# CAIN-A75-008 Robotics Release Gate Review

**Verdict:** VERIFIED

> Recomputes a robot-policy release decision from supplied evaluation numbers: binding, coverage, a Wilson lower bound, safety budgets, regression and sign-offs. It does not run the policy or the robot; SIMULATED evidence is never hardware validation.

| Policy | Version | Verdict | Scenarios (rate / Wilson lower) | Failed checks |
|---|---|---|---|---|
| 0 policy-pickplace | 3.2.0 | PASS | pick 0.97/0.915479; place 0.98/0.929987; handover 0.96/0.901628 |  |

## What this proves and what it does not
- Proves: the release arithmetic is consistent, each required scenario clears the gate on a Wilson lower bound, safety and regression budgets hold, required sign-offs are current, and a simulated run is not called hardware validation.
- Does not prove: that the evaluations are representative, that the policy is safe in the real world, or that the simulation matches the robot. The gate is only as good as the supplied trials.
- Verify offline: `python3 verify_offline.py run.json --pinned-key AM+iGbO7brn008RPLyU1TZ+kalBmqlrP2O/GXZjlRmI=`