Expand description
Chat-lowering bridge for verified model-precision contracts.
The bridge runs while assembling a request, before tokenisation and outside the decode hot path. It never invents a contract: both core registries must agree on profile ID, version, and specification identity.
Structs§
- Applied
Precision - Receipt returned when a verified precision contract changes a chat request.
Functions§
- apply_
active_ model_ precision - Resolve and apply the model’s verified runtime contract. A missing or stale contract is deliberately a no-op so ordinary chats retain their established behaviour until calibration is explicitly promoted.
- apply_
contract_ to_ parts - Apply verified model-precision contract across explicitly structured request parts.
- apply_
contract_ to_ prompt - Apply one already-verified contract. Public for deterministic tests and
alternate client surfaces; normal chat should use
apply_active_model_precision.