Semantic versioning.
- OpenAPI contract implemented end-to-end (resumes, ATS evaluate, AIBF analyze, HR feedback, model retraining) plus an admin flagged-feed and health check.
- Bias-detection engine: linear reference model, zero-baseline SHAP attributions, bias score, thresholds, explanations, flagging.
- Human-in-the-loop retraining with feedback-driven threshold recalibration.
- Synthetic, deterministic training/demo data (no real PII).
- Test suite (10 checks), runnable pipeline demo, Docker/Compose, docs.