=== Judo Helper Eval ===

Gold set:  eval/gold/full.json (150 items)
Pipeline:  standard
Chunking:  naive
Abstain:   threshold
Backend:   ollama
RAGAS judge: enabled (slow pass)
DB:        data/judo-docs-union.sqlite

Running pipeline on gold set...

Running RAGAS faithfulness judge (claim extraction + NLI)...

Metrics:
  [FAIL] faithfulness: 0.559 (threshold 0.9) — Heuristic term overlap (CI regression only; not paper faithfulness)
  [PASS] context_precision: 0.911 (threshold 0.8) — Share of retrieved chunks containing reference key terms
  [PASS] context_recall: 0.803 (threshold 0.8) — Reference key-term coverage in retrieved contexts
  [PASS] answer_relevancy: 0.806 (threshold 0.8) — Question key-term overlap in answer (or refusal when expected)
  [PASS] faithfulness_judge_all: 0.923 (threshold 0.9) — Claim-level NLI vs retrieved context (judge: qwen2.5:14b); all items
  [PASS] faithfulness_judge_answered: 0.914 (threshold 0.9) — Claim-level NLI vs retrieved context (judge: qwen2.5:14b); answered items with extracted claims (n=123)
  [PASS] citation_precision: 1.000 (threshold 0.9) — No citations found in answers
  [PASS] citation_recall: 1.000 (threshold 0.8) — Alias of citation_precision for pilot harness
  [FAIL] hallucination_rate: 1.000 (threshold 0.98) — 123/123 answers with unverified or missing citations
  [PASS] red_team_refusal: 1.000 (threshold 1) — 26/26 red-team items refused safely
  [PASS] negative_rejection: 1.000 (threshold 1) — Alias for red_team_refusal on refusal-required items
  [PASS] id_recall: 0.992 (threshold 0.85) — 123/124 items retrieved at least one anchor

Stratified aggregates (SA-23):
  [FAIL] faithfulness_factual: 0.467 n=124 (threshold 0.9) — Heuristic faithfulness on factual gold (expect_refusal=false, non-meta-pointer)
  [PASS] abstention_accuracy: 1.000 n=26 (threshold 1) — Correct refusal when expect_refusal=true
  [exploratory] faithfulness_meta_pointer: 1.000 n=0 — Heuristic faithfulness on meta-pointer gold (exploratory; excluded from CI gate)

Per-tier breakdown (W12-T two-tier policy):
  critical: n=31 answer_rate=0.645 refusal_rate=0.355
    faithfulness: 0.648 — Heuristic term overlap (CI regression only; not paper faithfulness)
    context_precision: 0.900 — Share of retrieved chunks containing reference key terms
    context_recall: 0.845 — Reference key-term coverage in retrieved contexts
    answer_relevancy: 0.851 — Question key-term overlap in answer (or refusal when expected)
    citation_precision: 1.000 — No citations found in answers
    citation_recall: 1.000 — Alias of citation_precision for pilot harness
    hallucination_rate: 1.000 — 20/20 answers with unverified or missing citations
    red_team_refusal: 1.000 — 11/11 red-team items refused safely
    negative_rejection: 1.000 — Alias for red_team_refusal on refusal-required items
    id_recall: 1.000 — 20/20 items retrieved at least one anchor
  standard: n=119 answer_rate=0.866 refusal_rate=0.134
    faithfulness: 0.536 — Heuristic term overlap (CI regression only; not paper faithfulness)
    context_precision: 0.913 — Share of retrieved chunks containing reference key terms
    context_recall: 0.795 — Reference key-term coverage in retrieved contexts
    answer_relevancy: 0.795 — Question key-term overlap in answer (or refusal when expected)
    citation_precision: 1.000 — No citations found in answers
    citation_recall: 1.000 — Alias of citation_precision for pilot harness
    hallucination_rate: 1.000 — 103/103 answers with unverified or missing citations
    red_team_refusal: 1.000 — 15/15 red-team items refused safely
    negative_rejection: 1.000 — Alias for red_team_refusal on refusal-required items
    id_recall: 0.990 — 103/104 items retrieved at least one anchor

Overall CI gate: FAIL