feat: LLM accuracy testing framework with 17-test catalog

Framework runs a category/subcategory test catalog against multiple
LLMs through an OpenAI-compatible endpoint. Deterministic evaluators
score each response. Reports render as table, markdown, or JSON.

internal/report/json.go

@@ -0,0 +1,60 @@
+package report
+
+import (
+	"encoding/json"
+	"io"
+
+	"github.com/lukaszraczylo/llm-testbench/internal/runner"
+	"github.com/lukaszraczylo/llm-testbench/internal/testkit"
+)
+
+// jsonResult is the JSON-serializable projection of a runner.Result, joined
+// with its test's category/subcategory for self-contained output.
+type jsonResult struct {
+	Model            string  `json:"model"`
+	TestID           string  `json:"test_id"`
+	Category         string  `json:"category,omitempty"`
+	Subcategory      string  `json:"subcategory,omitempty"`
+	Detail           string  `json:"detail,omitempty"`
+	Error            string  `json:"error,omitempty"`
+	Score            float64 `json:"score"`
+	LatencyMS        int64   `json:"latency_ms"`
+	PromptTokens     int     `json:"prompt_tokens"`
+	CompletionTokens int     `json:"completion_tokens"`
+	Tokens           int     `json:"tokens"`
+	Skipped          bool    `json:"skipped"`
+}
+
+// renderJSON dumps raw results as a JSON array, joined with test metadata.
+func renderJSON(w io.Writer, tests []testkit.Test, results []runner.Result) error {
+	testByID := make(map[string]testkit.Test, len(tests))
+	for _, t := range tests {
+		testByID[t.ID] = t
+	}
+
+	out := make([]jsonResult, 0, len(results))
+	for _, r := range results {
+		t := testByID[r.TestID]
+		jr := jsonResult{
+			Model:            r.Model,
+			TestID:           r.TestID,
+			Category:         t.Category,
+			Subcategory:      t.Subcategory,
+			Score:            r.Score.Value,
+			Detail:           r.Score.Detail,
+			Skipped:          r.Score.Skipped,
+			LatencyMS:        r.Latency.Milliseconds(),
+			PromptTokens:     r.PromptTokens,
+			CompletionTokens: r.CompletionTokens,
+			Tokens:           r.TotalTokens(),
+		}
+		if r.Err != nil {
+			jr.Error = r.Err.Error()
+		}
+		out = append(out, jr)
+	}
+
+	enc := json.NewEncoder(w)
+	enc.SetIndent("", "  ")
+	return enc.Encode(out)
+}