# CAIN EPICENTER — Test Report

| Suite | Command | Result |
|---|---|---|
| EPICENTER tests | `pytest tests/test_epicenter.py` | **12/12 passed** |
| Formal properties | `run_property_suite_epicenter()` | **30/30 passed (P211–P240)** |
| Adversarial lab | `run_epicenter_lab(cases=100000)` | **99,995 executed / 99,995 blocked / 0 violations**, 35 categories |
| Mutations | `run_epicenter_mutations()` | **24/24 killed** |
| Mission chaos | `MissionChaosEngine.run()` | **21/21 faults fail-closed** |
| Clean-room verifier | `python3 cain-mission-verify/verify.py contract\|certificate\|bundle` | valid accepted; tampered rejected; bundle recomputes |
| Latency benchmark | `python3 scripts/cain42_evolution/bench_epicenter.py` | per-op p50/p95/p99 in `artifacts/epicenter/EPICENTER_BENCH.json` |
| Live wiring | `install_hosted_pipeline()` | **default_stages_v20 installed by default (includes 3 mission stages)** |

## What is exercised
mission contract integrity/binding/expiry/authority; mission graph four-level separation; evidence-gated state machine;
success predicates + agent-claim rejection; evidence sufficiency (independent roots) + claim-evidence graph;
false-completion defense (13 modes); drift + objective substitution; governed replan + plan differential; authority
budget + decay; recovery governance + false-rollback rejection; completion certificate semantics (missing subgoals,
insufficient evidence, untrustworthy prediction, tamper, replay); trajectory automaton + next-step risk; multi-agent
authority conservation + collective-completion defense; chaos fail-closed; clean-room reproduction; hosted stages.

## Limits
Same-project tests only; no third-party validation; mission reliability is not a guarantee; no real physical system;
the lab and mutations are self-defined by the project; the benchmark is a local in-process micro-benchmark.
