# CAIN CRUCIBLE — Proof of Control

`cain/crucible/proof.py`. The central CRUCIBLE primitive.

A `ProofOfControl` binds: control + digest, policy/authority/runtime/model versions, agent identity, adversarial
scenario, execution path, environment, test-input digest, expected prohibited effect, actual outcome, effect-level
outcome, observations, evidence, harness identity, result, nonce and timestamp.

It answers: WHAT CONTROL WAS TESTED? WHAT WAS ATTACKED? WHAT PATH? WHAT SHOULD HAVE HAPPENED? WHAT ACTUALLY HAPPENED?
DID THE PROHIBITED EFFECT OCCUR? WHAT EVIDENCE SUPPORTS THE RESULT?

## Results
`CONTROL_HELD · CONTROL_HELD_WITH_LIMITATION · CONTROL_DEGRADED · CONTROL_FAILED · UNKNOWN · NOT_TESTABLE`.

## Effect-level outcomes (critical)
`BLOCKED_BEFORE_EFFECT · BLOCKED_AT_EFFECT_BOUNDARY · PARTIALLY_EXECUTED · EFFECT_OCCURRED · EFFECT_UNKNOWN ·
EFFECT_VERIFIED_SAFE · EFFECT_VIOLATION`. A control that logs an attack while allowing the effect is **not effective**.

## Verification
`verify_proof` recomputes the digest and signature, checks required bindings, result and effect-outcome enums, and
rejects `PROOF_HELD_BUT_EFFECT_OCCURRED` (a proof cannot claim CONTROL_HELD while the prohibited effect occurred), plus
nonce replay and control mismatch (P161, P162; lab `fake_proof`, `held_but_effect`, `replay_proof`).

## Status
IMPLEMENTED, TESTED, LIVE (the `crucible_proof` restriction-only hosted stage).
