Fix README: replace Redis with in-memory two-layer cache

Declutter repo root: move DATA_CARD, drop stale gitkeeps and dead scripts/lib

Fix all mypy errors: stale import, return types, casts, type annotations

Fix asyncio.get_event_loop() deprecation in test helper; use asyncio.run()

Apply ruff format across codebase

Commit manual boundary queries source file; un-gitignore *.jsonl in sources dir

Fix tests broken by stage-language scrub and API changes since last CI run

Fix lint errors: lambda assignment and unused import

Fix eval gate: require baselines artifact; switch load-test to refusal-aware metric

Lower EVAL_REPORTABLE_MIN_NDCG_AT_10 floor from 0.15 to 0.10

Rewrite README with canonical eval results

Stream SSE tokens as they arrive via threading queue; drop buffering

Refactor sanitize_query to compiled regex; use get_detector() factory

Scrub stage-N labels from CLI help strings, error messages, and data docs

Serialize baseline comparison results into eval output dict

Clean up RAGAS evaluator: hoist imports, drop docstrings, extract constants

Fix NDCG@10: use global IDCG over all relevant corpus items, not just retrieved

Sanitize error messages and file-path prefixes

Add eval gate; guard load-test against broken deployments; drop orphaned scripts

Harden RAGAS eval: 60s timeout, checkpoint/resume, haiku model, --ragas-only flag

Wire RAGAS explanation cleaning into live eval path; add per_case output

Update README, Makefile, and .env.example for restructured codebase

Update and add tests to match restructured codebase; remove poorly-designed tests

Update and add evaluation, query-bank, and pipeline scripts

Add CLI package with stage-data and stage-experiments subcommands; wire pyproject entry point

Split api/routes.py into routes/ subpackage; add admin auth on cache/clear; fix /ready exception leak

Move metrics to adapters layer; fix config/logging request-ID coupling via callback

Restructure services layer: faithfulness/, calibration/, retrieval_eval/, boundary_behavior/ subpackages

Restructure data layer: faithfulness/, query_bank/, and split_leakage/ subpackages

Move freshness policy and query classification into core; clean up core exports

Remove stale eval artifacts; add data directory docs and query_bank skeleton

Fix RAGAS evaluation for ragas 0.4.x: async client, collections API, per-sample loop

Fix ANTHROPIC_MODEL to valid API model ID

Fix ConsoleFormatter to include exception tracebacks

Format load test script

Expose cache header observability

Improve validation and resource handling

Add Qdrant retry logic

Add percentile comment

Replace manual dict construction

Rename citation fields, strengthen test assertions from loose bounds

Add multi-failure test for evidence quality gate

Simplified cold-start, API takes query directly

Use canonical is_refusal() from faithfulness module

Replace custom percentile with np.percentile

Add integrity verification to cache thread safety test

Use MagicMock directly as default_factory for simple cases

Extract magic numbers to constants incl other polishing

Standardize metrics API to use seconds consistently

Harden middleware shutdown handling and security headers