MCM repaired original — race reached, input freshness prerequisite
One original Motocross Madness demo reached textured T-rific quarry practice with the published queued Flip repair. Driving and steering consequence are not established: all five proposed ArrowUp commands were refused before dispatch by the existing 12-second scene-freshness gate. No arrow key went down; no held input was retried. The first runtime prerequisite is personally reviewed immutable race pixels followed by unchanged live owner/module revalidation completing inside that existing admission bound. This assignment ends at that prerequisite; it does not relax the bound or claim an emulator input defect.
Evidence: scratch/runs/20261107-mcm-repaired-original-driving/, especially
observations.json, runtime-pins.json, original/*-binding.json,
original/evaluation-times.jsonl, exact numbered commands/replies,
probe-read.json, and terminal-retrieval-verification.json.
Original identity and visible result
The runtime used Git commit dcbc8248a606d9594634213242063960316bdb4d, including
the tested production asynchronous queued Flip repair. The relevant tracked
source patch is empty; the existing untracked lib/wininit-boot.js is retained
as a new-file binary patch. Small run-specific drivers/probe have new-file
patches. All 892 staged file SHA-256/length pins matched before launch and at
termination. Original assets, registry configuration, normal working directory
and all 311 registered inputs were preserved. The 310 companion media records
match the preceding original run exactly. Git is the source archive; evidence
contains no source/media archives or repeated media copies.
The actual guest and owning renderer each recovered one live executing module,
SHA-256 e41511fe861363fbe1639d80f20cbcdd7240f46937c93a801eaa256ca094c467.
Guest target: E48C4B50B70CA194B153B2DEE34DB2CF; renderer target:
4C90BE0BA8D9C399A0ABC3620E701781; endpoint 1. The independent renderer epoch
matches the probe receipt. JS repair sources are separately pinned in
runtime-pins.json; an unchanged WAT module hash alone does not identify them.
Personally reviewed video-memory OK, RIDER name, Single Player Event, Stunt
Quarry, Practice/two opponents, T-rific, default rider/yellow bike, and Start.
review/quarry-ready.png and original/quarry-no-input.png show textured terrain,
sky, upright centered rider/yellow bike, and a near-zero speed gauge. The latter
also shows both opponents on the terrain. These are an idle comparison, not
a before/after control consequence. No throttle or steering receipt exists.
No FPS, audio, parity or general playability assertion follows from these pixels.
Environment: baremetal boat provider, Chrome 151.0.7922.108 / V8 15.1.206.16, 1024×768 page, owning legacy D3DIM Worker, WebGL 2.0. Actual GL renderer is ANGLE Vulkan SwiftShader, not the earlier Intel UHD 620 baseline. No hardware-matched performance improvement can be inferred.
Exact admission boundary
All 27 numbered commands (0–26) have successful exact-ID ACKs. Seven ACKs are explicit zero-effect stale refusals: two menu proposals and five ArrowUp proposals. Each refusal was followed only by a new capture, as required. The menu commands that did dispatch retained independent scheduled key release, full owner/window/module/epoch checks and immutable personally reviewed pixels. Sticky UNKNOWN handling and transport were unchanged; there was no UNKNOWN or partial-input replay.
The post-screenshot live identity read took 3,586, 4,080, 3,995 and 5,426 ms for
the four retained refreshed race scenes. For throttle-fresh, the host request
began when the pixels were 9,392 ms old and returned the refusal at 13,256 ms.
For throttle3, those bounds were 9,806–13,011 ms; for throttle4,
10,866–14,905 ms. These are recorded host request/reply bounds, not invented
exact gate-check timestamps. Earlier proposals were already stale at request.
The final live validation can exhaust freshness even after a prompt review.
Resolving that acquisition/admission latency while preserving every identity
and freshness check is the prerequisite before another authorized driving run.
Bounded natural wait observations and limits
The run-only span wrapper observes Promise settlement and returns the original Promise unchanged. GL wrappers call each original method once, preserving arguments, receiver and return value. It adds no GL synchronization/readback operation, guest call, clock/configuration change or dispatch bypass. A bounded 100-ms timer observes renderer task progress. The observer was disposed.
The probe armed for 3,000 ms with capacity 4,096 and retained 762 completed
records, zero drops: 733 natural clientWaitSync calls, each 0–0.050 ms,
spanning 2,996.870 ms, and 29 renderer timer tasks spanning 2,799.540 ms.
This establishes Worker event-loop progress during the poll observation window.
It does not measure a logical frame or simulation step.
No complete Flip, replay, readPixels, getBufferSubData or writeback span entered
this short window. Installation occurred inside an already executing operation;
record sequence/ordinal/context are null. Therefore total Flip/readback wait is
unknown, and the probe's pendingFlip:false means only that no new wrapped
Flip entered its own record set. It must not be treated as an independent
production pending-state assertion. This is a measurement limitation, not a
zero wait or stalled-forever claim. Independent capture receipts show completed
fence sequence advancing 74→110→153→196→232, with async reads 1→7→14→21→27;
closure reports sequence 303 / 38 async reads. These counters establish
completion progress, not FPS or the duration of any particular Flip.
Closure and checks
Fresh UUID-idempotent boat bx_hmpdck8g was created once. The immutable runtime
cap was 14:21:52.105–14:46:52.105Z; only one original was launched, at
14:22:14.243Z, and ordinary quit completed before the cap. Before quit all
physical/async keys were released and the input queue was empty. Chrome exited
0; all 16 recorded processes disappeared, both ports refused connections, and
profile/writer lock were absent. All 241 terminal files, 11,299,084 bytes, were
retrieved and SHA-256/length verified before delegated normal stop. Stop completed
14:28:48.172Z, forced:false, state archived, active expiry null; original
archive deadline was 14:51:52.323Z. No old boat was resumed, stopped or deleted.
Sampled local free space stayed above 2 GB (initial 3,664,236,544 bytes; pre-retrieval 2,831,392,768 bytes); remote preflight free space was 58,041,843,712 bytes. Allocation has an explicit 2-GB guard. Continuous disk telemetry is not claimed. No build or production edit was needed. Driver/probe syntax, exact ACK/zero-effect assertions, module/epoch agreement, bounded record checks, media reference identity, terminal retrieval/cleanup assertions, and owned documentation whitespace pass. Accepted fixture matrices were not rerun.
The run-only probe check also passes original Promise identity, success/rejection settlement, one-for-one GL calls and restoration. The global evidence validator reports the existing Alice dialogue containment error; this run has no errors, missing artifacts or invalid metadata.