Changelog

All notable changes to this project are documented in this file.

[Unreleased]

Research wizard (/research) — guided health → train → trust → leakage → external validation → export
Paper curves — ROC / PR / calibration points in evaluation_report.json; Analytics Chart.js plots; GET /v1/reports/curves; bootstrap AUC CIs + quality notes
Analytics — sex/prevalence charts; per-chart and bulk PNG export; print/PDF layout
Predict — export session JSON (features + score)
Trust packaging — per-run trust_pack.json; run-scoped leakage/SHAP/external-validate/ZIP/methods
Live e2e (tests/test_e2e_features.py) and optional Playwright smoke (web/e2e)
Docs/book/paper notes updated for the complete research loop

Working package openhealth with CLI ehr-ai (start, train, doctor, …)
Task YAML presets under tasks/ (diabetes, heart failure, custom)
Dataset health checks, schema map-preview/import, Config Center (config/workspace.yaml)
Named experiment runs under reports/runs/ with promote-to-active
Multi-model compare, leakage audit, SHAP, fairness job paths
Thin OMOP / FHIR subset adapters; clinical-research worklist prototype
Honest scope docs: LIMITATIONS.md, WHY_THIS_FRAMEWORK.md

Primary UI: Angular researcher workbench (web/) — Streamlit legacy UI removed
FastAPI researcher/framework routes
Docker Compose one-command stack (UI :8080, API :8000)
Downloadable results ZIP; academic verification via local research-paper/ package (make -C research-paper paper-quick)

Synthetic software-verification track under local-only research-paper/reports/ (package is gitignored)
MIMIC lock path: make -C research-paper mimic-lock + docs/mimic_lock_checklist.md (credentialed extract required; never commit PHI)
Citation metadata: CITATION.cff