Motocross Madness fresh scene attempt — 2026-11-07

The registered original reached a personally inspected, settled T-rific quarry race with rider, yellow bike, textured terrain, RIDER HUD and speedometer. Throttle and distinct steering remain unproved. The first new causal blocker was Boat exec HTTP 500 on the ordinary throttle request, with no successful ACK. No emulator mismatch is demonstrated; no production code was changed.

Evidence: scratch/runs/20261107-mcm-fresh-scene/.

The corrected checked scene-step.py was reused. Local controls executed the actual request caller with refusal and successful ACKs, proving refusal exits 3 and successful ACK permits subsequent retrieval. The actual scene-step refusal control proves no later retrieval or original launch. Seven controls executed the production direct initializer; actual renderer/input executor controls covered uppercase RIDER, balanced Enter, literal menu points, throttle and separate steering holds. Atomic ACK, UNKNOWN/no-replay, invalid-duration and prerequisite fail-fast controls passed before allocation.

One fresh box, bx_w6rehtfv, was allocated at 07:09:45 UTC with a persisted, fsynced UUID Idempotency-Key. Remote Surface4/7 COM/vtable-tail checks passed against a retained, prepared and consumed module. Native Chrome preflights proved trusted RIDER character triples, balanced Enter and arrow key pairs, and invalid-name/duration zero effects on about:blank.

The generated dx_boids guest ran in a Worker with an actual legacy/WebGL owner; completed fences advanced 50 → 58, error 0, adapter present, failure null, fallback 0 and completed software draws 0. I inspected its sample boundary scene. I also inspected the real generated COMCTL property sheet and its disappearance after balanced ordinary OK at (173,418). Both generated browsers closed normally before the original.

Exactly one original browser launched at 07:12:05.242 UTC, through the production ?debug&threads&d3d-renderer=webgl&app=mcm route. I inspected the startup video-memory prompt. Its capture expired during retrieval, so a read-only capture refreshed the scene. The refreshed pixels were teal while the modal log remained, as documented in the supported direct-launch notes; ordinary Enter accepted default OK. Each subsequent action followed a fresh, personally inspected actual scene and received a successful numbered ACK: Enter Name → uppercase RIDER → Enter → Single Player → Stunt Quarry → Event Options Next → T-rific quarry Next → rider/bike Start. No forced state, configuration, registry, media or API outcome was used. All eight admitted inputs were 8.622–10.434 seconds old at dispatch, within the unchanged 12-second guard (freshness-verification.json).

After a read-only 35-second settling capture, 07:14:57.543 UTC review/race-settled.png shows the rendered quarry race. I personally viewed it before requesting an ordinary two-second Up hold. The Boat request returned Internal Server Error, HTTP 500, without command ACK. The checked scene step stopped and did not invoke post-action retrieval. A premature local view_image call while that subprocess was pending found no after-throttle file; it obtained no image and issued no remote retrieval. This local ordering mistake does not establish an input outcome.

Read-only command-record inspection at 07:16:06.415 UTC found no id-throttle, command 10, consumption marker, input-admission record, ACK, browser failure or resulting capture. Latest remained race-settled. The throttle request was never replayed, and no steering request followed. The transport outcome remains UNKNOWN; absence of remote command records is recorded separately from an ACK. Ordinary terminal quit then successfully used command 10, confirming no throttle command occupied that slot.

The original owning renderer was target 4B2B060CE19A95DD8291A3ABDDAD41D9, session C31DEDA2B66802163F2512FF9B2F7A16, endpoint 1, legacy/WebGL. At the settled race its completed counter was 110, error 0, adapter present, endpoint failure null, guest fallback 0 and completed software draws 0. Actual captured stats report ANGLE / Mesa Intel UHD Graphics 620 (KBL GT2), WebGL 2.0, rather than inferring the route from the UI label. The guest and renderer Workers both executed module SHA-256 e41511fe861363fbe1639d80f20cbcdd7240f46937c93a801eaa256ca094c467, 2,182,493 bytes, equal to the staged standard module. Chrome was 151.0.7922.108, V8 15.1.206.16. Staging records main 2661845f93cb4578362a62bf8f87e9544b1aca7b, binary dirty patches and the untracked wininit-boot patch; these identify source inputs without asserting fresh-build provenance for the retained standard module.

All 311 authentic registered inputs, 45,460,823 bytes, matched the previous authentic inventory and before/after hashes. MCM.EXE remains f0370d1d526be46eaf6ea5f941f185a393a8432baec695e1529f8672a314a71b. Installation-root VFS paths and working directory were preserved. The RIDER profile was created through ordinary guest UI only. All 938 staged pins matched remotely before original and at cleanup. All 1,657 complete successful HTTP responses match pins with no unmapped success. Non-200s identify build-info, auth-user, optional icons, favicon and optional font-bundle endpoints.

All 188 remote artifacts were recovered or deduplicated against identical owned bytes and hash-verified before delegated shutdown. Chrome exits were 0. The later terminal census verified all six browser/controller PIDs absent, all six ports refusing and all three private profiles and writer locks absent. Delegated normal stop stop_837e0f9a0b576551 completed at 07:16:49.915 UTC, forced false, error null. The box is archived, active expiry null. Allocation through completed stop took about 425 seconds; the whole assignment stayed within 35 minutes. Local free space stayed above 2 GB. No original input was replayed, no second box/original was launched, and no stopped box was resumed. No FPS, audio or full27 claim follows.

Next work needs a fresh authorized run after the Boat exec failure, preserving the checked ACK-before-retrieval rule. Race entry is now evidenced; ordinary throttle movement and separate steering still need direct visual verification.