Re-Volt checkpoint/lap attempt, 7 November 2026

No new checkpoint or lap was verified. The original demo reached its default RC Bandit / Toys in the Hood 1 race, then stopped at a concrete route boundary: the player remained beside the wooden fence and start-line post. Ordinary steering produced a side-on wall contact, followed by an upright resting car; later forward, brake/reverse and steering pulses did not establish an exit. Lap 1/3 remained unchanged. No emulator correctness defect was demonstrated, and no physics, guest state, bindings or production code was changed.

Evidence: scratch/runs/20261107-revolt-checkpoint-lap/. Candidate re-volt-demo; registered app re_volt_demo.

Original controls and route

The original package's shipped readme.txt was read using small HTTP ranges, without downloading or copying the game package. Its default controls are arrows for driving, End for flip, Home for reposition, and Escape for pause. End differs from a secondary guide's Enter mapping. The readme also explicitly describes Configure Controller as a sequential assignment wizard. That wizard was never entered in this run; defaults were preserved.

The released readiness/deadline, identity, exact-ACK and independent-release controls were reused. Only run paths changed in the selected browser/input helpers. Current relevant runtime files were selected separately from the earlier readiness run; 13 non-run pins had changed, including the canonical module and renderer/input/audio files. The original 443 installed files and separate browser manifest matched their inherited hashes.

The one original browser followed the native main menu, default player name, RC Bandit, Toys in the Hood 1 and summary. Some short Enter pulses produced exact trusted events but no visible selection. A 12-second bounded Enter advanced to name entry; later confirmations and passive captures completed the route. Exact native event delivery does not establish a guest input poll.

The personally inspected race captures establish these limits:

Capture in review/ Observed player/scene HUD consequence
grid-review.png Upright grid beneath start sign; driveway exit ahead 0 mph, lap 1/3
throttle-one.png Player and follow camera displaced amid nearby cars 4 mph, lap 1/3
throttle-two.png Player faces wooden fence beside start-line post 7 mph, lap 1/3
steer-right-one.png After three-second Right, player on its side against fence/post 3 mph, lap 1/3
pause-settled.png Actual native Resume / Restart Race / Options / Quit menu; player appears upright 0 mph, lap 1/3
resume-long.png Resume confirmed; player remains in fence/post gap 0 mph, lap 1/3
fence-forward.png Six-second Up does not establish passage through gap 1 mph, lap 1/3
back-from-post.png Post no longer occludes lane; player upright beside fence 0 mph, lap 1/3
along-fence.png Eight-second Up leaves player against fence 0 mph, lap 1/3
reverse-clearance.png After eight-second Down, no established reverse clearance 1 mph, lap 1/3
reverse-steer-left.png Final short Left leaves same fence-side route boundary 1 mph, lap 1/3

Nearby car motion and contact confound the initial displacement and turn. Later captures show the player and camera against identifiable fence/post/house geometry, but the visible lap counter never changes. Elapsed race/lap timers and opponent-distance figures are not accepted as player checkpoint progress. No separate internal checkpoint counter was identified or verified.

Review error and recovery limits

One semantic review was wrong. pause-recovery.png still showed the race after a two-second Escape pulse. The next command, select-restart, was constructed in the same tool call that returned that image, before its pixels were inspected, and falsely described an open pause menu. Its ordinary 500 ms Down therefore went to the race. The original false review is retained alongside review-correction.json. No restart was selected or completed. This command must not be described as correctly reviewed admission.

Subsequent dependent inputs followed separate image inspection. A longer Escape and passive settle visibly opened the actual pause menu; ordinary Enter eventually selected Resume. There was no controller configuration, forced reposition, private handler, memory/configuration edit or input replay. The first throttle review also understated the displayed speed as 3 mph; the inspected capture actually shows 4 mph, as recorded above.

The selected single-key driver did not exercise simultaneous throttle and steering, Home or End. Its input.cjs key mapper does not admit Home/End, although the separate released native helper has Home support. Those keys were not guessed, substituted or sent through another input owner. A useful next attempt needs qualified shipped recovery-key coverage and ordinary combined throttle/steering through the existing driver, then a reviewed route away from this fence. The present observations do not justify a physics fix.

The local post-run recovery-key-capability.json confirms both Home and End are rejected by the selected mapper, without sending guest input. This is an unmet driver-coverage prerequisite, not a demonstrated emulator blocker or a completed native recovery. The coordinator's 23:58:34Z request to exercise Home/End arrived after the original browser had already closed at 23:57:50Z; it was read after the normal box stop. Its instruction not to replay or resume an ended runtime was honored. The supplied original PC manual capture was then inspected and independently agreed with the already-read shipped readme.

Runtime identity and evidence recovery

The selected source checkout is 941f63c3f1d4dbe54859033421c2a208857b62e3. Relevant tracked changes were empty; the retained untracked boot-helper patch and final run-helper patches complete the recorded inputs. This task reused the existing canonical build; it did not perform or claim a fresh code build.

Both loaded guest and renderer module bytecode hash to f9650607181301b5d2a9990420f437bd9f1ca10d40ac73067339f07499d53121. CDP evidence binds the actual guest Worker/process instance and owning legacy render Worker/endpoint, with WebGL 2 and ANGLE Intel Mesa UHD Graphics 620. These are live owner/backend observations, not toolbar inference. Chrome was 151.0.7922.108 with an unemulated 1024×712 viewport and default guest Workers.

runtime-pins.json, media-identity.json, source SHA and binary diffs reference recovery inputs without a source or media archive. The loaded module, unique captures, served hashes, commands, native events and terminal logs are retained. All 1,069 staged pins matched again at shutdown, including original media.

The run produced 28 exact command ACKs and 45 trusted native events: 21 balanced key pairs, one balanced mouse pair and one mouse move. All 22 pulses had independent release armed before DOWN and exact DOWN/UP ACKs. There was no original UNKNOWN command or replay. This transport/input result does not erase the one semantic review error described above.

The initial immediate post-quit clearance found Chrome absent, ports closed, profile/writer removed and input released, but Node still draining its close timers. Its failed assertions are retained. The later read-only custody check confirmed Node PID 5906 and Chrome PID 5918 absent, HTTP/CDP ports 33907/40255 closed, and no profile, writer or held input. No original command or input was replayed for that check. All 316 evidence references (260 unique hashes, 11,868,737 reference bytes) were retrieved and SHA-256/size verified before normal stop. Recovery and stop records are retrieval-verified.json and normal-stop-confirmed.json.

One fresh UUID created bx_426fzwgf at 23:42:34.091Z, with initial expiry 2026-11-08 00:17:33.939Z. The immutable browser/watchdog deadline was 2026-11-08 00:07:34.101Z. Independent normal-stop and deadline delegates were armed under a detached supervisor before allocation, with durable heartbeats. Chrome exited normally at 23:57:50.233Z. The delegated normal stop completed at 23:59:31.801Z, receipt stop_751ce88f3a43b81b, with forced:false. Final API state is archived, active expiry null; no resume, delete or additional box operation occurred. Both lifecycle delegates exited zero and their supervisor recorded no remaining children. Local free space stayed above the 5 GB launch floor and 1.5 GB operating floor. The messageboard watcher is released at finalization.

There is no checkpoint/lap, completed recovery, win, broad performance, audio, parity or full-corpus claim.