v0.4: Add filter_supported_claims_by_relevance to normalize extraction across model sizes

Upload src/verifier.py

v0.3.2: malformed line tracking, trace logging behind flag

v0.3.2: realistic mock, rebuild_errors logging, 100 gold cases (167 tests pass)

v0.3.2: code-quality patch — dedup fix, filter_stats, schema validation, tighter prompt (167 tests)

v0.3.1: inference detector (4-tier), contradiction question-fallback, improved batch prompt (154 tests pass)

v0.3: fix all reviewed issues (bugs #1-3, issues #4-8, 125 tests pass)

v0.3: 2-call batch verifier, numeric consistency check, slot-aware relevance filter

v0.2.1: claim quality filter, relevance check, stricter over-abstention, grounded_accept_rate

v0.2: split verifier, span_matcher, calibration, frame-based contradictions

v0.1.4: frame-based contradictions, span-match upgrade, false_contradiction_rate, concise hypothesis

v0.1.2b: improved verifier prompt + deferral postprocessing for partial_answer_coverage

v0.1.2: slot-typed contradictions, NOT_IN_EVIDENCE postprocess, partial_hypothesis, punctuation fix

v0.1 patch: strengthen verifier/gate/metrics

Add src/verifier.py