				Concepts:      []string{"performance"},
+				FilesModified: []string{"handler.go"},
+			},
+			olderObs: &Observation{
+				Concepts:      []string{"testing"},
+				FilesModified: []string{"handler.go"},
+			},
+			expectConflict: false,
+		},
+		{
+			name: "no overlapping concepts or files - no conflict",
+			newerObs: &Observation{
+				Concepts:      []string{"security"},
+				FilesModified: []string{"auth.go"},
+			},
+			olderObs: &Observation{
+				Concepts:      []string{"testing"},
+				FilesModified: []string{"test.go"},
+			},
+			expectConflict: false,
+		},
+	}
+
+	for _, tt := range tests {
+		s.Run(tt.name, func() {
+			found, _ := DetectConceptTagMismatch(tt.newerObs, tt.olderObs)
+			s.Equal(tt.expectConflict, found)
+		})
+	}
+}
+
+// TestDetectConflict tests comprehensive conflict detection.
+func (s *ConflictSuite) TestDetectConflict() {
+	// Test explicit correction takes precedence
+	newer := &Observation{
+		ID:        2,
+		Project:   "test",
+		Narrative: sql.NullString{String: "Actually, that was wrong. We should use a different approach.", Valid: true},
+	}
+	older := &Observation{
+		ID:      1,
+		Project: "test",
+	}
+
+	result := DetectConflict(newer, older)
+	s.True(result.HasConflict)
+	s.Equal(ConflictContradicts, result.Type)
+	s.Equal(ResolutionPreferNewer, result.Resolution)
+	s.Contains(result.Reason, "Explicit correction")
+}
+
+// TestDetectConflictsWithExisting tests conflict detection against multiple observations.
+func (s *ConflictSuite) TestDetectConflictsWithExisting() {
+	newer := &Observation{
+		ID:            3,
+		Project:       "test",
+		Narrative:     sql.NullString{String: "Actually, that was wrong", Valid: true},
+		Concepts:      []string{"security"},
+		FilesModified: []string{"auth.go"},
+	}
+
+	existing := []*Observation{
+		{
+			ID:            1,
+			Project:       "test",
+			Concepts:      []string{"security"},
+			FilesModified: []string{"auth.go"},
+		},
+		{
+			ID:            2,
+			Project:       "test",
+			Concepts:      []string{"testing"},
+			FilesModified: []string{"test.go"},
+		},
+		{
+			ID:      3, // Same as newer - should be skipped
+			Project: "test",
+		},
+	}
+
+	results := DetectConflictsWithExisting(newer, existing)
+
+	// Should find conflicts with obs 1 (concepts + files overlap + correction language)
+	// but not with obs 2 (different concepts/files) or obs 3 (same ID)
+	s.GreaterOrEqual(len(results), 1)
+
+	// At least one result should reference obs 1
+	foundObs1 := false
+	for _, r := range results {
+		for _, id := range r.OlderObsIDs {
+			if id == 1 {
+				foundObs1 = true
+				break
+			}
+		}
+	}
+	s.True(foundObs1)
+}
+
+// TestDetectConflictsWithExisting_DifferentProjects tests that different projects don't conflict.
+func (s *ConflictSuite) TestDetectConflictsWithExisting_DifferentProjects() {
+	newer := &Observation{
+		ID:        2,
+		Project:   "project-a",
+		Scope:     ScopeProject,
+		Narrative: sql.NullString{String: "Actually, that was wrong", Valid: true},
+	}
+
+	existing := []*Observation{
+		{
+			ID:      1,
+			Project: "project-b",
+			Scope:   ScopeProject,
+		},
+	}
+
+	results := DetectConflictsWithExisting(newer, existing)
+	s.Empty(results) // Different projects should not conflict
+}
+
+// TestDetectConflictsWithExisting_GlobalScope tests that global observations can conflict across projects.
+func (s *ConflictSuite) TestDetectConflictsWithExisting_GlobalScope() {
+	newer := &Observation{
+		ID:            2,
+		Project:       "project-a",
+		Scope:         ScopeGlobal,
+		Narrative:     sql.NullString{String: "Actually, that was wrong", Valid: true},
+		Concepts:      []string{"security"},
+		FilesModified: []string{"auth.go"},
+	}
+
+	existing := []*Observation{
+		{
+			ID:            1,
+			Project:       "project-b",
+			Scope:         ScopeGlobal, // Global scope allows cross-project conflict detection
+			Concepts:      []string{"security"},
+			FilesModified: []string{"auth.go"},
+		},
+	}
+
+	results := DetectConflictsWithExisting(newer, existing)
+	s.GreaterOrEqual(len(results), 1) // Global scope allows conflict detection
+}
+
+// TestCorrectionPatterns_Compiled ensures all patterns compile correctly.
+func TestCorrectionPatterns_Compiled(t *testing.T) {
+	// This test verifies that all correction patterns are valid regexps
+	// If any pattern fails to compile, the package won't load
+	assert.NotEmpty(t, CorrectionPatterns)
+	for i, pattern := range CorrectionPatterns {
+		assert.NotNil(t, pattern, "Pattern %d should not be nil", i)
+	}
+}
+
+// TestOpposingChangePatterns tests the opposing change pattern map.
+func TestOpposingChangePatterns(t *testing.T) {
+	assert.NotEmpty(t, OpposingChangePatterns)
+	assert.Equal(t, "remove", OpposingChangePatterns["add"])
+	assert.Equal(t, "removed", OpposingChangePatterns["added"])
+	assert.Equal(t, "delete", OpposingChangePatterns["create"])
+	assert.Equal(t, "disable", OpposingChangePatterns["enable"])
+}
+
+// TestObservationConflict_Fields tests field access.
+func TestObservationConflict_Fields(t *testing.T) {
+	conflict := &ObservationConflict{
+		ID:              1,
+		NewerObsID:      10,
+		OlderObsID:      5,
+		ConflictType:    ConflictSuperseded,
+		Resolution:      ResolutionPreferNewer,
+		Reason:          "Test reason",          //nolint:govet
+		DetectedAt:      "2024-01-01T00:00:00Z", //nolint:govet
+		DetectedAtEpoch: 1704067200000,          //nolint:govet
+		Resolved:        false,
+	}
+
+	assert.Equal(t, int64(1), conflict.ID)
+	assert.Equal(t, int64(10), conflict.NewerObsID)
+	assert.Equal(t, int64(5), conflict.OlderObsID)
+	assert.Equal(t, ConflictSuperseded, conflict.ConflictType)
+	assert.Equal(t, ResolutionPreferNewer, conflict.Resolution)
+	assert.False(t, conflict.Resolved)
+}

pkg/models/observation.go

@@ -121,38 +121,44 @@ func (j JSONInt64Map) Value() (driver.Value, error) {
 
 // Observation represents a learning extracted from a Claude Code session.
 type Observation struct {
-	ID              int64            `db:"id" json:"id"`
+	FileMtimes      JSONInt64Map     `db:"file_mtimes" json:"file_mtimes,omitempty"`
 	SDKSessionID    string           `db:"sdk_session_id" json:"sdk_session_id"`
 	Project         string           `db:"project" json:"project"`
 	Scope           ObservationScope `db:"scope" json:"scope"`
 	Type            ObservationType  `db:"type" json:"type"`
-	Title           sql.NullString   `db:"title" json:"title,omitempty"`
+	CreatedAt       string           `db:"created_at" json:"created_at"`
 	Subtitle        sql.NullString   `db:"subtitle" json:"subtitle,omitempty"`
-	Facts           JSONStringArray  `db:"facts" json:"facts,omitempty"`
+	Title           sql.NullString   `db:"title" json:"title,omitempty"`
 	Narrative       sql.NullString   `db:"narrative" json:"narrative,omitempty"`
 	Concepts        JSONStringArray  `db:"concepts" json:"concepts,omitempty"`
 	FilesRead       JSONStringArray  `db:"files_read" json:"files_read,omitempty"`
 	FilesModified   JSONStringArray  `db:"files_modified" json:"files_modified,omitempty"`
-	FileMtimes      JSONInt64Map     `db:"file_mtimes" json:"file_mtimes,omitempty"`
+	Facts           JSONStringArray  `db:"facts" json:"facts,omitempty"`
 	PromptNumber    sql.NullInt64    `db:"prompt_number" json:"prompt_number,omitempty"`
+	LastRetrievedAt sql.NullInt64    `db:"last_retrieved_at_epoch" json:"last_retrieved_at_epoch,omitempty"`
+	ScoreUpdatedAt  sql.NullInt64    `db:"score_updated_at_epoch" json:"score_updated_at_epoch,omitempty"`
 	DiscoveryTokens int64            `db:"discovery_tokens" json:"discovery_tokens"`
-	CreatedAt       string           `db:"created_at" json:"created_at"`
+	ID              int64            `db:"id" json:"id"`
 	CreatedAtEpoch  int64            `db:"created_at_epoch" json:"created_at_epoch"`
+	ImportanceScore float64          `db:"importance_score" json:"importance_score"`
+	UserFeedback    int              `db:"user_feedback" json:"user_feedback"`
+	RetrievalCount  int              `db:"retrieval_count" json:"retrieval_count"`
 	IsStale         bool             `db:"-" json:"is_stale,omitempty"`
+	IsSuperseded    bool             `db:"is_superseded" json:"is_superseded,omitempty"`
 }
 
 // ParsedObservation represents an observation parsed from SDK response XML.
 type ParsedObservation struct {
+	FileMtimes    map[string]int64
 	Type          ObservationType
 	Title         string
 	Subtitle      string
-	Facts         []string
 	Narrative     string
+	Scope         ObservationScope
+	Facts         []string
 	Concepts      []string
 	FilesRead     []string
 	FilesModified []string
-	FileMtimes    map[string]int64 // File path -> mtime epoch ms
-	Scope         ObservationScope // Optional: if empty, will be auto-determined
 }
 
 // ToStoredObservation converts a ParsedObservation to the stored Observation format.
@@ -187,24 +193,30 @@ func DetermineScope(concepts []string) ObservationScope {
 // ObservationJSON is a JSON-friendly representation of Observation.
 // It converts sql.NullString to plain strings for clean JSON output.
 type ObservationJSON struct {
-	ID              int64            `json:"id"`
+	FileMtimes      map[string]int64 `json:"file_mtimes,omitempty"`
+	Subtitle        string           `json:"subtitle,omitempty"`
 	SDKSessionID    string           `json:"sdk_session_id"`
-	Project         string           `json:"project"`
 	Scope           ObservationScope `json:"scope"`
 	Type            ObservationType  `json:"type"`
 	Title           string           `json:"title,omitempty"`
-	Subtitle        string           `json:"subtitle,omitempty"`
-	Facts           []string         `json:"facts,omitempty"`
+	CreatedAt       string           `json:"created_at"`
 	Narrative       string           `json:"narrative,omitempty"`
+	Project         string           `json:"project"`
 	Concepts        []string         `json:"concepts,omitempty"`
+	Facts           []string         `json:"facts,omitempty"`
 	FilesRead       []string         `json:"files_read,omitempty"`
 	FilesModified   []string         `json:"files_modified,omitempty"`
-	FileMtimes      map[string]int64 `json:"file_mtimes,omitempty"`
-	PromptNumber    int64            `json:"prompt_number,omitempty"`
-	DiscoveryTokens int64            `json:"discovery_tokens"`
-	CreatedAt       string           `json:"created_at"`
 	CreatedAtEpoch  int64            `json:"created_at_epoch"`
+	DiscoveryTokens int64            `json:"discovery_tokens"`
+	ID              int64            `json:"id"`
+	PromptNumber    int64            `json:"prompt_number,omitempty"`
+	ImportanceScore float64          `json:"importance_score"`
+	UserFeedback    int              `json:"user_feedback"`
+	RetrievalCount  int              `json:"retrieval_count"`
+	LastRetrievedAt int64            `json:"last_retrieved_at_epoch,omitempty"`
+	ScoreUpdatedAt  int64            `json:"score_updated_at_epoch,omitempty"`
 	IsStale         bool             `json:"is_stale,omitempty"`
+	IsSuperseded    bool             `json:"is_superseded,omitempty"`
 }
 
 // MarshalJSON implements json.Marshaler for Observation.
@@ -225,6 +237,12 @@ func (o *Observation) MarshalJSON() ([]byte, error) {
 		CreatedAt:       o.CreatedAt,
 		CreatedAtEpoch:  o.CreatedAtEpoch,
 		IsStale:         o.IsStale,
+		// Importance scoring fields
+		ImportanceScore: o.ImportanceScore,
+		UserFeedback:    o.UserFeedback,
+		RetrievalCount:  o.RetrievalCount,
+		// Conflict detection fields
+		IsSuperseded: o.IsSuperseded,
 	}
 	if o.Title.Valid {
 		j.Title = o.Title.String
@@ -238,6 +256,12 @@ func (o *Observation) MarshalJSON() ([]byte, error) {
 	if o.PromptNumber.Valid {
 		j.PromptNumber = o.PromptNumber.Int64
 	}
+	if o.LastRetrievedAt.Valid {
+		j.LastRetrievedAt = o.LastRetrievedAt.Int64
+	}
+	if o.ScoreUpdatedAt.Valid {
+		j.ScoreUpdatedAt = o.ScoreUpdatedAt.Int64
+	}
 	return json.Marshal(j)
 }
 
@@ -268,9 +292,28 @@ func NewObservation(sdkSessionID, project string, parsed *ParsedObservation, pro
 		DiscoveryTokens: discoveryTokens,
 		CreatedAt:       now.Format(time.RFC3339),
 		CreatedAtEpoch:  now.UnixMilli(),
+		// Importance scoring: new observations start with score 1.0
+		ImportanceScore: 1.0,
+		UserFeedback:    0,
+		RetrievalCount:  0,
 	}
 }
 
+// ToMap converts the observation to a map for JSON response building.
+// This allows adding extra fields like similarity scores.
+func (o *Observation) ToMap() map[string]interface{} {
+	// Marshal to JSON then unmarshal to map (uses MarshalJSON for proper conversion)
+	data, err := json.Marshal(o)
+	if err != nil {
+		return map[string]interface{}{"id": o.ID, "error": err.Error()}
+	}
+	var result map[string]interface{}
+	if err := json.Unmarshal(data, &result); err != nil {
+		return map[string]interface{}{"id": o.ID, "error": err.Error()}
+	}
+	return result
+}
+
 // CheckStaleness checks if an observation is stale based on current file mtimes.
 // Returns true if any tracked file has been modified since the observation was created.
 func (o *Observation) CheckStaleness(currentMtimes map[string]int64) bool {

pkg/models/observation_test.go

@@ -50,8 +50,8 @@ func (s *ObservationSuite) TestGlobalizableConcepts() {
 func (s *ObservationSuite) TestDetermineScope_TableDriven() {
 	tests := []struct {
 		name     string
-		concepts []string
 		expected ObservationScope
+		concepts []string
 	}{
 		{
 			name:     "empty concepts - project scope",
@@ -106,8 +106,8 @@ func (s *ObservationSuite) TestDetermineScope_TableDriven() {
 // TestParsedObservation_FileMtimesJSON tests FileMtimes JSON serialization.
 func (s *ObservationSuite) TestParsedObservation_FileMtimesJSON() {
 	obs := &ParsedObservation{
-		Type:       ObsTypeDiscovery,
-		Title:      "Test",
+		Type:       ObsTypeDiscovery, //nolint:govet
+		Title:      "Test",           //nolint:govet
 		FileMtimes: map[string]int64{"file1.go": 1234567890, "file2.go": 1234567891},
 	}
 
@@ -121,9 +121,9 @@ func (s *ObservationSuite) TestParsedObservation_FileMtimesJSON() {
 // TestObservation_CheckStaleness_TableDriven tests staleness checking.
 func (s *ObservationSuite) TestObservation_CheckStaleness_TableDriven() {
 	tests := []struct {
-		name          string
 		storedMtimes  map[string]int64
 		currentMtimes map[string]int64
+		name          string
 		expectedStale bool
 	}{
 		{
@@ -221,9 +221,9 @@ func (s *ObservationSuite) TestParsedObservation_Fields() {
 func (s *ObservationSuite) TestObservation_NullFields() {
 	// Test with null fields
 	obs := &Observation{
-		ID:        1,
-		Project:   "test",
-		Type:      ObsTypeDiscovery,
+		ID:        1,                //nolint:govet
+		Project:   "test",           //nolint:govet
+		Type:      ObsTypeDiscovery, //nolint:govet
 		Title:     sql.NullString{Valid: false},
 		Subtitle:  sql.NullString{Valid: false},
 		Narrative: sql.NullString{Valid: false},
@@ -235,9 +235,9 @@ func (s *ObservationSuite) TestObservation_NullFields() {
 
 	// Test with valid fields
 	obs2 := &Observation{
-		ID:        2,
-		Project:   "test",
-		Type:      ObsTypeBugfix,
+		ID:        2,             //nolint:govet
+		Project:   "test",        //nolint:govet
+		Type:      ObsTypeBugfix, //nolint:govet
 		Title:     sql.NullString{String: "Fix bug", Valid: true},
 		Subtitle:  sql.NullString{String: "Memory leak", Valid: true},
 		Narrative: sql.NullString{String: "Fixed memory leak in handler", Valid: true},
@@ -300,10 +300,10 @@ func TestParsedObservation_ToStoredObservation(t *testing.T) {
 // TestJSONStringArray tests JSONStringArray scanning.
 func TestJSONStringArray(t *testing.T) {
 	tests := []struct {
-		name     string
 		input    interface{}
-		wantErr  bool
+		name     string
 		expected JSONStringArray
+		wantErr  bool
 	}{
 		{
 			name:     "nil input",
@@ -348,10 +348,10 @@ func TestJSONStringArray(t *testing.T) {
 // TestJSONInt64Map tests JSONInt64Map scanning.
 func TestJSONInt64Map(t *testing.T) {
 	tests := []struct {
-		name     string
 		input    interface{}
-		wantErr  bool
 		expected JSONInt64Map
+		name     string
+		wantErr  bool
 	}{
 		{
 			name:     "nil input",

pkg/models/pattern.go

@@ -0,0 +1,398 @@
+// Package models contains domain models for claude-mnemonic.
+package models
+
+import (
+	"database/sql"
+	"database/sql/driver"
+	"encoding/json"
+	"time"
+)
+
+// PatternType represents the category of detected pattern.
+type PatternType string
+
+const (
+	// PatternTypeBug represents recurring bug patterns (e.g., "nil handling oversight").
+	PatternTypeBug PatternType = "bug"
+	// PatternTypeRefactor represents recurring refactoring approaches (e.g., "interface extraction").
+	PatternTypeRefactor PatternType = "refactor"
+	// PatternTypeArchitecture represents consistent architectural patterns.
+	PatternTypeArchitecture PatternType = "architecture"
+	// PatternTypeAntiPattern represents identified anti-patterns to avoid.
+	PatternTypeAntiPattern PatternType = "anti-pattern"
+	// PatternTypeBestPractice represents best practices that work consistently.
+	PatternTypeBestPractice PatternType = "best-practice"
+)
+
+// PatternStatus represents the lifecycle status of a pattern.
+type PatternStatus string
+
+const (
+	// PatternStatusActive means the pattern is actively being tracked and can be referenced.
+	PatternStatusActive PatternStatus = "active"
+	// PatternStatusDeprecated means the pattern has been superseded or is no longer relevant.
+	PatternStatusDeprecated PatternStatus = "deprecated"
+	// PatternStatusMerged means this pattern was merged into another pattern.
+	PatternStatusMerged PatternStatus = "merged"
+)
+
+// Pattern represents a recurring pattern detected across observations.
+// This enables Claude to reference historical insights: "I've encountered this pattern 12 times."
+type Pattern struct {
+	Status         PatternStatus   `db:"status" json:"status"`
+	Name           string          `db:"name" json:"name"`
+	Type           PatternType     `db:"type" json:"type"`
+	CreatedAt      string          `db:"created_at" json:"created_at"`
+	LastSeenAt     string          `db:"last_seen_at" json:"last_seen_at"`
+	Signature      JSONStringArray `db:"signature" json:"signature"`
+	Projects       JSONStringArray `db:"projects" json:"projects"`
+	ObservationIDs JSONInt64Array  `db:"observation_ids" json:"observation_ids"`
+	Recommendation sql.NullString  `db:"recommendation" json:"recommendation"`
+	Description    sql.NullString  `db:"description" json:"description"`
+	MergedIntoID   sql.NullInt64   `db:"merged_into_id" json:"merged_into_id,omitempty"`
+	Frequency      int             `db:"frequency" json:"frequency"`
+	Confidence     float64         `db:"confidence" json:"confidence"`
+	ID             int64           `db:"id" json:"id"`
+	LastSeenEpoch  int64           `db:"last_seen_at_epoch" json:"last_seen_at_epoch"`
+	CreatedAtEpoch int64           `db:"created_at_epoch" json:"created_at_epoch"`
+}
+
+// JSONInt64Array is a custom type for handling JSON int64 arrays in SQLite.
+type JSONInt64Array []int64
+
+// Scan implements sql.Scanner for JSONInt64Array.
+func (j *JSONInt64Array) Scan(src interface{}) error {
+	if src == nil {
+		*j = nil
+		return nil
+	}
+
+	var data []byte
+	switch v := src.(type) {
+	case string:
+		data = []byte(v)
+	case []byte:
+		data = v
+	default:
+		return nil
+	}
+
+	if len(data) == 0 {
+		*j = nil
+		return nil
+	}
+
+	return json.Unmarshal(data, j)
+}
+
+// Value implements driver.Valuer for JSONInt64Array.
+func (j JSONInt64Array) Value() (driver.Value, error) {
+	if j == nil {
+		return nil, nil
+	}
+	return json.Marshal(j)
+}
+
+// PatternJSON is a JSON-friendly representation of Pattern.
+type PatternJSON struct {
+	Status         PatternStatus `json:"status"`
+	Name           string        `json:"name"`
+	Type           PatternType   `json:"type"`
+	Description    string        `json:"description,omitempty"`
+	CreatedAt      string        `json:"created_at"`
+	Recommendation string        `json:"recommendation,omitempty"`
+	LastSeenAt     string        `json:"last_seen_at"`
+	Signature      []string      `json:"signature,omitempty"`
+	ObservationIDs []int64       `json:"observation_ids,omitempty"`
+	Projects       []string      `json:"projects,omitempty"`
+	MergedIntoID   int64         `json:"merged_into_id,omitempty"`
+	Confidence     float64       `json:"confidence"`
+	Frequency      int           `json:"frequency"`
+	LastSeenEpoch  int64         `json:"last_seen_at_epoch"`
+	ID             int64         `json:"id"`
+	CreatedAtEpoch int64         `json:"created_at_epoch"`
+}
+
+// MarshalJSON implements json.Marshaler for Pattern.
+func (p *Pattern) MarshalJSON() ([]byte, error) {
+	j := PatternJSON{
+		ID:             p.ID,
+		Name:           p.Name,
+		Type:           p.Type,
+		Signature:      p.Signature,
+		Frequency:      p.Frequency,
+		Projects:       p.Projects,
+		ObservationIDs: p.ObservationIDs,
+		Status:         p.Status,
+		Confidence:     p.Confidence,
+		LastSeenAt:     p.LastSeenAt,
+		LastSeenEpoch:  p.LastSeenEpoch,
+		CreatedAt:      p.CreatedAt,
+		CreatedAtEpoch: p.CreatedAtEpoch,
+	}
+	if p.Description.Valid {
+		j.Description = p.Description.String
+	}
+	if p.Recommendation.Valid {
+		j.Recommendation = p.Recommendation.String
+	}
+	if p.MergedIntoID.Valid {
+		j.MergedIntoID = p.MergedIntoID.Int64
+	}
+	return json.Marshal(j)
+}
+
+// NewPattern creates a new pattern from detected data.
+func NewPattern(name string, patternType PatternType, description string, signature []string, project string, observationID int64) *Pattern {
+	now := time.Now()
+	return &Pattern{
+		Name:           name,
+		Type:           patternType,
+		Description:    sql.NullString{String: description, Valid: description != ""},
+		Signature:      signature,
+		Frequency:      1,
+		Projects:       []string{project},
+		ObservationIDs: []int64{observationID},
+		Status:         PatternStatusActive,
+		Confidence:     0.5, // Initial confidence
+		LastSeenAt:     now.Format(time.RFC3339),
+		LastSeenEpoch:  now.UnixMilli(),
+		CreatedAt:      now.Format(time.RFC3339),
+		CreatedAtEpoch: now.UnixMilli(),
+	}
+}
+
+// AddOccurrence records a new occurrence of this pattern.
+func (p *Pattern) AddOccurrence(project string, observationID int64) {
+	p.Frequency++
+
+	// Add project if not already tracked
+	found := false
+	for _, proj := range p.Projects {
+		if proj == project {
+			found = true
+			break
+		}
+	}
+	if !found {
+		p.Projects = append(p.Projects, project)
+	}
+
+	// Add observation ID if not already tracked
+	for _, id := range p.ObservationIDs {
+		if id == observationID {
+			return
+		}
+	}
+	p.ObservationIDs = append(p.ObservationIDs, observationID)
+
+	// Update confidence based on frequency and cross-project occurrence
+	p.updateConfidence()
+
+	// Update last seen timestamp
+	now := time.Now()
+	p.LastSeenAt = now.Format(time.RFC3339)
+	p.LastSeenEpoch = now.UnixMilli()
+}
+
+// updateConfidence adjusts confidence based on frequency and cross-project validation.
+func (p *Pattern) updateConfidence() {
+	// Base confidence from frequency (logarithmic scaling)
+	freqConfidence := 0.3 + (0.4 * (float64(min(p.Frequency, 10)) / 10.0))
+
+	// Cross-project bonus: patterns seen across multiple projects are more reliable
+	projectBonus := 0.0
+	if len(p.Projects) >= 2 {
+		projectBonus = 0.1
+	}
+	if len(p.Projects) >= 5 {
+		projectBonus = 0.2
+	}
+
+	p.Confidence = min(1.0, freqConfidence+projectBonus)
+}
+
+// PatternMatch represents a match between an observation and a potential pattern.
+type PatternMatch struct {
+	MatchedOn     string  `json:"matched_on"`
+	SuggestedName string  `json:"suggested_name,omitempty"`
+	PatternID     int64   `json:"pattern_id"`
+	Score         float64 `json:"score"`
+	IsNew         bool    `json:"is_new"`
+}
+
+// PatternSignatureKeywords are common keywords used in pattern detection.
+var PatternSignatureKeywords = map[PatternType][]string{
+	PatternTypeBug: {
+		"nil", "null", "undefined", "panic", "crash", "error handling",
+		"race condition", "deadlock", "memory leak", "overflow",
+		"off-by-one", "boundary", "timeout", "concurrency",
+	},
+	PatternTypeRefactor: {
+		"extract", "inline", "rename", "move", "split", "merge",
+		"interface", "abstraction", "decouple", "simplify",
+		"consolidate", "modularize", "encapsulate",
+	},
+	PatternTypeArchitecture: {
+		"layer", "service", "repository", "controller", "handler",
+		"middleware", "dependency injection", "factory", "singleton",
+		"observer", "strategy", "adapter", "facade", "builder",
+	},
+	PatternTypeAntiPattern: {
+		"god class", "spaghetti", "copy paste", "magic number",
+		"hardcoded", "circular dependency", "premature optimization",
+		"over-engineering", "feature envy", "data clump",
+	},
+	PatternTypeBestPractice: {
+		"test", "validation", "logging", "monitoring", "documentation",
+		"error handling", "retry", "timeout", "circuit breaker",
+		"graceful shutdown", "health check", "metrics",
+	},
+}
+
+// DetectPatternType analyzes concepts and content to determine pattern type.
+func DetectPatternType(concepts []string, title, narrative string) PatternType {
+	// Check concepts first
+	for _, concept := range concepts {
+		switch concept {
+		case "anti-pattern":
+			return PatternTypeAntiPattern
+		case "best-practice":
+			return PatternTypeBestPractice
+		case "architecture":
+			return PatternTypeArchitecture
+		case "refactor":
+			return PatternTypeRefactor
+		}
+	}
+
+	// Check for bug-related patterns in content
+	content := title + " " + narrative
+	for _, keyword := range PatternSignatureKeywords[PatternTypeBug] {
+		if containsIgnoreCase(content, keyword) {
+			return PatternTypeBug
+		}
+	}
+
+	// Default to refactor for other patterns
+	return PatternTypeRefactor
+}
+
+// containsIgnoreCase checks if text contains substr (case-insensitive).
+func containsIgnoreCase(text, substr string) bool {
+	textLower := toLower(text)
+	substrLower := toLower(substr)
+	return contains(textLower, substrLower)
+}
+
+// Simple implementations to avoid strings package dependency in this file
+func toLower(s string) string {
+	b := make([]byte, len(s))
+	for i := 0; i < len(s); i++ {
+		c := s[i]
+		if 'A' <= c && c <= 'Z' {
+			c += 'a' - 'A'
+		}
+		b[i] = c
+	}
+	return string(b)
+}
+
+func contains(s, substr string) bool {
+	if len(substr) > len(s) {
+		return false
+	}
+	for i := 0; i <= len(s)-len(substr); i++ {
+		if s[i:i+len(substr)] == substr {
+			return true
+		}
+	}
+	return false
+}
+
+// ExtractSignature creates a signature from observation content.
+func ExtractSignature(concepts []string, title, narrative string) []string {
+	var signature []string
+
+	// Add all concepts
+	signature = append(signature, concepts...)
+
+	// Extract key terms from title (simple word extraction)
+	for _, word := range splitWords(title) {
+		if len(word) > 3 && isSignificantWord(word) {
+			signature = append(signature, toLower(word))
+		}
+	}
+
+	return uniqueStrings(signature)
+}
+
+// splitWords is a simple word splitter.
+func splitWords(s string) []string {
+	var words []string
+	word := ""
+	for _, r := range s {
+		if r == ' ' || r == '-' || r == '_' || r == '.' || r == ',' {
+			if word != "" {
+				words = append(words, word)
+				word = ""
+			}
+		} else {
+			word += string(r)
+		}
+	}
+	if word != "" {
+		words = append(words, word)
+	}
+	return words
+}
+
+// isSignificantWord filters out common stop words.
+func isSignificantWord(word string) bool {
+	stopWords := map[string]bool{
+		"the": true, "and": true, "for": true, "with": true, "that": true,
+		"this": true, "from": true, "have": true, "not": true, "are": true,
+		"was": true, "but": true, "all": true, "can": true, "had": true,
+		"were": true, "been": true, "will": true, "when": true, "what": true,
+	}
+	return !stopWords[toLower(word)]
+}
+
+// uniqueStrings returns a slice with duplicate strings removed.
+func uniqueStrings(s []string) []string {
+	seen := make(map[string]bool)
+	var result []string
+	for _, v := range s {
+		if !seen[v] {
+			seen[v] = true
+			result = append(result, v)
+		}
+	}
+	return result
+}
+
+// CalculateMatchScore computes similarity between two signatures.
+func CalculateMatchScore(sig1, sig2 []string) float64 {
+	if len(sig1) == 0 || len(sig2) == 0 {
+		return 0.0
+	}
+
+	set1 := make(map[string]bool)
+	for _, s := range sig1 {
+		set1[toLower(s)] = true
+	}
+
+	matches := 0
+	for _, s := range sig2 {
+		if set1[toLower(s)] {
+			matches++
+		}
+	}
+
+	// Jaccard similarity
+	unionSize := len(sig1) + len(sig2) - matches
+	if unionSize == 0 {
+		return 0.0
+	}
+	return float64(matches) / float64(unionSize)
+}

pkg/models/pattern_test.go

@@ -0,0 +1,357 @@
+package models
+
+import (
+	"database/sql"
+	"encoding/json"
+	"testing"
+	"time"
+)
+
+func TestNewPattern(t *testing.T) {
+	pattern := NewPattern(
+		"Test Pattern",
+		PatternTypeBug,
+		"A test pattern description",
+		[]string{"nil", "error", "handling"},
+		"test-project",
+		123,
+	)
+
+	if pattern.Name != "Test Pattern" {
+		t.Errorf("Expected name 'Test Pattern', got '%s'", pattern.Name)
+	}
+	if pattern.Type != PatternTypeBug {
+		t.Errorf("Expected type PatternTypeBug, got '%s'", pattern.Type)
+	}
+	if !pattern.Description.Valid || pattern.Description.String != "A test pattern description" {
+		t.Errorf("Description not set correctly")
+	}
+	if len(pattern.Signature) != 3 {
+		t.Errorf("Expected 3 signature elements, got %d", len(pattern.Signature))
+	}
+	if pattern.Frequency != 1 {
+		t.Errorf("Expected frequency 1, got %d", pattern.Frequency)
+	}
+	if len(pattern.Projects) != 1 || pattern.Projects[0] != "test-project" {
+		t.Errorf("Projects not set correctly")
+	}
+	if len(pattern.ObservationIDs) != 1 || pattern.ObservationIDs[0] != 123 {
+		t.Errorf("ObservationIDs not set correctly")
+	}
+	if pattern.Status != PatternStatusActive {
+		t.Errorf("Expected status Active, got '%s'", pattern.Status)
+	}
+	if pattern.Confidence != 0.5 {
+		t.Errorf("Expected initial confidence 0.5, got %f", pattern.Confidence)
+	}
+}
+
+func TestPattern_AddOccurrence(t *testing.T) {
+	pattern := NewPattern("Test", PatternTypeBug, "desc", []string{"test"}, "project1", 1)
+
+	// Add same project occurrence
+	pattern.AddOccurrence("project1", 2)
+	if pattern.Frequency != 2 {
+		t.Errorf("Expected frequency 2, got %d", pattern.Frequency)
+	}
+	if len(pattern.Projects) != 1 {
+		t.Errorf("Expected 1 project (no duplicates), got %d", len(pattern.Projects))
+	}
+
+	// Add different project occurrence
+	pattern.AddOccurrence("project2", 3)
+	if pattern.Frequency != 3 {
+		t.Errorf("Expected frequency 3, got %d", pattern.Frequency)
+	}
+	if len(pattern.Projects) != 2 {
+		t.Errorf("Expected 2 projects, got %d", len(pattern.Projects))
+	}
+
+	// Add duplicate observation ID - should not duplicate
+	pattern.AddOccurrence("project2", 3)
+	if len(pattern.ObservationIDs) != 3 {
+		t.Errorf("Expected 3 observation IDs (no duplicate), got %d", len(pattern.ObservationIDs))
+	}
+
+	// Check confidence increased
+	if pattern.Confidence <= 0.5 {
+		t.Errorf("Expected confidence to increase above 0.5, got %f", pattern.Confidence)
+	}
+}
+
+func TestPattern_ConfidenceCalculation(t *testing.T) {
+	tests := []struct {
+		name          string
+		frequency     int
+		projectCount  int
+		minConfidence float64
+		maxConfidence float64
+	}{
+		{"low_frequency", 2, 1, 0.3, 0.5},
+		{"high_frequency", 10, 1, 0.6, 0.8},
+		{"multi_project", 3, 3, 0.4, 0.7},
+		{"high_freq_multi_proj", 10, 5, 0.7, 1.0},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			pattern := NewPattern("Test", PatternTypeBug, "", []string{}, "proj1", 1)
+
+			// Simulate occurrences
+			for i := 1; i < tt.frequency; i++ {
+				projIdx := i % tt.projectCount
+				if projIdx == 0 {
+					projIdx = 1
+				}
+				pattern.AddOccurrence("proj"+string(rune('0'+projIdx)), int64(i+1))
+			}
+
+			if pattern.Confidence < tt.minConfidence || pattern.Confidence > tt.maxConfidence {
+				t.Errorf("Expected confidence between %f and %f, got %f",
+					tt.minConfidence, tt.maxConfidence, pattern.Confidence)
+			}
+		})
+	}
+}
+
+func TestPatternType_Detection(t *testing.T) {
+	tests := []struct {
+		title     string
+		narrative string
+		expected  PatternType
+		concepts  []string
+	}{
+		{concepts: []string{"anti-pattern"}, title: "", narrative: "", expected: PatternTypeAntiPattern},
+		{concepts: []string{"best-practice"}, title: "", narrative: "", expected: PatternTypeBestPractice},
+		{concepts: []string{"architecture"}, title: "", narrative: "", expected: PatternTypeArchitecture},
+		{concepts: []string{"refactor"}, title: "", narrative: "", expected: PatternTypeRefactor},
+		{concepts: []string{}, title: "nil pointer bug", narrative: "", expected: PatternTypeBug},
+		{concepts: []string{}, title: "Deadlock in concurrent code", narrative: "", expected: PatternTypeBug},
+		{concepts: []string{}, title: "Extract interface", narrative: "", expected: PatternTypeRefactor},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.title+"_"+tt.expected.String(), func(t *testing.T) {
+			result := DetectPatternType(tt.concepts, tt.title, tt.narrative)
+			if result != tt.expected {
+				t.Errorf("Expected %s, got %s", tt.expected, result)
+			}
+		})
+	}
+}
+
+func (pt PatternType) String() string {
+	return string(pt)
+}
+
+func TestExtractSignature(t *testing.T) {
+	concepts := []string{"error-handling", "security"}
+	title := "Nil Pointer Validation Pattern"
+	narrative := "Always validate before dereferencing"
+
+	signature := ExtractSignature(concepts, title, narrative)
+
+	// Should contain concepts
+	found := false
+	for _, s := range signature {
+		if s == "error-handling" {
+			found = true
+			break
+		}
+	}
+	if !found {
+		t.Errorf("Expected signature to contain concepts, got %v", signature)
+	}
+
+	// Should contain significant words from title
+	found = false
+	for _, s := range signature {
+		if s == "validation" || s == "pattern" || s == "pointer" {
+			found = true
+			break
+		}
+	}
+	if !found {
+		t.Errorf("Expected signature to contain title keywords, got %v", signature)
+	}
+}
+
+func TestCalculateMatchScore(t *testing.T) {
+	tests := []struct {
+		name     string
+		sig1     []string
+		sig2     []string
+		minScore float64
+		maxScore float64
+	}{
+		{"identical", []string{"a", "b", "c"}, []string{"a", "b", "c"}, 1.0, 1.0},
+		{"partial", []string{"a", "b", "c"}, []string{"a", "b", "d"}, 0.4, 0.6},
+		{"no_match", []string{"a", "b", "c"}, []string{"x", "y", "z"}, 0.0, 0.0},
+		{"empty", []string{}, []string{"a", "b"}, 0.0, 0.0},
+		{"subset", []string{"a", "b"}, []string{"a", "b", "c", "d"}, 0.4, 0.6},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			score := CalculateMatchScore(tt.sig1, tt.sig2)
+			if score < tt.minScore || score > tt.maxScore {
+				t.Errorf("Expected score between %f and %f, got %f",
+					tt.minScore, tt.maxScore, score)
+			}
+		})
+	}
+}
+
+func TestPattern_MarshalJSON(t *testing.T) {
+	pattern := &Pattern{
+		ID:             1,
+		Name:           "Test Pattern",
+		Type:           PatternTypeBug,
+		Description:    sql.NullString{String: "A description", Valid: true},
+		Signature:      []string{"a", "b"},
+		Recommendation: sql.NullString{String: "Do this", Valid: true},
+		Frequency:      5,
+		Projects:       []string{"proj1", "proj2"},
+		ObservationIDs: []int64{1, 2, 3},
+		Status:         PatternStatusActive,
+		MergedIntoID:   sql.NullInt64{Int64: 0, Valid: false},
+		Confidence:     0.8,
+		LastSeenAt:     time.Now().Format(time.RFC3339),
+		LastSeenEpoch:  time.Now().UnixMilli(),
+		CreatedAt:      time.Now().Format(time.RFC3339),
+		CreatedAtEpoch: time.Now().UnixMilli(),
+	}
+
+	data, err := json.Marshal(pattern)
+	if err != nil {
+		t.Fatalf("Failed to marshal pattern: %v", err)
+	}
+
+	var result PatternJSON
+	if err := json.Unmarshal(data, &result); err != nil {
+		t.Fatalf("Failed to unmarshal pattern: %v", err)
+	}
+
+	if result.Name != pattern.Name {
+		t.Errorf("Expected name %s, got %s", pattern.Name, result.Name)
+	}
+	if result.Description != pattern.Description.String {
+		t.Errorf("Expected description %s, got %s", pattern.Description.String, result.Description)
+	}
+	if result.Frequency != pattern.Frequency {
+		t.Errorf("Expected frequency %d, got %d", pattern.Frequency, result.Frequency)
+	}
+	if result.MergedIntoID != 0 {
+		t.Errorf("Expected merged_into_id 0 for invalid NullInt64, got %d", result.MergedIntoID)
+	}
+}
+
+func TestJSONInt64Array_Scan(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    interface{}
+		expected JSONInt64Array
+		wantErr  bool
+	}{
+		{"string_array", "[1, 2, 3]", JSONInt64Array{1, 2, 3}, false},
+		{"bytes_array", []byte("[4, 5, 6]"), JSONInt64Array{4, 5, 6}, false},
+		{"nil", nil, nil, false},
+		{"empty_string", "", nil, false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			var arr JSONInt64Array
+			err := arr.Scan(tt.input)
+			if (err != nil) != tt.wantErr {
+				t.Errorf("Scan() error = %v, wantErr %v", err, tt.wantErr)
+				return
+			}
+			if len(arr) != len(tt.expected) {
+				t.Errorf("Expected length %d, got %d", len(tt.expected), len(arr))
+			}
+		})
+	}
+}
+
+func TestJSONInt64Array_Value(t *testing.T) {
+	arr := JSONInt64Array{1, 2, 3}
+	val, err := arr.Value()
+	if err != nil {
+		t.Fatalf("Value() error = %v", err)
+	}
+
+	bytes, ok := val.([]byte)
+	if !ok {
+		t.Fatalf("Expected []byte, got %T", val)
+	}
+
+	var result []int64
+	if err := json.Unmarshal(bytes, &result); err != nil {
+		t.Fatalf("Failed to unmarshal: %v", err)
+	}
+
+	if len(result) != 3 || result[0] != 1 || result[1] != 2 || result[2] != 3 {
+		t.Errorf("Expected [1, 2, 3], got %v", result)
+	}
+}
+
+func TestPatternSignatureKeywords(t *testing.T) {
+	// Verify keywords exist for each type
+	types := []PatternType{
+		PatternTypeBug,
+		PatternTypeRefactor,
+		PatternTypeArchitecture,
+		PatternTypeAntiPattern,
+		PatternTypeBestPractice,
+	}
+
+	for _, pt := range types {
+		keywords := PatternSignatureKeywords[pt]
+		if len(keywords) == 0 {
+			t.Errorf("No keywords defined for pattern type %s", pt)
+		}
+	}
+}
+
+func TestUniqueStrings(t *testing.T) {
+	tests := []struct {
+		input    []string
+		expected int
+	}{
+		{[]string{"a", "b", "c"}, 3},
+		{[]string{"a", "a", "b"}, 2},
+		{[]string{"a", "a", "a"}, 1},
+		{[]string{}, 0},
+	}
+
+	for _, tt := range tests {
+		result := uniqueStrings(tt.input)
+		if len(result) != tt.expected {
+			t.Errorf("uniqueStrings(%v) = %v (len=%d), expected len=%d",
+				tt.input, result, len(result), tt.expected)
+		}
+	}
+}
+
+func TestContainsIgnoreCase(t *testing.T) {
+	tests := []struct {
+		text     string
+		substr   string
+		expected bool
+	}{
+		{"Hello World", "hello", true},
+		{"Hello World", "WORLD", true},
+		{"Hello World", "xyz", false},
+		{"", "a", false},
+		{"a", "", true},
+	}
+
+	for _, tt := range tests {
+		result := containsIgnoreCase(tt.text, tt.substr)
+		if result != tt.expected {
+			t.Errorf("containsIgnoreCase(%q, %q) = %v, expected %v",
+				tt.text, tt.substr, result, tt.expected)
+		}
+	}
+}

pkg/models/prompt.go

@@ -3,18 +3,18 @@ package models
 
 // UserPrompt represents a user prompt captured during a session.
 type UserPrompt struct {
-	ID                  int64  `db:"id" json:"id"`
 	ClaudeSessionID     string `db:"claude_session_id" json:"claude_session_id"`
-	PromptNumber        int    `db:"prompt_number" json:"prompt_number"`
 	PromptText          string `db:"prompt_text" json:"prompt_text"`
-	MatchedObservations int    `db:"matched_observations" json:"matched_observations"`
 	CreatedAt           string `db:"created_at" json:"created_at"`
+	ID                  int64  `db:"id" json:"id"`
+	PromptNumber        int    `db:"prompt_number" json:"prompt_number"`
+	MatchedObservations int    `db:"matched_observations" json:"matched_observations"`
 	CreatedAtEpoch      int64  `db:"created_at_epoch" json:"created_at_epoch"`
 }
 
 // UserPromptWithSession includes session context for search results.
 type UserPromptWithSession struct {
-	UserPrompt
 	Project      string `db:"project" json:"project"`
 	SDKSessionID string `db:"sdk_session_id" json:"sdk_session_id"`
+	UserPrompt
 }

pkg/models/relation.go

@@ -0,0 +1,497 @@
+// Package models contains domain models for claude-mnemonic.
+package models
+
+import (
+	"strings"
+	"time"
+)
+
+// RelationType represents the type of relationship between observations.
+type RelationType string
+
+const (
+	// RelationCauses means source observation caused target observation.
+	// Example: "This architectural decision caused this bug"
+	RelationCauses RelationType = "causes"
+	// RelationFixes means source observation fixes target observation.
+	// Example: "This bugfix addresses that discovered issue"
+	RelationFixes RelationType = "fixes"
+	// RelationSupersedes means source observation supersedes target observation.
+	// Example: "This new approach replaces the old workaround"
+	RelationSupersedes RelationType = "supersedes"
+	// RelationDependsOn means source observation depends on target observation.
+	// Example: "This feature relies on that architectural decision"
+	RelationDependsOn RelationType = "depends_on"
+	// RelationRelatesTo means observations are related but no causal relationship.
+	// Example: "Both deal with authentication"
+	RelationRelatesTo RelationType = "relates_to"
+	// RelationEvolvesFrom means source observation evolved from target observation.
+	// Example: "This refined pattern evolved from that initial discovery"
+	RelationEvolvesFrom RelationType = "evolves_from"
+)
+
+// AllRelationTypes is the list of all valid relation types.
+var AllRelationTypes = []RelationType{
+	RelationCauses,
+	RelationFixes,
+	RelationSupersedes,
+	RelationDependsOn,
+	RelationRelatesTo,
+	RelationEvolvesFrom,
+}
+
+// RelationDetectionSource indicates how a relationship was detected.
+type RelationDetectionSource string
+
+const (
+	// DetectionSourceFileOverlap means relationship was detected via shared file references.
+	DetectionSourceFileOverlap RelationDetectionSource = "file_overlap"
+	// DetectionSourceEmbeddingSimilarity means relationship was detected via vector similarity.
+	DetectionSourceEmbeddingSimilarity RelationDetectionSource = "embedding_similarity"
+	// DetectionSourceTemporalProximity means relationship was detected via close timestamps.
+	DetectionSourceTemporalProximity RelationDetectionSource = "temporal_proximity"
+	// DetectionSourceNarrativeMention means relationship was detected via explicit mentions.
+	DetectionSourceNarrativeMention RelationDetectionSource = "narrative_mention"
+	// DetectionSourceConceptOverlap means relationship was detected via shared concepts.
+	DetectionSourceConceptOverlap RelationDetectionSource = "concept_overlap"
+	// DetectionSourceTypeProgression means relationship was detected via type progression pattern.
+	DetectionSourceTypeProgression RelationDetectionSource = "type_progression"
+)
+
+// ObservationRelation represents a directed relationship between two observations.
+type ObservationRelation struct {
+	RelationType    RelationType            `db:"relation_type" json:"relation_type"`
+	DetectionSource RelationDetectionSource `db:"detection_source" json:"detection_source"`
+	Reason          string                  `db:"reason" json:"reason,omitempty"`
+	CreatedAt       string                  `db:"created_at" json:"created_at"`
+	ID              int64                   `db:"id" json:"id"`
+	SourceID        int64                   `db:"source_id" json:"source_id"`
+	TargetID        int64                   `db:"target_id" json:"target_id"`
+	Confidence      float64                 `db:"confidence" json:"confidence"`
+	CreatedAtEpoch  int64                   `db:"created_at_epoch" json:"created_at_epoch"`
+}
+
+// NewObservationRelation creates a new observation relation.
+func NewObservationRelation(sourceID, targetID int64, relType RelationType, confidence float64, source RelationDetectionSource, reason string) *ObservationRelation {
+	now := time.Now()
+	return &ObservationRelation{
+		SourceID:        sourceID,
+		TargetID:        targetID,
+		RelationType:    relType,
+		Confidence:      confidence,
+		DetectionSource: source,
+		Reason:          reason,
+		CreatedAt:       now.Format(time.RFC3339),
+		CreatedAtEpoch:  now.UnixMilli(),
+	}
+}
+
+// RelationDetectionResult contains the result of relation detection.
+type RelationDetectionResult struct {
+	RelationType    RelationType
+	DetectionSource RelationDetectionSource
+	Reason          string
+	SourceID        int64
+	TargetID        int64
+	Confidence      float64
+}
+
+// DetectFileOverlapRelation checks if observations share file references and determines relationship type.
+func DetectFileOverlapRelation(newer, older *Observation) *RelationDetectionResult {
+	// Check for overlapping modified files
+	newerModified := make(map[string]bool)
+	for _, f := range newer.FilesModified {
+		newerModified[f] = true
+	}
+
+	olderModified := make(map[string]bool)
+	for _, f := range older.FilesModified {
+		olderModified[f] = true
+	}
+
+	// Files modified by both
+	var sharedModified []string
+	for f := range newerModified {
+		if olderModified[f] {
+			sharedModified = append(sharedModified, f)
+		}
+	}
+
+	// Files that newer reads which older modified
+	var newerReadsOlderModified []string
+	for _, f := range newer.FilesRead {
+		if olderModified[f] {
+			newerReadsOlderModified = append(newerReadsOlderModified, f)
+		}
+	}
+
+	// Calculate overlap score
+	overlap := len(sharedModified) + len(newerReadsOlderModified)
+	if overlap == 0 {
+		return nil
+	}
+
+	// Determine relationship type based on observation types and file overlap
+	relType := RelationRelatesTo
+	confidence := 0.5 + float64(overlap)*0.1 // Base 0.5, +0.1 per overlapping file
+
+	// Type-based relationship inference
+	switch {
+	case newer.Type == ObsTypeBugfix && (older.Type == ObsTypeDecision || older.Type == ObsTypeFeature):
+		relType = RelationFixes
+		confidence += 0.2
+	case newer.Type == ObsTypeRefactor && older.Type == ObsTypeDiscovery:
+		relType = RelationEvolvesFrom
+		confidence += 0.15
+	case newer.Type == older.Type && len(sharedModified) > 0:
+		relType = RelationSupersedes
+		confidence += 0.1
+	case newer.Type == ObsTypeFeature && older.Type == ObsTypeDecision:
+		relType = RelationDependsOn
+		confidence += 0.15
+	}
+
+	if confidence > 1.0 {
+		confidence = 1.0
+	}
+
+	reason := buildFileOverlapReason(sharedModified, newerReadsOlderModified)
+
+	return &RelationDetectionResult{
+		SourceID:        newer.ID,
+		TargetID:        older.ID,
+		RelationType:    relType,
+		Confidence:      confidence,
+		DetectionSource: DetectionSourceFileOverlap,
+		Reason:          reason,
+	}
+}
+
+// buildFileOverlapReason creates a human-readable reason for file overlap relation.
+func buildFileOverlapReason(shared, readsModified []string) string {
+	parts := []string{}
+	if len(shared) > 0 {
+		parts = append(parts, "both modified: "+strings.Join(truncateList(shared, 3), ", "))
+	}
+	if len(readsModified) > 0 {
+		parts = append(parts, "reads files modified by older: "+strings.Join(truncateList(readsModified, 3), ", "))
+	}
+	return strings.Join(parts, "; ")
+}
+
+// DetectConceptOverlapRelation checks if observations share concepts.
+func DetectConceptOverlapRelation(newer, older *Observation) *RelationDetectionResult {
+	newerConcepts := make(map[string]bool)
+	for _, c := range newer.Concepts {
+		newerConcepts[c] = true
+	}
+
+	var shared []string
+	for _, c := range older.Concepts {
+		if newerConcepts[c] {
+			shared = append(shared, c)
+		}
+	}
+
+	if len(shared) == 0 {
+		return nil
+	}
+
+	// Calculate confidence based on overlap ratio
+	totalUniqueConcepts := len(newerConcepts)
+	for _, c := range older.Concepts {
+		if !newerConcepts[c] {
+			totalUniqueConcepts++
+		}
+	}
+
+	overlapRatio := float64(len(shared)) / float64(totalUniqueConcepts)
+	confidence := 0.3 + overlapRatio*0.5 // Base 0.3, scale with overlap
+
+	// Boost for important concepts
+	for _, c := range shared {
+		if isHighValueConcept(c) {
+			confidence += 0.1
+		}
+	}
+	if confidence > 1.0 {
+		confidence = 1.0
+	}
+
+	return &RelationDetectionResult{
+		SourceID:        newer.ID,
+		TargetID:        older.ID,
+		RelationType:    RelationRelatesTo,
+		Confidence:      confidence,
+		DetectionSource: DetectionSourceConceptOverlap,
+		Reason:          "shared concepts: " + strings.Join(truncateList(shared, 5), ", "),
+	}
+}
+
+// isHighValueConcept returns true for concepts that strongly indicate relationships.
+func isHighValueConcept(concept string) bool {
+	highValue := map[string]bool{
+		"security":       true,
+		"architecture":   true,
+		"gotcha":         true,
+		"anti-pattern":   true,
+		"best-practice":  true,
+		"error-handling": true,
+	}
+	return highValue[concept]
+}
+
+// DetectTypeProgressionRelation checks for natural type progressions.
+// Example: discovery -> decision -> feature -> bugfix
+func DetectTypeProgressionRelation(newer, older *Observation) *RelationDetectionResult {
+	// Define natural type progressions
+	progressions := map[ObservationType][]ObservationType{
+		ObsTypeBugfix:   {ObsTypeDiscovery, ObsTypeFeature, ObsTypeDecision},
+		ObsTypeFeature:  {ObsTypeDiscovery, ObsTypeDecision},
+		ObsTypeRefactor: {ObsTypeDiscovery, ObsTypeFeature, ObsTypeBugfix},
+		ObsTypeDecision: {ObsTypeDiscovery},
+		ObsTypeChange:   {ObsTypeDiscovery, ObsTypeDecision},
+	}
+
+	validPredecessors, ok := progressions[newer.Type]
+	if !ok {
+		return nil
+	}
+
+	isValidProgression := false
+	for _, pred := range validPredecessors {
+		if older.Type == pred {
+			isValidProgression = true
+			break
+		}
+	}
+
+	if !isValidProgression {
+		return nil
+	}
+
+	// Determine relationship type based on progression
+	var relType RelationType
+	var confidence float64 = 0.4
+
+	switch {
+	case newer.Type == ObsTypeBugfix && older.Type == ObsTypeDiscovery:
+		relType = RelationFixes
+		confidence = 0.6
+	case newer.Type == ObsTypeBugfix && older.Type == ObsTypeFeature:
+		relType = RelationFixes
+		confidence = 0.5
+	case newer.Type == ObsTypeFeature && older.Type == ObsTypeDecision:
+		relType = RelationDependsOn
+		confidence = 0.6
+	case newer.Type == ObsTypeRefactor:
+		relType = RelationEvolvesFrom
+		confidence = 0.5
+	default:
+		relType = RelationRelatesTo
+	}
+
+	return &RelationDetectionResult{
+		SourceID:        newer.ID,
+		TargetID:        older.ID,
+		RelationType:    relType,
+		Confidence:      confidence,
+		DetectionSource: DetectionSourceTypeProgression,
+		Reason:          string(older.Type) + " -> " + string(newer.Type) + " progression",
+	}
+}
+
+// DetectTemporalProximityRelation checks if observations are temporally close (same session).
+func DetectTemporalProximityRelation(newer, older *Observation) *RelationDetectionResult {
+	// Only relate observations from the same session
+	if newer.SDKSessionID != older.SDKSessionID {
+		return nil
+	}
+
+	// Check temporal proximity (within 5 minutes)
+	timeDiffMs := newer.CreatedAtEpoch - older.CreatedAtEpoch
+	if timeDiffMs < 0 {
+		timeDiffMs = -timeDiffMs
+	}
+
+	fiveMinutesMs := int64(5 * 60 * 1000)
+	if timeDiffMs > fiveMinutesMs {
+		return nil
+	}
+
+	// Calculate confidence based on temporal proximity
+	// Closer = higher confidence
+	proximityRatio := 1.0 - (float64(timeDiffMs) / float64(fiveMinutesMs))
+	confidence := 0.3 + proximityRatio*0.4
+
+	return &RelationDetectionResult{
+		SourceID:        newer.ID,
+		TargetID:        older.ID,
+		RelationType:    RelationRelatesTo,
+		Confidence:      confidence,
+		DetectionSource: DetectionSourceTemporalProximity,
+		Reason:          "same session, close timestamps",
+	}
+}
+
+// NarrativeMentionPatterns are patterns that indicate explicit relationships in narratives.
+var NarrativeMentionPatterns = []struct {
+	Pattern      string
+	RelationType RelationType
+	ConfBoost    float64
+}{
+	{" caused ", RelationCauses, 0.3},
+	{" causes ", RelationCauses, 0.3},
+	{" because of ", RelationCauses, 0.25},
+	{" due to ", RelationCauses, 0.2},
+	{" fixes ", RelationFixes, 0.3},
+	{" fixed ", RelationFixes, 0.3},
+	{" resolves ", RelationFixes, 0.3},
+	{" addresses ", RelationFixes, 0.25},
+	{" replaces ", RelationSupersedes, 0.3},
+	{" supersedes ", RelationSupersedes, 0.35},
+	{" instead of ", RelationSupersedes, 0.25},
+	{" depends on ", RelationDependsOn, 0.3},
+	{" requires ", RelationDependsOn, 0.25},
+	{" builds on ", RelationDependsOn, 0.25},
+	{" based on ", RelationDependsOn, 0.2},
+	{" related to ", RelationRelatesTo, 0.2},
+	{" similar to ", RelationRelatesTo, 0.2},
+	{" evolved from ", RelationEvolvesFrom, 0.3},
+	{" improved from ", RelationEvolvesFrom, 0.25},
+	{" refined from ", RelationEvolvesFrom, 0.25},
+}
+
+// DetectNarrativeMentionRelation checks if newer observation's narrative mentions relationship.
+func DetectNarrativeMentionRelation(newer, older *Observation) *RelationDetectionResult {
+	if !newer.Narrative.Valid || newer.Narrative.String == "" {
+		return nil
+	}
+
+	narrative := strings.ToLower(newer.Narrative.String)
+
+	// Check for patterns
+	for _, p := range NarrativeMentionPatterns {
+		if strings.Contains(narrative, p.Pattern) {
+			// Found a pattern - this is a potential relationship
+			confidence := 0.4 + p.ConfBoost
+			if confidence > 1.0 {
+				confidence = 1.0
+			}
+
+			return &RelationDetectionResult{
+				SourceID:        newer.ID,
+				TargetID:        older.ID,
+				RelationType:    p.RelationType,
+				Confidence:      confidence,
+				DetectionSource: DetectionSourceNarrativeMention,
+				Reason:          "narrative contains '" + strings.TrimSpace(p.Pattern) + "' language",
+			}
+		}
+	}
+
+	return nil
+}
+
+// DetectRelationsWithExisting checks a new observation against existing ones and returns detected relations.
+// This is the main entry point for relation detection.
+func DetectRelationsWithExisting(newer *Observation, existing []*Observation, minConfidence float64) []*RelationDetectionResult {
+	var results []*RelationDetectionResult
+	seen := make(map[int64]bool)
+
+	for _, older := range existing {
+		// Skip self
+		if older.ID == newer.ID {
+			continue
+		}
+
+		// Skip if already superseded
+		if older.IsSuperseded {
+			continue
+		}
+
+		// Only compare within same project (or both global)
+		if newer.Project != older.Project && newer.Scope != ScopeGlobal && older.Scope != ScopeGlobal {
+			continue
+		}
+
+		// Run all detection methods and keep highest confidence result per target
+		var bestResult *RelationDetectionResult
+
+		// 1. File overlap detection
+		if result := DetectFileOverlapRelation(newer, older); result != nil && result.Confidence >= minConfidence {
+			if bestResult == nil { //nolint:govet
+				bestResult = result
+			} else if result.Confidence > bestResult.Confidence {
+				bestResult = result
+			}
+		}
+
+		// 2. Concept overlap detection
+		if result := DetectConceptOverlapRelation(newer, older); result != nil && result.Confidence >= minConfidence {
+			if bestResult == nil {
+				bestResult = result
+			} else if result.Confidence > bestResult.Confidence {
+				bestResult = result
+			}
+		}
+
+		// 3. Type progression detection
+		if result := DetectTypeProgressionRelation(newer, older); result != nil && result.Confidence >= minConfidence {
+			if bestResult == nil {
+				bestResult = result
+			} else if result.Confidence > bestResult.Confidence {
+				bestResult = result
+			}
+		}
+
+		// 4. Temporal proximity detection
+		if result := DetectTemporalProximityRelation(newer, older); result != nil && result.Confidence >= minConfidence {
+			// Only use temporal proximity if no better detection found
+			if bestResult == nil {
+				bestResult = result
+			}
+		}
+
+		// 5. Narrative mention detection (can upgrade relation type)
+		if result := DetectNarrativeMentionRelation(newer, older); result != nil && result.Confidence >= minConfidence {
+			if bestResult == nil {
+				bestResult = result
+			} else if result.Confidence > bestResult.Confidence {
+				bestResult = result
+			}
+		}
+
+		// Add best result if found and not already seen
+		if bestResult != nil && !seen[older.ID] {
+			results = append(results, bestResult)
+			seen[older.ID] = true
+		}
+	}
+
+	return results
+}
+
+// truncateList truncates a list to maxLen items.
+func truncateList(items []string, maxLen int) []string {
+	if len(items) <= maxLen {
+		return items
+	}
+	result := items[:maxLen]
+	return append(result, "...")
+}
+
+// RelationWithDetails contains a relation with its observation details.
+type RelationWithDetails struct {
+	Relation    *ObservationRelation `json:"relation"`
+	SourceTitle string               `json:"source_title"`
+	TargetTitle string               `json:"target_title"`
+	SourceType  ObservationType      `json:"source_type"`
+	TargetType  ObservationType      `json:"target_type"`
+}
+
+// RelationGraph represents a graph of related observations.
+type RelationGraph struct {
+	Relations []*RelationWithDetails `json:"relations"`
+	CenterID  int64                  `json:"center_id"`
+}