# CAIN VIGIL — Test Report

| Suite | Command | Result |
|---|---|---|
| VIGIL tests | `pytest tests/test_vigil.py` | **13/13 passed** |
| Formal properties | `run_property_suite9()` | **15/15 passed** |
| Adversarial lab | `run_vigil_lab(cases=100000)` | **100,000 executed / 100,000 blocked / 0 violations**, 32 categories |
| Mutations | `run_vigil_mutations()` | **12/12 killed** |
| Regression (hermes/atlas/continuum) | `pytest tests/test_hermes.py tests/test_atlas.py tests/test_atlas_federation.py tests/test_cain_continuum_*.py` | **113/113 passed** |

## Exercises
state machine monotonicity/revocation/recovery; envelope bounds; uncertainty/observability budgets; release decisions
(envelope, unobservable, irreversible, drift, capability mismatch); drift + change point; world-state false consensus;
prediction + velocity; containment + override; fleet correlated failure; physical SIMULATED; end-to-end evaluator;
property suite; lab; mutations; hosted stages; HTTP surface.

## Lab categories (32)
slow_drift, fast_drift, concept_drift, distribution_shift, sensor_spoofing, world_model_poisoning, memory_poisoning,
tool_poisoning, policy_poisoning, delegation_laundering, authority_laundering, collusion, sybil_agents,
false_consensus, evidence_correlation, observability_loss, monitor_evasion, adversarial_adaptation, reward_hacking,
goal_drift, capability_escalation, self_modification, model_replacement, skill_replacement, environment_change,
time_of_check_time_of_use, uncertainty_laundering, reversibility_masking, consequence_escalation, stale_evidence,
containment_bypass, quarantine_escape.

## Limits
Same-project tests only; no third-party audit; prediction is a heuristic; physical SIMULATED.
