Wayfinder sits above control and below intelligence. It does not retrain your models or redesign your autonomy stack. It observes, assesses, modulates, enforces — and signs the record.
Read left to right. Policy envelopes enter from above. The evidence artifact exits to the regulator below.
Context ingestion
The governance layer reads the adaptive system's full decision context — inputs, model state, recent parameter deltas, and the operational envelope it is supposed to act within. Read-only. Zero interference with the underlying stack.
Envelope check
Each proposed decision is evaluated against the active regulatory and operational envelope — rate filings, ASIL targets, PCCP boundaries, fairness constraints, reserve policies. Breach margin is measured, not guessed.
Bounded authority shaping
Where the proposed action is inside the envelope, it passes. Where it approaches a boundary, the layer reduces authority — throttling, narrowing, or deferring — rather than vetoing outright. Graceful degradation, not a circuit breaker.
Runtime intervention
Out-of-envelope actions are intercepted before they reach actuation or bind. Learning updates that would move the model outside safety properties are rolled forward to the last compliant version. The system stays running; it just stays safe.
Hash-chain signature
Every decision — passed, modulated, or intercepted — is signed and appended to a tamper-evident audit chain. The chain is the evidence artifact. It is what the regulator, the auditor, and the reinsurer all read from.
One record per governed action. Human-readable, machine-readable, regulator-readable. Cryptographically chained. Exported in whatever filing language the authority on the other side of the table speaks.
We do not replace your underwriting engine, your autonomy stack, or your model catalogue. Wayfinder runs alongside — reading state, signing decisions, and producing evidence. Removal is a configuration change, not a re-architecture.
Fifteen minutes. What plugs in on your side. What the pilot looks like.
