Deploy: dashboard naive-vs-rigorous reveal + sync serving to origin/main

calibrate(jury): v1.1+v1.1.1 — fix weighting bugs; recency-position paraphrase clause

fix(calibration): per-corpus dispatch in generate-outputs (#19)

fix(judges,calibration,harness): three Codex adversarial-review findings

fix(judges,calibration): five review follow-ups (items 5, 6, 7, 9, 10)

fix(judges): four review-blocking bugs (review items 1–4 + 8)

refactor(harness): migrate to per-dimension Judge layer (drop faithfulness/correctness)

feat(config): add evaluation.judge_dimensions field

feat(calibration): generate_kappa_table with strict/warn modes

feat(scripts): run_calibration.py orchestrator for Steps A/C/D

test(calibration): sklearn-parity fixtures + cross-check CI test

feat(calibration): hand-rolled cohen_kappa, gwets_ac2, bootstrap_ci

feat(variance): PermutedJudge + Jury — N permutations and multi-judge aggregator

feat(judges): CitationFaithfulnessJudge with all-or-nothing aggregation

feat(judges): CompletenessJudge + three-point reference-based rubric

feat(judges): RelevanceJudge + anchored three-point rubric

feat(judges): GroundednessJudge + anchored binary rubric

feat(judges): _call_judge_with_retry helper with strict-reprompt + abstain

feat(judges): MockJudge with LookupError on missing keys

feat(judges): Judge ABC with judge_id derived from model + dimension

feat(judges): Rubric markdown loader with aggressive validation

feat(judges): ScoreResult + abstain-reason constants

test: scaffold tests/evaluation/ directory for judge-layer tests

docs+test: round-2 incident response — Google API key format scrub

feat(eval): Week 1 step 5 — 25-question K8s golden dataset + grounded_refusal fix

docs(eval): Fix 2 SearchTool query expansion — attempted and reverted

feat: extend GoldenQuestion with source_pages and source_sections

fix: batch-3 adversarial review findings

refactor: address batch-2 review feedback

feat: parameterized system prompt template

feat: SSE meta event carries corpus + corpus_label

feat: per-request corpus routing via Literal validation

security: fail-closed on secret extraction and env var leakage

refactor: address batch-1 review feedback

feat: support multi-corpus golden dataset schema

feat: multi-corpus construction with nested corpus×provider composition

feat: add CorpusConfig for multi-corpus support

fix: stream stage events live, thread source_chunks, fix LangChain wrapper

style: fix ruff lint — import sorting, line length

test: update security integration mock for _orchestrator_done event

feat: route handler emits meta, injection, output_validation SSE events

fix: address remaining review issues (3/5)

feat: orchestrator.run_stream emits per-stage SSE events

feat: enrich SearchTool metadata with scores, previews, PII count

feat: expose reranker scores through retrieval pipeline

fix: ruff lint — import sorting, unused imports, line length, naming

fix(security): buffer stream for output validation, store filtered answer

fix(security): output validation on /ask/stream, correct audit endpoint

fix(security): add security to /ask/stream, wire PII redactor into SearchTool

feat(security): wire injection detection, output validation, audit into pipeline