# CAIN-A75-009 Autonomous R&D Change Review

**Verdict:** VERIFIED

> Recomputes an autonomous R&D change claim: diff and authority binding, preregistration, CI-clearing gain, a test that failed then passes, regressions, held-out evidence and claim honesty. It does not re-run the experiment or the code.

| Change | Verdict | Gain / 95% CI | Fixed tests | Regressions | Failed checks |
|---|---|---|---|---|---|
| 0 change-parser-roundtrip | ACCEPT | 0.09 / 0.02 | 1 | 0 |  |

## What this proves and what it does not
- Proves: the change is bound to its files and experiment, the gain clears the stated 95% interval rather than being a point estimate, a test failed before and passes after, nothing regressed, the held-out check is independent, and the record does not overstate itself.
- Does not prove: that the experiment was run as described, that the metric is meaningful, or that the code is correct beyond the named tests.
- Verify offline: `python3 verify_offline.py run.json --pinned-key UuwlUy36flMvwAXAkmsNwprs1LullQjv/4CVuce90ag=`