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

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

feat(calibration): six row configs for the κ ablation table

feat(eval): K8s refusal_threshold sweep against 25Q set — 0.015 validated

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

chore(eval): pin gpt-4o-mini snapshot + wire fastapi golden_dataset + pre-commit tolerances

feat(eval): K8s refusal_threshold 0.02 → 0.015 empirical calibration

feat: K8s pilot corpus — 8 pages + config entry + JSON rewrite

fix: batch-3 adversarial review findings

feat: K8s corpus config entry, ingestion target, curation policy

feat(security): add security config models to AppConfig

feat: infrastructure sprint — vLLM/Modal, Helm, Terraform (#8)

feat: Anthropic Haiku benchmark + README with provider comparison

feat: add SQLite conversation sessions with session_id

fix: production config with reranker disabled for 512MB free tier

chore: surface retry and rate_limit_rpm config in default.yaml

feat: add cross-encoder reranking with feature flag

feat: add grounded refusal gate based on retrieval score threshold

feat: Day 1 — repo scaffolding, provider abstraction, config, tests