# Evolution 16 — release gates

Statuses: PASS, FAIL, BLOCKED, OPEN, NOT APPLICABLE. A PASS means the named evidence exists and the named command
succeeded on the author's machine. It never means independently reproduced or externally reviewed.

| Gate | Item | Status | Evidence / reason |
|---|---|---|---|
| A | Baseline reproducibility | **PASS** | `baseline.md`: Evolution 15 suites 44/44 before any change and on a clean checkout. Noted: 12 historical tests fail on a clean checkout because the compiled native enforcer is not in git |
| B | Evolution 15 regression preservation | **PASS** | `regression_results.json`: 44 passed; long-horizon rerun 0 / 550 attacks, 200 / 200 benign; Evolution 13 fix check exit 0; historical 500 passed |
| C | Mission-contract integrity | **PASS** | 4 contract scenarios x 30, thread test, classifier properties, mutations M16-07..10, 23, 25 |
| D | Authorization-to-effect integrity | **PASS** (in the simulator) | 0 / 690 bad outcomes; exploration I1-I3; verifier `admission.*`, `dispatch.*`. **Not wired to MCPGate** |
| E | Transaction recovery and idempotency | **PASS** (in the simulator) | crash sweep 71 / 71, 0 duplicate writes; F16-01 fixed |
| F | Self-improvement admission controls | **OPEN** | nothing built; the existing lifecycle's tests pass (149 with the mission suite) but it was not audited or attacked here |
| G | Mission drift detection | **OPEN** | only an evidence validity window and the F16-02 fix were added; no new drift detection; no model in the loop |
| H | Delegation and revocation integrity | **PASS** for the durable path; **OPEN** overall | dispatch-time re-check and the verifier's independent chain check pass; F15-03/04 (direct callers of the graph and supervisor) remain open |
| I | Independent evidence verification | **PASS** as a clean-room program; **OPEN** as independence | 570 / 570 verdicts as specified; no CAIN imports. Same author, same project; no outside party has run it |
| J | Adversarial evaluation reproducibility | **PASS** | seeded; `EVOLUTION16_REPRODUCE.sh`; not reproduced by anyone else |
| K | Formal-property coverage | **OPEN** | bounded exploration (16,104 sequences), 25 / 25 mutants, differential tests. No model checker, no proof |
| L | Top-100 registry integrity | **PASS** | registry unchanged; mappings only (`top100_delta.md`) |
| M | Invention deduplication and prior-art documentation | **PASS** with a stated limit | six families, each with overlaps listed; three candidates rejected. Prior art read at abstract level; **no patent database search** |
| N | IP valuation arithmetic and evidence discipline | **PASS** | arithmetic shown; method unchanged; v46 untouched; all six families E4, none higher |
| O | Supply-chain provenance | **OPEN** | manifest with SHA-256 of sources and artifacts; **self-asserted, unsigned**, files uncommitted; no SBOM, no build attestation; no new dependency |
| P | Performance and reliability | **OPEN** | measured: durable journaling is about 28x slower at the median (5.0 ms vs 0.18 ms); no soak, no load test, no multi-process test |
| Q | External review | **BLOCKED** | no independent reviewer has looked at any of this |
| R | Production deployment readiness | **BLOCKED** | not wired to any live path; single file on a single host; open items U16-01..08; Q blocked |

**Production: NOT APPROVED.** Nothing in Evolution 16 was deployed, published to the three sites, added to the
auditor kit, or committed.

## Rollback
Evolution 16 is uncommitted. To remove it:
```
git checkout -- cain/execution_chain/coordinator.py cain/mission_control/governed_runtime.py
rm -r cain/mission_control/{durable_runtime,recovery_lab,recovery_lab_mutations,recovery_explore}.py \
      cain-mission-journal-verify tests/test_evolution16_durable_runtime.py artifacts/evolution16 \
      scripts/cain42_evolution/build_evolution16_evidence.py
```
Reverting the two tracked files also removes the F16-01 and F16-02 fixes. The valuation files (v47, v42) are separate;
register v48 from another session builds on v47, so removing v47 would break v48.