Expectrail review: the same checks before and after a change

REJECTED

The change fails a required check, or the verifier rejected it.

the verifier rejected the candidate: regression check player-stays-above-ground failed on smoke; regression check player-always-visible failed on smoke; held-out check player-stays-above-ground failed on held-out-routes

Evidence: RECORDED LIVE, an archived live run (the run labelled it FRESH LIVE) Candidate from a fixture patch Verifier decision: reject

Recorded evidence: SAM probe run 41123758768, verify mode. This page shows that run as it wrote its record; nothing was run again.

Each check, before and after the change

CheckRoleBeforeAfterResult
player-stays-above-groundrequestedfailfailUNRESOLVED
player-motion-on-smokeguardfailpassREPAIRED
coins-visibleguardpasspassPRESERVED
player-always-visibleguardfailfailUNRESOLVED

Before is the game with the reported bug and no fix; after is the same game with the candidate. A requested check is one the repair request names as failing; only those can show a repair. Guards must pass after the change. The verifier also checked the clean game, separately.

The first failing frame, before and after

Before the change, frame 5
Before the changerun review-20261030T203750Z-22490, failing base, frame 5, replay 919a9e67ce0f
After the change, frame 5
After the changerun review-20261030T203750Z-22490, candidate, frame 5, replay 4223f770e21f

Private routes

1 check on the private routes: 0 passed, 1 failed, 0 need review (source: harness). Route names and frames stay private.

How the review ran

Diagnostics

Identities

Review run
review-20261030T203750Z-22490 (live mode)
Repair request
rr-gate1, SHA256 ac8f0e926683dc03a0f7dc314a2dd3302b2239f5fdd8660aaff9ce69bbcb62be
Candidate patch
30a53ff1a9b53211f978818faff8bc74251054acae23a6f1d776f7dd35e7cb30
Verification file
060ea6c62f1a53ea652b497201967980b7afb1646387db537e9b983a7fdb6e7a
Failing base
harness, observed by SAM 3.1, replay 919a9e67ce0f
Candidate
harness, observed by SAM 3.1, replay 4223f770e21f
Run it again
python3 -m agents.review_change --request <request.json> --candidate <candidate.patch> --mode live --out <new folder>

This page shows evidence for these checks on this route. It is not proof that the game is correct, and no status here approves a merge or a release. Generated from change-evidence.json.