perf(planner): trim relevant tables (+cache) to cut latency; keep repair loop & rich traces

refactor(core): trace schema upgrade, verifier/executor sync, benchmark plot polish

feat(trace): enrich StageTrace (sql_length/row_count/verified/error_type/repair_attempts/skipped) and propagate in normalization; tag EmptySQL; annotate repair attempts

feat(bench): gold-aware EM/SM/ExecAcc + p50/p95; write per-stage means; richer plots

feat(core): refine pipeline & verifier; improve Spider benchmark accuracy

feat(bench): auto-detect latest run and plot per-stage latency + metrics summary

feat(core): always emit per-stage traces in Pipeline.run (fallback when StageResult.trace is empty)

chore(factory): safely load .env via dotenv (with fallback under CI)

chore(factory): safely load .env via dotenv (with fallback under CI)

fix(grafana): move nl2sql.json into provisioning folder and fix dashboard mount path

fix(grafana): move nl2sql.json into provisioning folder and fix dashboard mount path

feat(observability): add Prometheus-Grafana stack with auto-provisioning and docs

feat(metrics): instrument per-stage and pipeline_total latency; count pipeline_runs_total (ok/error/ambiguous)

feat(metrics): instrument per-stage and pipeline_total latency; count pipeline_runs_total (ok/error/ambiguous)

chore(pre-commit): add types-PyYAML to mypy additional_dependencies; clean & rewire hooks

feat(metrics): integrate Prometheus; expose /metrics; instrument safety & verifier

refactor(core): DI-ready Pipeline; add registry + YAML factory + typed trace/result

refactor(core): DI-ready Pipeline; add registry + YAML factory + typed trace/result

fix(verifier): robust aggregate detection and projection-level semantic check

feat(trace): standardize StageTrace (add summary) and coerce duration_ms to int at API boundary

refactor(core): DI-ready Pipeline; add registry + YAML factory + typed trace/result

refactor(core): DI-ready Pipeline; add registry + YAML factory + typed trace/result

test-mode: stub runner in router; factory stubs accept positional calls

fix(router): lazy pipeline build for pytest; fix(factory): stubs accept positional/keyword args

refactor(core): DI-ready Pipeline; add registry + YAML factory + typed trace/result

refactor(core): DI-ready Pipeline; add registry + YAML factory + typed trace/result

refactor(core): DI-ready Pipeline; add registry + YAML factory + typed trace/result

refactor(core): DI-ready Pipeline; add registry + YAML factory + typed trace/result

refactor(core): DI-ready Pipeline; add registry + YAML factory + typed trace/result

fix(api): make traces.duration_ms an int via TraceItem model; coerce in handler

fix(router): coerce trace.duration_ms to int in response (_round_trace)

test(router): add full coverage for nl2sql_handler (clarify, error, success, db_id, crash, invalid-type, 422, trace rounding)

refactor(router): inject pipeline via Depends(get_runner); preserve error dump logging

test(router): use _RUN hook to fake pipeline.run in router tests

test(router): monkeypatch _pipeline.run instead of nl2sql.Pipeline.run

fix(router): re-export Pipeline via alias to satisfy test monkeypatch; remove duplicate import

fix(router): avoid Pipeline redefinition; expose Pipeline via import for test monkeypatch

test(benchmarks): add black-box tests for evaluate_spider outputs and trace normalization

feat(benchmarks): add pro evaluator with EM, structural match, execution accuracy, and safety consistency metrics

fix(pipeline): allow optional schema_preview in run() for better typing compatibility with benchmarks

feat(benchmarks): align Spider eval with config-driven Pipeline and native Safety; log per-stage trace; add CSV summary

feat(benchmarks): align Spider eval with config-driven Pipeline and native Safety; log per-stage trace; add CSV summary

feat(safety): harden SQL validation (multi-CTE, recursive WITH, unicode normalization, precise errors, EXPLAIN gate)

docs(readme): add evolution note and improve intro narrative

fix(types): avoid mypy no-redef in nl2sql_handler by predeclaring pipeline_obj

chore(pre-commit): dedupe mypy and add types-requests; split commit vs push hooks

chore(pre-commit): add types-requests to mypy hook

fix(api): map FinalResult → HTTP (200/400) and stabilize nl2sql handler; prevent 500s in tests

fix(api): map FinalResult → HTTP (200/400) and stabilize nl2sql handler; prevent 500s in tests

chore(pre-commit): split auto-fix on commit vs checks on push; run mypy/pytest on push