# CAIN-A75-060 Research Toolchain Review

**Verdict:** VERIFIED

> Recomputes a research-toolchain boundary call log: exact allowlist, shell denial, strict host limits, a call limit, and executed only for ALLOW, with published counts equal to the replay.

| # | Verdict | Calls | Allowed | Denied | Executed | Boundary | Failed checks |
|---|---|---|---|---|---|---|---|
| 0 | SOUND | 4 | 2 | 2 | 2 | research-toolchain-boundary |  |

## What this proves and what it does not
- Proves: the boundary names a human operator, tenant, agent and exact allowlist; every call is recomputed in time order against the policy (allowlist, shell denial, host and path limits, human approval for gated tools, call limit); each recorded decision and reason equals the recomputation; an ALLOW executed and a DENY did not; and the published allowed/denied/executed counts match.
- Does not prove: that the log lists every call the agent attempted, that the downstream system enforced the boundary, or that an allowed tool is safe. It is a call-log review.
- Verify offline: `python3 verify_offline.py run.json --pinned-key XPO1rzC1vJcMoc1EXBRNurddIHqFJaytE5yvBIFmytc=`
