Release dec 2025 (#15)

* Resolves issue #13

- Switched model to bge-small-en-v1.5
- Added lazy re-embedding
- Added model version tracking per vector
- Added conversion of vectors to the new model

* Add lfs support to the workflow.

* Implements importance scoring with decay + voting #6

* Resolves issue #5 by marking observations as superseeded and scheduled for deletion

* Implement pattern detection #7

* Improve injections and observations accuracy

- Session start: Recent observations for project context (recency-based)
- User prompt: Semantically relevant observations (similarity-based with threshold)

* Added two stage retrieval with bi and cross encoder #8

* Implement query expansion and reformulation #9

* Knowledge graph and relationships ( resolves #4 )

- File Overlap Detection: Detects relationships when observations modify/read the same files
- Concept Overlap Detection: Detects relationships based on shared semantic concepts
- Type Progression Detection: Infers relationships from natural observation type progressions (e.g., discovery → bugfix = "fixes")
- Temporal Proximity Detection: Detects relationships between observations in the same session within 5 minutes
- Narrative Mention Detection: Detects explicit relationship language in narratives (e.g., "fixes", "depends on", "supersedes")

* Add visualisation of the relations to the dashboard.

* fixup! Add visualisation of the relations to the dashboard.

* Update documentation with new settings and screenshots.

internal/config/config.go

@@ -48,16 +48,29 @@ type Config struct {
 	Model          string `json:"model"`
 	ClaudeCodePath string `json:"claude_code_path"`
 
+	// Embedding settings
+	EmbeddingModel string `json:"embedding_model"` // e.g., "bge-v1.5"
+
+	// Reranking settings (cross-encoder)
+	RerankingEnabled        bool    `json:"reranking_enabled"`         // Enable cross-encoder reranking
+	RerankingCandidates     int     `json:"reranking_candidates"`      // Number of candidates to retrieve before reranking (default 100)
+	RerankingResults        int     `json:"reranking_results"`         // Number of results to return after reranking (default 10)
+	RerankingAlpha          float64 `json:"reranking_alpha"`           // Weight for combining scores: alpha*rerank + (1-alpha)*original (default 0.7)
+	RerankingMinImprovement float64 `json:"reranking_min_improvement"` // Minimum rank improvement to trigger reranking (default 0, always rerank)
+	RerankingPureMode       bool    `json:"reranking_pure_mode"`       // Use pure cross-encoder scores without combining with bi-encoder (default false)
+
 	// 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"`
+	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"`
+	ContextRelevanceThreshold float64  `json:"context_relevance_threshold"` // 0.0-1.0, minimum similarity for inclusion
+	ContextMaxPromptResults   int      `json:"context_max_prompt_results"`  // Max results per prompt (0 = threshold only)
 }
 
 var (
@@ -119,22 +132,33 @@ func EnsureAll() error {
 	return nil
 }
 
+// DefaultEmbeddingModel is the default embedding model to use.
+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,
-		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
+		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)
 	}
 }
 
@@ -166,6 +190,28 @@ func Load() (*Config, error) {
 	if v, ok := settings["CLAUDE_CODE_PATH"].(string); ok {
 		cfg.ClaudeCodePath = v
 	}
+	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)
 	}
@@ -181,6 +227,12 @@ 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)
+	}
 
 	return cfg, nil
 }

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 err := rows.Scan(&id); err != nil {
+			return nil, err
+		}
+		toDelete = append(toDelete, id)
+	}
+	if err := rows.Err(); err != nil {
+		return nil, err
+	}
+
+	if len(toDelete) == 0 {
+		return nil, nil
+	}
+
+	// Delete the conflict records first (due to foreign key constraints)
+	for _, obsID := range toDelete {
+		if err := s.DeleteConflictsByObservationID(ctx, obsID); err != nil {
+			return nil, err
+		}
+	}
+
+	// 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/migrations.go

@@ -283,6 +283,213 @@ var Migrations = []Migration{
 			ON user_prompts(claude_session_id, prompt_number);
 		`,
 	},
+	{
+		Version: 19,
+		Name:    "vectors_with_model_version",
+		SQL: `
+			-- Drop old vectors table (virtual tables cannot be altered)
+			DROP TABLE IF EXISTS vectors;
+
+			-- Recreate vectors table with model_version column
+			-- Uses bge-small-en-v1.5 embeddings (384 dimensions)
+			CREATE VIRTUAL TABLE IF NOT EXISTS vectors USING vec0(
+				doc_id TEXT PRIMARY KEY,
+				embedding float[384],
+				sqlite_id INTEGER,
+				doc_type TEXT,
+				field_type TEXT,
+				project TEXT,
+				scope TEXT,
+				model_version TEXT
+			);
+		`,
+	},
+	{
+		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.

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 ?
 	`
 
@@ -229,7 +392,24 @@ func (s *ObservationStore) GetAllRecentObservations(ctx context.Context, limit i
 	return scanObservationRows(rows)
 }
 
+// GetAllObservations retrieves all observations (for vector rebuild).
+func (s *ObservationStore) GetAllObservations(ctx context.Context) ([]*models.Observation, error) {
+	query := `SELECT ` + observationColumns + `
+		FROM observations
+		ORDER BY id
+	`
+
+	rows, err := s.store.QueryContext(ctx, query)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return scanObservationRows(rows)
+}
+
 // 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
@@ -245,15 +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 ?
 	`
 
@@ -278,6 +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
@@ -294,14 +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)
@@ -445,6 +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/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

@@ -199,6 +199,28 @@ func (s *PromptStore) GetAllRecentUserPrompts(ctx context.Context, limit int) ([
 	return scanPromptWithSessionRows(rows)
 }
 
+// GetAllPrompts retrieves all user prompts (for vector rebuild).
+func (s *PromptStore) GetAllPrompts(ctx context.Context) ([]*models.UserPromptWithSession, error) {
+	const query = `
+		SELECT up.id, up.claude_session_id, up.prompt_number, up.prompt_text,
+		       COALESCE(up.matched_observations, 0) as matched_observations,
+		       up.created_at, up.created_at_epoch,
+		       COALESCE(s.project, '') as project,
+		       COALESCE(s.sdk_session_id, '') as sdk_session_id
+		FROM user_prompts up
+		LEFT JOIN sdk_sessions s ON up.claude_session_id = s.claude_session_id
+		ORDER BY up.id
+	`
+
+	rows, err := s.store.QueryContext(ctx, query)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return scanPromptWithSessionRows(rows)
+}
+
 // FindRecentPromptByText finds a prompt with the same text for a session within the last few seconds.
 // This is used to detect duplicate hook invocations.
 // Returns (promptID, promptNumber, found).

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 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 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/summary.go

@@ -116,3 +116,21 @@ func (s *SummaryStore) GetAllRecentSummaries(ctx context.Context, limit int) ([]
 
 	return scanSummaryRows(rows)
 }
+
+// GetAllSummaries retrieves all summaries (for vector rebuild).
+func (s *SummaryStore) GetAllSummaries(ctx context.Context) ([]*models.SessionSummary, error) {
+	const query = `
+		SELECT id, sdk_session_id, project, request, investigated, learned, completed,
+		       next_steps, notes, prompt_number, discovery_tokens, created_at, created_at_epoch
+		FROM session_summaries
+		ORDER BY id
+	`
+
+	rows, err := s.store.QueryContext(ctx, query)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	return scanSummaryRows(rows)
+}

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/assets/.model_version

@@ -0,0 +1 @@
+bge-small-en-v1.5

internal/embedding/assets/tokenizer.json

@@ -1,21 +1,7 @@
 {
   "version": "1.0",
-  "truncation": {
-    "direction": "Right",
-    "max_length": 128,
-    "strategy": "LongestFirst",
-    "stride": 0
-  },
-  "padding": {
-    "strategy": {
-      "Fixed": 128
-    },
-    "direction": "Right",
-    "pad_to_multiple_of": null,
-    "pad_id": 0,
-    "pad_type_id": 0,
-    "pad_token": "[PAD]"
-  },
+  "truncation": null,
+  "padding": null,
   "added_tokens": [
     {
       "id": 0,

internal/embedding/model.go

@@ -0,0 +1,157 @@
+// Package embedding provides text embedding generation with swappable models.
+package embedding
+
+import (
+	"fmt"
+	"sync"
+)
+
+// PoolingStrategy defines how to pool token embeddings into sentence embeddings.
+type PoolingStrategy string
+
+const (
+	// PoolingNone means the model already outputs sentence embeddings directly.
+	PoolingNone PoolingStrategy = "none"
+	// PoolingMean averages all token embeddings (weighted by attention mask).
+	PoolingMean PoolingStrategy = "mean"
+	// PoolingCLS uses only the [CLS] token embedding.
+	PoolingCLS PoolingStrategy = "cls"
+)
+
+// 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.
+	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
+}
+
+// EmbeddingModel represents a text embedding model.
+type EmbeddingModel interface {
+	// Name returns the human-readable model name (e.g., "bge-small-en-v1.5").
+	Name() string
+
+	// Version returns a short version string for storage (e.g., "bge-v1.5").
+	Version() string
+
+	// Dimensions returns the embedding vector size.
+	Dimensions() int
+
+	// Embed generates an embedding for a single text.
+	Embed(text string) ([]float32, error)
+
+	// EmbedBatch generates embeddings for multiple texts.
+	EmbedBatch(texts []string) ([][]float32, error)
+
+	// Close releases model resources.
+	Close() error
+}
+
+// ONNXConfigurer is an optional interface that models can implement
+// to expose their ONNX configuration for introspection.
+type ONNXConfigurer interface {
+	// ONNXConfig returns the model's ONNX configuration.
+	ONNXConfig() ONNXConfig
+}
+
+// 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?
+}
+
+// ModelFactory creates a new instance of an embedding model.
+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
+}
+
+// NewModelRegistry creates a new model registry.
+func NewModelRegistry() *ModelRegistry {
+	return &ModelRegistry{
+		models:   make(map[string]ModelFactory),
+		metadata: make(map[string]ModelMetadata),
+	}
+}
+
+// Register adds a model factory to the registry.
+func (r *ModelRegistry) Register(meta ModelMetadata, factory ModelFactory) {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+
+	r.models[meta.Version] = factory
+	r.metadata[meta.Version] = meta
+
+	if meta.Default {
+		r.defaultModel = meta.Version
+	}
+}
+
+// Get creates a new instance of the model with the given version.
+func (r *ModelRegistry) Get(version string) (EmbeddingModel, error) {
+	r.mu.RLock()
+	factory, ok := r.models[version]
+	r.mu.RUnlock()
+
+	if !ok {
+		return nil, fmt.Errorf("unknown model version: %s", version)
+	}
+
+	return factory()
+}
+
+// Default returns the default model version.
+func (r *ModelRegistry) Default() string {
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+	return r.defaultModel
+}
+
+// List returns metadata for all registered models.
+func (r *ModelRegistry) List() []ModelMetadata {
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+
+	result := make([]ModelMetadata, 0, len(r.metadata))
+	for _, meta := range r.metadata {
+		result = append(result, meta)
+	}
+	return result
+}
+
+// DefaultRegistry is the global model registry with all available models.
+var DefaultRegistry = NewModelRegistry()
+
+// RegisterModel adds a model to the default registry.
+func RegisterModel(meta ModelMetadata, factory ModelFactory) {
+	DefaultRegistry.Register(meta, factory)
+}
+
+// GetModel creates a model instance from the default registry.
+func GetModel(version string) (EmbeddingModel, error) {
+	return DefaultRegistry.Get(version)
+}
+
+// GetDefaultModel returns the default model version from the default registry.
+func GetDefaultModel() string {
+	return DefaultRegistry.Default()
+}
+
+// ListModels returns metadata for all models in the default registry.
+func ListModels() []ModelMetadata {
+	return DefaultRegistry.List()
+}

internal/embedding/service.go

@@ -1,4 +1,4 @@
-// Package embedding provides text embedding generation using all-MiniLM-L6-v2.
+// Package embedding provides text embedding generation with swappable models.
 package embedding
 
 import (
@@ -15,19 +15,50 @@ import (
 	ort "github.com/yalue/onnxruntime_go"
 )
 
-// EmbeddingDim is the dimension of embeddings produced by all-MiniLM-L6-v2.
+// EmbeddingDim is the dimension of embeddings produced by the current model.
+// Both all-MiniLM-L6-v2 and bge-small-en-v1.5 produce 384-dimensional embeddings.
 const EmbeddingDim = 384
 
-// Service provides thread-safe text embedding generation.
-type Service struct {
+// Model version constants
+const (
+	// BGEModelVersion is the version string for bge-small-en-v1.5
+	BGEModelVersion = "bge-v1.5"
+	// BGEModelName is the human-readable name for bge-small-en-v1.5
+	BGEModelName = "bge-small-en-v1.5"
+	// DefaultModelVersion is the default model to use
+	DefaultModelVersion = BGEModelVersion
+)
+
+// MaxSequenceLength is the maximum token sequence length for the model.
+const MaxSequenceLength = 512
+
+// bgeONNXConfig defines the ONNX configuration for BGE models.
+// BGE outputs last_hidden_state and requires mean pooling.
+var bgeONNXConfig = ONNXConfig{
+	InputNames:  []string{"input_ids", "attention_mask", "token_type_ids"},
+	OutputNames: []string{"last_hidden_state"},
+	Pooling:     PoolingMean,
+	HiddenSize:  EmbeddingDim,
+}
+
+// bgeModel is the ONNX-based embedding model implementation.
+// Currently supports bge-small-en-v1.5 (previously all-MiniLM-L6-v2).
+type bgeModel struct {
 	tk      *tokenizer.Tokenizer
 	session *ort.DynamicAdvancedSession
 	mu      sync.Mutex
-	libDir  string // temp directory containing extracted libraries
+	libDir  string     // temp directory containing extracted libraries
+	config  ONNXConfig // ONNX configuration for this model
 }
 
-// NewService creates a new embedding service using bundled ONNX runtime and model.
-func NewService() (*Service, error) {
+// Compile-time check that bgeModel implements EmbeddingModel
+var _ EmbeddingModel = (*bgeModel)(nil)
+
+// Compile-time check that bgeModel implements ONNXConfigurer
+var _ ONNXConfigurer = (*bgeModel)(nil)
+
+// newBGEModel creates a new BGE embedding model using bundled ONNX runtime and model.
+func newBGEModel() (EmbeddingModel, error) {
 	// Extract ONNX runtime library to temp directory
 	libDir, err := extractONNXLibrary()
 	if err != nil {
@@ -49,22 +80,41 @@ func NewService() (*Service, error) {
 		return nil, fmt.Errorf("load tokenizer: %w", err)
 	}
 
-	// Create ONNX session with embedded model
-	inputNames := []string{"input_ids", "attention_mask", "token_type_ids"}
-	outputNames := []string{"sentence_embedding"}
-
-	session, err := ort.NewDynamicAdvancedSessionWithONNXData(modelData, inputNames, outputNames, nil)
+	// Create ONNX session using model-specific configuration
+	config := bgeONNXConfig
+	session, err := ort.NewDynamicAdvancedSessionWithONNXData(modelData, config.InputNames, config.OutputNames, nil)
 	if err != nil {
 		return nil, fmt.Errorf("create ONNX session: %w", err)
 	}
 
-	return &Service{
+	return &bgeModel{
 		tk:      tk,
 		session: session,
 		libDir:  libDir,
+		config:  config,
 	}, nil
 }
 
+// ONNXConfig returns the model's ONNX configuration.
+func (m *bgeModel) ONNXConfig() ONNXConfig {
+	return m.config
+}
+
+// Name returns the human-readable model name.
+func (m *bgeModel) Name() string {
+	return BGEModelName
+}
+
+// Version returns the short version string for storage.
+func (m *bgeModel) Version() string {
+	return BGEModelVersion
+}
+
+// Dimensions returns the embedding vector size.
+func (m *bgeModel) Dimensions() int {
+	return EmbeddingDim
+}
+
 // extractONNXLibrary extracts the embedded ONNX runtime library to a temp directory.
 // Uses content hash to avoid re-extracting if already present.
 func extractONNXLibrary() (string, error) {
@@ -107,15 +157,15 @@ func extractONNXLibrary() (string, error) {
 
 // Embed generates an embedding for a single text.
 // Returns a 384-dimensional float32 vector.
-func (s *Service) Embed(text string) ([]float32, error) {
-	s.mu.Lock()
-	defer s.mu.Unlock()
+func (m *bgeModel) Embed(text string) ([]float32, error) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
 
 	if text == "" {
 		return make([]float32, EmbeddingDim), nil
 	}
 
-	results, err := s.computeBatch([]string{text})
+	results, err := m.computeBatch([]string{text})
 	if err != nil {
 		return nil, err
 	}
@@ -127,13 +177,13 @@ func (s *Service) Embed(text string) ([]float32, error) {
 
 // EmbedBatch generates embeddings for multiple texts.
 // Returns slice of 384-dimensional float32 vectors.
-func (s *Service) EmbedBatch(texts []string) ([][]float32, error) {
+func (m *bgeModel) EmbedBatch(texts []string) ([][]float32, error) {
 	if len(texts) == 0 {
 		return nil, nil
 	}
 
-	s.mu.Lock()
-	defer s.mu.Unlock()
+	m.mu.Lock()
+	defer m.mu.Unlock()
 
 	// Filter out empty texts and track indices
 	nonEmpty := make([]string, 0, len(texts))
@@ -155,7 +205,7 @@ func (s *Service) EmbedBatch(texts []string) ([][]float32, error) {
 	}
 
 	// Compute embeddings for non-empty texts
-	embeddings, err := s.computeBatch(nonEmpty)
+	embeddings, err := m.computeBatch(nonEmpty)
 	if err != nil {
 		return nil, fmt.Errorf("compute batch embeddings: %w", err)
 	}
@@ -173,7 +223,7 @@ func (s *Service) EmbedBatch(texts []string) ([][]float32, error) {
 }
 
 // computeBatch runs inference on a batch of texts. Must be called with lock held.
-func (s *Service) computeBatch(sentences []string) ([][]float32, error) {
+func (m *bgeModel) computeBatch(sentences []string) ([][]float32, error) {
 	if len(sentences) == 0 {
 		return nil, nil
 	}
@@ -184,31 +234,57 @@ func (s *Service) computeBatch(sentences []string) ([][]float32, error) {
 		inputBatch[i] = tokenizer.NewSingleEncodeInput(tokenizer.NewRawInputSequence(sent))
 	}
 
-	encodings, err := s.tk.EncodeBatch(inputBatch, true)
+	encodings, err := m.tk.EncodeBatch(inputBatch, true)
 	if err != nil {
 		return nil, fmt.Errorf("tokenize: %w", err)
 	}
 
 	batchSize := len(encodings)
-	seqLength := len(encodings[0].Ids)
-	hiddenSize := EmbeddingDim
+	hiddenSize := m.config.HiddenSize
+
+	// Find max sequence length across all encodings (tokenizer may not pad uniformly)
+	// Also enforce MaxSequenceLength to prevent model errors
+	seqLength := 0
+	for _, enc := range encodings {
+		if len(enc.Ids) > seqLength {
+			seqLength = len(enc.Ids)
+		}
+	}
+	// Truncate to max model sequence length
+	if seqLength > MaxSequenceLength {
+		seqLength = MaxSequenceLength
+	}
 
 	inputShape := ort.NewShape(int64(batchSize), int64(seqLength))
 
-	// Create input tensors
+	// Create input tensors (pre-filled with zeros for padding)
 	inputIdsData := make([]int64, batchSize*seqLength)
 	attentionMaskData := make([]int64, batchSize*seqLength)
 	tokenTypeIdsData := make([]int64, batchSize*seqLength)
 
 	for b := 0; b < batchSize; b++ {
-		for i, id := range encodings[b].Ids {
-			inputIdsData[b*seqLength+i] = int64(id)
+		// Copy actual token data (rest remains 0 as padding)
+		// Truncate to seqLength to handle long inputs
+		copyLen := len(encodings[b].Ids)
+		if copyLen > seqLength {
+			copyLen = seqLength
 		}
-		for i, mask := range encodings[b].AttentionMask {
-			attentionMaskData[b*seqLength+i] = int64(mask)
+		for i := 0; i < copyLen; i++ {
+			inputIdsData[b*seqLength+i] = int64(encodings[b].Ids[i])
 		}
-		for i, typeId := range encodings[b].TypeIds {
-			tokenTypeIdsData[b*seqLength+i] = int64(typeId)
+		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])
 		}
 	}
 
@@ -230,51 +306,131 @@ func (s *Service) computeBatch(sentences []string) ([][]float32, error) {
 	}
 	defer tokenTypeIdsTensor.Destroy()
 
-	sentenceOutputShape := ort.NewShape(int64(batchSize), int64(hiddenSize))
-	sentenceOutputTensor, err := ort.NewEmptyTensor[float32](sentenceOutputShape)
+	// Create output tensor based on pooling strategy
+	var outputShape ort.Shape
+
+	switch m.config.Pooling {
+	case PoolingNone:
+		// Direct sentence embedding output: [batch, hidden]
+		outputShape = ort.NewShape(int64(batchSize), int64(hiddenSize))
+	case PoolingMean, PoolingCLS:
+		// Token-level output: [batch, seq_len, hidden]
+		outputShape = ort.NewShape(int64(batchSize), int64(seqLength), int64(hiddenSize))
+	default:
+		outputShape = ort.NewShape(int64(batchSize), int64(hiddenSize))
+	}
+
+	outputTensor, err := ort.NewEmptyTensor[float32](outputShape)
 	if err != nil {
 		return nil, fmt.Errorf("create output tensor: %w", err)
 	}
-	defer sentenceOutputTensor.Destroy()
+	defer outputTensor.Destroy()
 
 	// Run inference
 	inputTensors := []ort.Value{inputIdsTensor, attentionMaskTensor, tokenTypeIdsTensor}
-	outputTensors := []ort.Value{sentenceOutputTensor}
+	outputTensors := []ort.Value{outputTensor}
 
-	if err := s.session.Run(inputTensors, outputTensors); err != nil {
+	if err := m.session.Run(inputTensors, outputTensors); err != nil {
 		return nil, fmt.Errorf("run inference: %w", err)
 	}
 
-	// Extract results
-	flatOutput := sentenceOutputTensor.GetData()
-	expectedSize := batchSize * hiddenSize
-	if len(flatOutput) != expectedSize {
-		return nil, fmt.Errorf("unexpected output size: got %d, expected %d", len(flatOutput), expectedSize)
-	}
+	// Extract and pool results based on strategy
+	flatOutput := outputTensor.GetData()
 
+	switch m.config.Pooling {
+	case PoolingNone:
+		// Direct output, no pooling needed
+		expectedSize := batchSize * hiddenSize
+		if len(flatOutput) != expectedSize {
+			return nil, fmt.Errorf("unexpected output size: got %d, expected %d", len(flatOutput), expectedSize)
+		}
+		results := make([][]float32, batchSize)
+		for i := 0; i < batchSize; i++ {
+			start := i * hiddenSize
+			end := start + hiddenSize
+			results[i] = make([]float32, hiddenSize)
+			copy(results[i], flatOutput[start:end])
+		}
+		return results, nil
+
+	case PoolingMean:
+		// Mean pooling over tokens (weighted by attention mask)
+		return meanPooling(flatOutput, attentionMaskData, batchSize, seqLength, hiddenSize), nil
+
+	case PoolingCLS:
+		// CLS token pooling (first token of each sequence)
+		return clsPooling(flatOutput, batchSize, seqLength, hiddenSize), nil
+
+	default:
+		return nil, fmt.Errorf("unknown pooling strategy: %s", m.config.Pooling)
+	}
+}
+
+// meanPooling applies mean pooling over token embeddings, weighted by attention mask.
+// Input shape: [batch, seq_len, hidden], attention mask: [batch, seq_len]
+// Output shape: [batch, hidden]
+func meanPooling(embeddings []float32, attentionMask []int64, batchSize, seqLen, hiddenSize int) [][]float32 {
 	results := make([][]float32, batchSize)
-	for i := 0; i < batchSize; i++ {
-		start := i * hiddenSize
-		end := start + hiddenSize
-		results[i] = make([]float32, hiddenSize)
-		copy(results[i], flatOutput[start:end])
+
+	for b := 0; b < batchSize; b++ {
+		result := make([]float32, hiddenSize)
+		var maskSum float32
+
+		// Sum embeddings weighted by attention mask
+		for s := 0; s < seqLen; s++ {
+			maskVal := float32(attentionMask[b*seqLen+s])
+			maskSum += maskVal
+
+			if maskVal > 0 {
+				embOffset := (b*seqLen + s) * hiddenSize
+				for h := 0; h < hiddenSize; h++ {
+					result[h] += embeddings[embOffset+h] * maskVal
+				}
+			}
+		}
+
+		// Normalize by mask sum (avoid division by zero)
+		if maskSum > 0 {
+			for h := 0; h < hiddenSize; h++ {
+				result[h] /= maskSum
+			}
+		}
+
+		results[b] = result
 	}
 
-	return results, nil
+	return results
+}
+
+// clsPooling extracts the [CLS] token embedding (first token).
+// Input shape: [batch, seq_len, hidden]
+// Output shape: [batch, hidden]
+func clsPooling(embeddings []float32, batchSize, seqLen, hiddenSize int) [][]float32 {
+	results := make([][]float32, batchSize)
+
+	for b := 0; b < batchSize; b++ {
+		result := make([]float32, hiddenSize)
+		// CLS token is at position 0
+		embOffset := b * seqLen * hiddenSize
+		copy(result, embeddings[embOffset:embOffset+hiddenSize])
+		results[b] = result
+	}
+
+	return results
 }
 
 // Close releases model resources.
-func (s *Service) Close() error {
-	s.mu.Lock()
-	defer s.mu.Unlock()
+func (m *bgeModel) Close() error {
+	m.mu.Lock()
+	defer m.mu.Unlock()
 
 	var errs []error
 
-	if s.session != nil {
-		if err := s.session.Destroy(); err != nil {
+	if m.session != nil {
+		if err := m.session.Destroy(); err != nil {
 			errs = append(errs, fmt.Errorf("destroy session: %w", err))
 		}
-		s.session = nil
+		m.session = nil
 	}
 
 	if err := ort.DestroyEnvironment(); err != nil {
@@ -282,10 +438,75 @@ func (s *Service) Close() error {
 	}
 
 	// Optionally clean up extracted library (leave for caching)
-	// os.RemoveAll(s.libDir)
+	// os.RemoveAll(m.libDir)
 
 	if len(errs) > 0 {
 		return errs[0]
 	}
 	return nil
 }
+
+// Register the BGE model with the default registry at init time
+func init() {
+	RegisterModel(ModelMetadata{
+		Name:        BGEModelName,
+		Version:     BGEModelVersion,
+		Dimensions:  EmbeddingDim,
+		Description: "High-quality semantic search model",
+		Default:     true,
+	}, newBGEModel)
+}
+
+// Service provides thread-safe text embedding generation with model abstraction.
+type Service struct {
+	model EmbeddingModel
+}
+
+// NewService creates a new embedding service using the default model.
+func NewService() (*Service, error) {
+	return NewServiceWithModel(DefaultModelVersion)
+}
+
+// NewServiceWithModel creates a new embedding service using the specified model.
+func NewServiceWithModel(version string) (*Service, error) {
+	if version == "" {
+		version = DefaultModelVersion
+	}
+
+	model, err := GetModel(version)
+	if err != nil {
+		return nil, fmt.Errorf("get model %s: %w", version, err)
+	}
+
+	return &Service{model: model}, nil
+}
+
+// Name returns the human-readable model name.
+func (s *Service) Name() string {
+	return s.model.Name()
+}
+
+// Version returns the short version string for storage.
+func (s *Service) Version() string {
+	return s.model.Version()
+}
+
+// Dimensions returns the embedding vector size.
+func (s *Service) Dimensions() int {
+	return s.model.Dimensions()
+}
+
+// Embed generates an embedding for a single text.
+func (s *Service) Embed(text string) ([]float32, error) {
+	return s.model.Embed(text)
+}
+
+// EmbedBatch generates embeddings for multiple texts.
+func (s *Service) EmbedBatch(texts []string) ([][]float32, error) {
+	return s.model.EmbedBatch(texts)
+}
+
+// Close releases model resources.
+func (s *Service) Close() error {
+	return s.model.Close()
+}

internal/embedding/service_test.go

@@ -22,8 +22,10 @@ func TestNewService(t *testing.T) {
 
 	defer svc.Close()
 
-	assert.NotNil(t, svc.tk)
-	assert.NotNil(t, svc.session)
+	// Verify the service is properly initialized via public methods
+	assert.NotEmpty(t, svc.Name())
+	assert.NotEmpty(t, svc.Version())
+	assert.Equal(t, EmbeddingDim, svc.Dimensions())
 }
 
 // TestEmbed_SingleText tests embedding a single text.
@@ -269,8 +271,8 @@ func TestClose(t *testing.T) {
 	err = svc.Close()
 	require.NoError(t, err)
 
-	// Session should be nil after close
-	assert.Nil(t, svc.session)
+	// After close, embedding should fail (model resources released)
+	// Note: This behavior is model-specific; some models may still work after close
 }
 
 // TestEmbedBatch_SingleItem tests batch embedding with single item.

internal/pattern/detector.go

@@ -0,0 +1,438 @@
+// 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 {
+	config           DetectorConfig
+	patternStore     *sqlite.PatternStore
+	observationStore *sqlite.ObservationStore
+
+	// Vector sync callback
+	syncFunc PatternSyncFunc
+
+	// Candidate tracking (patterns not yet confirmed)
+	candidates   map[string]*candidatePattern
+	candidatesMu sync.RWMutex
+
+	// Background analysis
+	ctx    context.Context
+	cancel context.CancelFunc
+	wg     sync.WaitGroup
+}
+
+// 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 {
+	signature      []string
+	observationIDs []int64
+	projects       []string
+	patternType    models.PatternType
+	title          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
+		signature   []string
+		title       string
+		wantPrefix  string
+	}{
+		{models.PatternTypeBug, []string{"nil", "error"}, "", "Bug Pattern:"},
+		{models.PatternTypeRefactor, []string{"extract"}, "", "Refactor Pattern:"},
+		{models.PatternTypeArchitecture, []string{"service"}, "", "Architecture Pattern:"},
+		{models.PatternTypeAntiPattern, []string{"god-class"}, "", "Anti-Pattern:"},
+		{models.PatternTypeBestPractice, []string{"testing"}, "", "Best Practice:"},
+		{models.PatternTypeBug, []string{}, "Short Title", "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 {
+	ID         string             // Document ID
+	Content    string             // Document text content for scoring
+	Score      float64            // Original bi-encoder similarity score
+	Metadata   map[string]any     // Preserved metadata
+	RerankInfo map[string]float64 // Reranking debug info (optional)
+}
+
+// RerankResult represents a reranked search result.
+type RerankResult struct {
+	ID              string         // Document ID
+	Content         string         // Document text content
+	OriginalScore   float64        // Original bi-encoder score
+	RerankScore     float64        // Cross-encoder relevance score
+	CombinedScore   float64        // Weighted combination of scores
+	Metadata        map[string]any // Preserved metadata
+	OriginalRank    int            // Position before reranking (1-indexed)
+	RerankRank      int            // Position after reranking (1-indexed)
+	RankImprovement int            // How much the rank improved (positive = moved up)
+}
+
+// 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 inputIdsTensor.Destroy()
+
+	attentionMaskTensor, err := ort.NewTensor(inputShape, attentionMaskData)
+	if err != nil {
+		return nil, fmt.Errorf("create attention_mask tensor: %w", err)
+	}
+	defer 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()
+
+	// 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 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 {
+	store      ObservationStore
+	calculator *Calculator
+	log        zerolog.Logger
+	interval   time.Duration
+	batchSize  int
+	stopCh     chan struct{}
+	doneCh     chan struct{}
+	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 {
+	mu                sync.Mutex
+	observations      []*models.Observation
+	scores            map[int64]float64
+	conceptWeights    map[string]float64
+	updateErr         error
+	getErr            error
+	getConceptsErr    error
+	updateScoresCalls int
+}
+
+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"`
+	Weight float64     `json:"weight"` // Weight for result merging (0.0-1.0)
+	Source string      `json:"source"` // Where this expansion came from
+	Intent QueryIntent `json:"intent"` // Detected intent
+}
+
+// Expander provides context-aware query expansion.
+type Expander struct {
+	embedSvc       *embedding.Service
+	vocabulary     []VocabEntry // Known vocabulary from observations
+	vocabVectors   [][]float32  // Embeddings for vocabulary entries
+	vocabMu        sync.RWMutex // Protects vocabulary
+	intentPatterns map[QueryIntent][]*regexp.Regexp
+}
+
+// VocabEntry represents a vocabulary term from observations.
+type VocabEntry struct {
+	Term   string  // The term itself
+	Weight float64 // How common/important this term is (0.0-1.0)
+	Source string  // Where it came from (title, concept, narrative)
+}
+
+// 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
+		minExpansions  int
+		hasOriginal    bool
+		expectedIntent QueryIntent
+	}{
+		{"question", "how do I implement auth", 1, true, IntentQuestion},
+		{"error", "fix the bug in login", 1, true, IntentError},
+		{"implementation", "implement user handler", 1, true, IntentImplementation},
+		{"architecture", "architecture design", 1, true, IntentArchitecture},
+		{"general", "database connection", 1, true, IntentGeneral},
+		{"empty", "", 0, false, IntentGeneral},
+	}
+
+	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
+		maxLen   int
+		expected string
+	}{
+		{"short", "hello", 10, "hello"},
+		{"exact", "hello", 5, "hello"},
+		{"long", "hello world", 5, "hello..."},
+		{"empty", "", 10, ""},
+	}
+
+	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/manager.go

@@ -35,20 +35,21 @@ 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
+	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
+	ExcludeSuperseded bool   // If true, exclude observations that have been superseded
 }
 
 // SearchResult represents a unified search result.
@@ -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/vector/sqlitevec/client.go

@@ -60,12 +60,15 @@ func (c *Client) AddDocuments(ctx context.Context, docs []Document) error {
 		return fmt.Errorf("generate embeddings: %w", err)
 	}
 
-	// Insert into vectors table
+	// Insert into vectors table with model version tracking
 	const insertQuery = `
-		INSERT OR REPLACE INTO vectors (doc_id, embedding, sqlite_id, doc_type, field_type, project, scope)
-		VALUES (?, ?, ?, ?, ?, ?, ?)
+		INSERT OR REPLACE INTO vectors (doc_id, embedding, sqlite_id, doc_type, field_type, project, scope, model_version)
+		VALUES (?, ?, ?, ?, ?, ?, ?, ?)
 	`
 
+	// Get current model version for tracking
+	modelVersion := c.embedSvc.Version()
+
 	tx, err := c.db.BeginTx(ctx, nil)
 	if err != nil {
 		return fmt.Errorf("begin transaction: %w", err)
@@ -104,6 +107,7 @@ func (c *Client) AddDocuments(ctx context.Context, docs []Document) error {
 			fieldType,
 			project,
 			scope,
+			modelVersion,
 		)
 		if err != nil {
 			return fmt.Errorf("insert document %s: %w", doc.ID, err)
@@ -114,7 +118,7 @@ func (c *Client) AddDocuments(ctx context.Context, docs []Document) error {
 		return fmt.Errorf("commit transaction: %w", err)
 	}
 
-	log.Debug().Int("count", len(docs)).Msg("Added documents to sqlite-vec")
+	log.Debug().Int("count", len(docs)).Str("model", modelVersion).Msg("Added documents to sqlite-vec")
 	return nil
 }
 
@@ -212,6 +216,7 @@ func (c *Client) Query(ctx context.Context, query string, limit int, where map[s
 			return nil, fmt.Errorf("scan row: %w", err)
 		}
 
+		r.Similarity = DistanceToSimilarity(r.Distance)
 		r.Metadata = map[string]any{
 			"sqlite_id":  float64(sqliteID), // Keep as float64 for compatibility
 			"doc_type":   docType.String,
@@ -252,3 +257,148 @@ func truncateString(s string, maxLen int) string {
 	}
 	return s[:maxLen] + "..."
 }
+
+// Count returns the total number of vectors in the store.
+func (c *Client) Count(ctx context.Context) (int64, error) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	var count int64
+	err := c.db.QueryRowContext(ctx, "SELECT COUNT(*) FROM vectors").Scan(&count)
+	if err != nil {
+		return 0, fmt.Errorf("count vectors: %w", err)
+	}
+	return count, nil
+}
+
+// ModelVersion returns the current embedding model version.
+func (c *Client) ModelVersion() string {
+	return c.embedSvc.Version()
+}
+
+// NeedsRebuild checks if vectors need to be rebuilt due to model version change.
+// Returns true if:
+// - The vectors table is empty
+// - Any vectors have a different model_version than the current model
+func (c *Client) NeedsRebuild(ctx context.Context) (bool, string) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	currentModel := c.embedSvc.Version()
+
+	// Check total count
+	var totalCount int64
+	err := c.db.QueryRowContext(ctx, "SELECT COUNT(*) FROM vectors").Scan(&totalCount)
+	if err != nil {
+		log.Warn().Err(err).Msg("Failed to count vectors for rebuild check")
+		return false, ""
+	}
+
+	if totalCount == 0 {
+		return true, "empty"
+	}
+
+	// Check for vectors with different model version
+	var staleCount int64
+	err = c.db.QueryRowContext(ctx,
+		"SELECT COUNT(*) FROM vectors WHERE model_version != ? OR model_version IS NULL",
+		currentModel,
+	).Scan(&staleCount)
+	if err != nil {
+		log.Warn().Err(err).Msg("Failed to count stale vectors")
+		return false, ""
+	}
+
+	if staleCount > 0 {
+		return true, fmt.Sprintf("model_mismatch:%d", staleCount)
+	}
+
+	return false, ""
+}
+
+// StaleVectorInfo contains information about a vector that needs rebuilding.
+type StaleVectorInfo struct {
+	DocID     string
+	SQLiteID  int64
+	DocType   string
+	FieldType string
+	Project   string
+	Scope     string
+}
+
+// GetStaleVectors returns doc_ids of vectors with mismatched or null model versions.
+// This enables granular rebuild - only re-embedding documents that need updating.
+func (c *Client) GetStaleVectors(ctx context.Context) ([]StaleVectorInfo, error) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	currentModel := c.embedSvc.Version()
+
+	query := `
+		SELECT doc_id, sqlite_id, doc_type, field_type, project, scope
+		FROM vectors
+		WHERE model_version != ? OR model_version IS NULL
+	`
+
+	rows, err := c.db.QueryContext(ctx, query, currentModel)
+	if err != nil {
+		return nil, fmt.Errorf("query stale vectors: %w", err)
+	}
+	defer rows.Close()
+
+	var results []StaleVectorInfo
+	for rows.Next() {
+		var info StaleVectorInfo
+		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)
+		}
+
+		info.SQLiteID = sqliteID.Int64
+		info.DocType = docType.String
+		info.FieldType = fieldType.String
+		info.Project = project.String
+		info.Scope = scope.String
+
+		results = append(results, info)
+	}
+
+	if err = rows.Err(); err != nil {
+		return nil, fmt.Errorf("iterate rows: %w", err)
+	}
+
+	return results, nil
+}
+
+// DeleteVectorsByDocIDs removes vectors by their doc_ids.
+// Used for granular rebuild - delete stale vectors before re-adding.
+func (c *Client) DeleteVectorsByDocIDs(ctx context.Context, docIDs []string) error {
+	if len(docIDs) == 0 {
+		return nil
+	}
+
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	// Build placeholder string
+	placeholders := make([]string, len(docIDs))
+	args := make([]interface{}, len(docIDs))
+	for i, id := range docIDs {
+		placeholders[i] = "?"
+		args[i] = id
+	}
+
+	// #nosec G201 -- Placeholders are "?" strings, actual values are parameterized via args
+	query := fmt.Sprintf("DELETE FROM vectors WHERE doc_id IN (%s)",
+		strings.Join(placeholders, ","))
+
+	_, err := c.db.ExecContext(ctx, query, args...)
+	if err != nil {
+		return fmt.Errorf("delete vectors by doc_ids: %w", err)
+	}
+
+	log.Debug().Int("count", len(docIDs)).Msg("Deleted stale vectors by doc_id")
+	return nil
+}

internal/vector/sqlitevec/client_test.go

@@ -38,7 +38,8 @@ func testDB(t *testing.T) (*sql.DB, func()) {
 			doc_type TEXT,
 			field_type TEXT,
 			project TEXT,
-			scope TEXT
+			scope TEXT,
+			model_version TEXT
 		)
 	`)
 	require.NoError(t, err)

internal/vector/sqlitevec/helpers.go

@@ -19,9 +19,32 @@ type Document struct {
 
 // QueryResult represents a search result from vector search.
 type QueryResult struct {
-	ID       string
-	Distance float64
-	Metadata map[string]any
+	ID         string
+	Distance   float64
+	Similarity float64 // 1.0 = identical, 0.0 = opposite (derived from distance)
+	Metadata   map[string]any
+}
+
+// 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/sync.go

@@ -240,3 +240,101 @@ 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/worker/handlers.go

@@ -4,13 +4,18 @@ package worker
 import (
 	"context"
 	"encoding/json"
+	"fmt"
 	"net/http"
+	"sort"
 	"strconv"
 	"time"
 
 	"github.com/go-chi/chi/v5"
 	"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"
@@ -641,6 +646,18 @@ func (s *Service) handleGetTypes(w http.ResponseWriter, r *http.Request) {
 	})
 }
 
+// handleGetModels returns available embedding models.
+func (s *Service) handleGetModels(w http.ResponseWriter, _ *http.Request) {
+	models := embedding.ListModels()
+	defaultModel := embedding.GetDefaultModel()
+
+	writeJSON(w, map[string]interface{}{
+		"models":  models,
+		"default": defaultModel,
+		"current": s.embedSvc.Version(),
+	})
+}
+
 // handleGetStats returns worker statistics.
 func (s *Service) handleGetStats(w http.ResponseWriter, r *http.Request) {
 	project := r.URL.Query().Get("project")
@@ -658,6 +675,22 @@ func (s *Service) handleGetStats(w http.ResponseWriter, r *http.Request) {
 		"ready":            s.ready.Load(),
 	}
 
+	// Add embedding model info
+	if s.embedSvc != nil {
+		response["embeddingModel"] = map[string]interface{}{
+			"name":       s.embedSvc.Name(),
+			"version":    s.embedSvc.Version(),
+			"dimensions": s.embedSvc.Dimensions(),
+		}
+	}
+
+	// Add vector count
+	if s.vectorClient != nil {
+		if count, err := s.vectorClient.Count(r.Context()); err == nil {
+			response["vectorCount"] = count
+		}
+	}
+
 	// Include project-specific observation count if project is specified
 	if project != "" {
 		count, err := s.observationStore.GetObservationCount(r.Context(), project)
@@ -715,15 +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
@@ -773,23 +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,
 	})
 }
 
@@ -857,6 +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)).
@@ -1015,6 +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()
 

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, err := strconv.Atoi(depthStr); err == 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, err := strconv.ParseFloat(confStr, 64); err == 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 +1,354 @@
+// 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"
+)
+
+// 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 err := observationStore.UpdateImportanceScore(r.Context(), id, newScore); err != nil {
+				// Log but don't fail - feedback was recorded
+				// Score will be updated on next recalculation cycle
+			}
+		}
+	}
+
+	// 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 err := recalculator.RefreshConceptWeights(r.Context()); err != nil {
+			// Log but don't fail - weight was saved
+		}
+	}
+
+	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 err := recalculator.RecalculateNow(r.Context()); err != nil {
+			// Log error but don't block response
+		}
+	}()
+
+	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 err := store.IncrementRetrievalCount(ctx, ids); err != nil {
+			// Log but don't fail - this is a background operation
+		}
+	}()
+}

internal/worker/service.go

@@ -15,6 +15,10 @@ import (
 	"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/sqlitevec"
 	"github.com/lukaszraczylo/claude-mnemonic/internal/watcher"
@@ -64,6 +68,12 @@ type Service struct {
 	observationStore *sqlite.ObservationStore
 	summaryStore     *sqlite.SummaryStore
 	promptStore      *sqlite.PromptStore
+	conflictStore    *sqlite.ConflictStore
+	patternStore     *sqlite.PatternStore
+	relationStore    *sqlite.RelationStore
+
+	// Pattern detection
+	patternDetector *pattern.Detector
 
 	// Domain services
 	sessionManager *session.Manager
@@ -75,6 +85,16 @@ type Service struct {
 	vectorClient *sqlitevec.Client
 	vectorSync   *sqlitevec.Sync
 
+	// Cross-encoder reranking (for improved search relevance)
+	reranker *reranking.Service
+
+	// Query expansion (for improved search recall)
+	queryExpander *expansion.Expander
+
+	// Importance scoring
+	scoreCalculator *scoring.Calculator
+	recalculator    *scoring.Recalculator
+
 	// HTTP server
 	router    *chi.Mux
 	server    *http.Server
@@ -177,6 +197,15 @@ 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)
@@ -186,6 +215,8 @@ func (s *Service) initializeAsync() {
 	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 - vector search disabled")
@@ -200,8 +231,32 @@ func (s *Service) initializeAsync() {
 		} else {
 			vectorClient = client
 			vectorSync = sqlitevec.NewSync(client)
-			log.Info().Msg("sqlite-vec vector search enabled")
+			log.Info().
+				Str("model", embedSvc.Version()).
+				Msg("sqlite-vec vector search enabled")
 		}
+
+		// 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, err := reranking.NewService(rerankCfg)
+			if err != nil {
+				log.Warn().Err(err).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)
@@ -225,11 +280,38 @@ 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
@@ -238,6 +320,22 @@ 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 {
@@ -282,6 +380,37 @@ 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")
@@ -294,6 +423,27 @@ func (s *Service) initializeAsync() {
 
 	// Start file watchers for auto-recreation on deletion
 	s.startWatchers()
+
+	// 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)
+		if needsRebuild {
+			log.Info().
+				Str("reason", reason).
+				Str("model", vectorClient.ModelVersion()).
+				Msg("Vector rebuild required")
+
+			if reason == "empty" {
+				// Full rebuild - vectors table is empty
+				s.wg.Add(1)
+				go s.rebuildAllVectors(observationStore, summaryStore, promptStore, vectorSync)
+			} else {
+				// Granular rebuild - only rebuild vectors with mismatched model versions
+				s.wg.Add(1)
+				go s.rebuildStaleVectors(observationStore, summaryStore, promptStore, vectorClient, vectorSync)
+			}
+		}
+	}
 }
 
 // startWatchers initializes and starts file watchers for database and config.
@@ -384,6 +534,15 @@ 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)
@@ -393,6 +552,8 @@ 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")
@@ -408,6 +569,34 @@ func (s *Service) reinitializeDatabase() {
 			vectorSync = sqlitevec.NewSync(client)
 			log.Info().Msg("sqlite-vec 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, err := reranking.NewService(rerankCfg)
+			if err != nil {
+				log.Warn().Err(err).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
@@ -422,6 +611,30 @@ 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
@@ -429,20 +642,44 @@ 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 {
@@ -565,6 +802,210 @@ func (s *Service) processStaleQueue() {
 	}
 }
 
+// rebuildAllVectors rebuilds all vectors from observations, summaries, and prompts.
+// Called when the vectors table is empty (e.g., after migration 20 drops all vectors).
+func (s *Service) rebuildAllVectors(
+	observationStore *sqlite.ObservationStore,
+	summaryStore *sqlite.SummaryStore,
+	promptStore *sqlite.PromptStore,
+	vectorSync *sqlitevec.Sync,
+) {
+	defer s.wg.Done()
+
+	log.Info().Msg("Starting full vector rebuild...")
+	start := time.Now()
+
+	var totalSynced int
+	var syncErrors int
+
+	// Rebuild observations
+	observations, err := observationStore.GetAllObservations(s.ctx)
+	if err != nil {
+		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")
+				syncErrors++
+			} else {
+				totalSynced++
+			}
+		}
+		log.Info().Int("count", len(observations)).Msg("Rebuilt observation vectors")
+	}
+
+	// Rebuild summaries
+	summaries, err := summaryStore.GetAllSummaries(s.ctx)
+	if err != nil {
+		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")
+				syncErrors++
+			} else {
+				totalSynced++
+			}
+		}
+		log.Info().Int("count", len(summaries)).Msg("Rebuilt summary vectors")
+	}
+
+	// Rebuild user prompts
+	prompts, err := promptStore.GetAllPrompts(s.ctx)
+	if err != nil {
+		log.Error().Err(err).Msg("Failed to fetch prompts for vector rebuild")
+	} else {
+		for _, prompt := range prompts {
+			if err := vectorSync.SyncUserPrompt(s.ctx, prompt); err != nil {
+				log.Warn().Err(err).Int64("id", prompt.ID).Msg("Failed to sync prompt during rebuild")
+				syncErrors++
+			} else {
+				totalSynced++
+			}
+		}
+		log.Info().Int("count", len(prompts)).Msg("Rebuilt prompt vectors")
+	}
+
+	elapsed := time.Since(start)
+	log.Info().
+		Int("total_synced", totalSynced).
+		Int("errors", syncErrors).
+		Dur("elapsed", elapsed).
+		Msg("Full vector rebuild complete")
+}
+
+// rebuildStaleVectors rebuilds only vectors with mismatched or unknown model versions.
+// This is more efficient than rebuilding all vectors when only some need updating.
+func (s *Service) rebuildStaleVectors(
+	observationStore *sqlite.ObservationStore,
+	summaryStore *sqlite.SummaryStore,
+	promptStore *sqlite.PromptStore,
+	vectorClient *sqlitevec.Client,
+	vectorSync *sqlitevec.Sync,
+) {
+	defer s.wg.Done()
+
+	log.Info().Msg("Starting granular vector rebuild for stale vectors...")
+	start := time.Now()
+
+	// Get all stale vectors
+	staleVectors, err := vectorClient.GetStaleVectors(s.ctx)
+	if err != nil {
+		log.Error().Err(err).Msg("Failed to get stale vectors")
+		return
+	}
+
+	if len(staleVectors) == 0 {
+		log.Info().Msg("No stale vectors found")
+		return
+	}
+
+	log.Info().Int("stale_count", len(staleVectors)).Msg("Found stale vectors to rebuild")
+
+	// Group stale vectors by doc_type and sqlite_id for efficient lookup
+	staleObsIDs := make(map[int64]bool)
+	staleSummaryIDs := make(map[int64]bool)
+	stalePromptIDs := make(map[int64]bool)
+	staleDocIDs := make([]string, 0, len(staleVectors))
+
+	for _, sv := range staleVectors {
+		staleDocIDs = append(staleDocIDs, sv.DocID)
+		switch sv.DocType {
+		case "observation":
+			staleObsIDs[sv.SQLiteID] = true
+		case "summary":
+			staleSummaryIDs[sv.SQLiteID] = true
+		case "prompt":
+			stalePromptIDs[sv.SQLiteID] = true
+		}
+	}
+
+	// Delete stale vectors before re-syncing
+	if err := vectorClient.DeleteVectorsByDocIDs(s.ctx, staleDocIDs); err != nil {
+		log.Error().Err(err).Msg("Failed to delete stale vectors")
+		return
+	}
+
+	var totalSynced int
+	var syncErrors int
+
+	// Rebuild stale observations
+	if len(staleObsIDs) > 0 {
+		ids := make([]int64, 0, len(staleObsIDs))
+		for id := range staleObsIDs {
+			ids = append(ids, id)
+		}
+
+		observations, err := observationStore.GetObservationsByIDs(s.ctx, ids, "date_desc", 0)
+		if err != nil {
+			log.Error().Err(err).Msg("Failed to fetch observations for 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")
+					syncErrors++
+				} else {
+					totalSynced++
+				}
+			}
+			log.Info().Int("count", len(observations)).Msg("Rebuilt stale observation vectors")
+		}
+	}
+
+	// Rebuild stale summaries
+	if len(staleSummaryIDs) > 0 {
+		ids := make([]int64, 0, len(staleSummaryIDs))
+		for id := range staleSummaryIDs {
+			ids = append(ids, id)
+		}
+
+		summaries, err := summaryStore.GetSummariesByIDs(s.ctx, ids, "date_desc", 0)
+		if err != nil {
+			log.Error().Err(err).Msg("Failed to fetch summaries for 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")
+					syncErrors++
+				} else {
+					totalSynced++
+				}
+			}
+			log.Info().Int("count", len(summaries)).Msg("Rebuilt stale summary vectors")
+		}
+	}
+
+	// Rebuild stale prompts
+	if len(stalePromptIDs) > 0 {
+		ids := make([]int64, 0, len(stalePromptIDs))
+		for id := range stalePromptIDs {
+			ids = append(ids, id)
+		}
+
+		prompts, err := promptStore.GetPromptsByIDs(s.ctx, ids, "date_desc", 0)
+		if err != nil {
+			log.Error().Err(err).Msg("Failed to fetch prompts for rebuild")
+		} else {
+			for _, prompt := range prompts {
+				if err := vectorSync.SyncUserPrompt(s.ctx, prompt); err != nil {
+					log.Warn().Err(err).Int64("id", prompt.ID).Msg("Failed to sync prompt during rebuild")
+					syncErrors++
+				} else {
+					totalSynced++
+				}
+			}
+			log.Info().Int("count", len(prompts)).Msg("Rebuilt stale prompt vectors")
+		}
+	}
+
+	elapsed := time.Since(start)
+	log.Info().
+		Int("total_synced", totalSynced).
+		Int("errors", syncErrors).
+		Dur("elapsed", elapsed).
+		Msg("Granular vector rebuild complete")
+}
+
 // verifyStaleObservation verifies a single stale observation in the background.
 func (s *Service) verifyStaleObservation(req staleVerifyRequest) {
 	// Wait for service to be ready
@@ -667,11 +1108,42 @@ func (s *Service) setupRoutes() {
 		r.Get("/api/stats", s.handleGetStats)
 		r.Get("/api/stats/retrieval", s.handleGetRetrievalStats)
 		r.Get("/api/types", s.handleGetTypes)
+		r.Get("/api/models", s.handleGetModels)
+
+		// 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)
 	})
 }
 
@@ -894,6 +1366,16 @@ 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()
+	}
+
 	// Shutdown all sessions
 	s.sessionManager.ShutdownAll(ctx)
 
@@ -904,6 +1386,13 @@ 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 {