# CAIN ORACLE — Reality Override & Simulator Admissibility

`cain/oracle/reality.py`.

## Simulator admissibility
A simulator must not become trusted merely because it produced a prediction. `SimulationAdmissibilityEngine.assess`
evaluates provenance, version, calibration, known failure modes, environment/action/state-space coverage, real-world
validations, sim-to-real error, age and unsupported assumptions → `ADMISSIBLE · CONDITIONALLY_ADMISSIBLE · DEGRADED ·
STALE · UNSUPPORTED · REJECTED`, each with an authority ceiling. `may_authorize` requires `ADMISSIBLE` for HIGH/CRITICAL
(P184). A prediction from a rejected simulator cannot silently authorize a high-consequence action.

## Reality override
**OBSERVED REALITY OUTRANKS PREDICTED REALITY.** `RealityOverrideEngine.reconcile(prediction, observation)`:

- an untrustworthy observation (`integrity_ok=False`, unsourced, or stale) does **not** override (P186);
- a contradicting observation invalidates affected authorization and drives: STOP_TRUSTING_PREDICTION,
  INVALIDATE_AFFECTED_AUTHORIZATION, RECALCULATE_CONSEQUENCE/AUTHORITY, GENERATE_EVIDENCE, UPDATE_MODEL_CALIBRATION,
  and (for high/critical or irreversible) INITIATE_RECOVERY + REVOKE_ADDITIONAL_AUTHORITY (P185).

## Status
IMPLEMENTED, TESTED.
