API
FastAPI endpoints cover environment lifecycle, orchestration, policy inference, evaluation, training metrics, case retrieval, and evidence queries.
Product Inference
POST /policy/inferselects one legal candidate using the active Transformers model when available, with deterministic safety-ranker fallback.GET /policy/model_statusreports active-model configuration and load status.
The active model is controlled by checkpoints/active/active_model_manifest.json, generated by scripts/activate_sweep_model.py. The default production path prefers the GRPO adapter for qwen-qwen2-5-0-5b-instruct, then falls back to the merged Qwen 0.5B artifact or SFT adapter if required.
Example:
curl http://127.0.0.1:8200/policy/model_status
curl -X POST http://127.0.0.1:8200/policy/infer