TuxRacer retained handoff publication and one bounded progression attempt
Historical source last updated, 2026-10-30. Runtime and implementation claims below describe that dated source; see current app notes and source for subsequent changes.
The retained independently tested handoff and original MainMenu evidence are
published. Packaging commit 1f6289cac96efc851d5e9d233919b8a76cbfdeb2 removes
only the real trailing blank line in the previous supervisor. It was remotely
SHA-verified before runtime. All pinned retained artifacts were independently
checked again by SHA/bytes; broad acceptance and full compilation were not rerun. The original missing skia-canvas error and previous SIGKILL remain in the parent.
The new display helper uses already-installed pngjs 7.0.0. Real PNG read and invalid-image rejection passed. Node startup with 256MiB heap passed first; the unchanged 512MiB/preload canonical manifest check enumerated all 157 files, including the comma-name check. DATA/AS were unlimited in Node, CPU60 and FSIZE12MiB remained enforced. The pre-guest ordinary-action transport smoke tested a real stdin pipe to a fixture, but did not validate the production schema.
Precisely one original guest ran. The worker inspected its actual title, then
mistakenly submitted {action:"keydown",code:13}. The runner rejected this;
no key was queued. Within the same execution the worker sent the supported
cmd:keydown:13 and cmd:keyup:13 forms. Both were dequeued and dispatched
at batches15888/15893. The real frame at batch15974 was personally inspected:
Enter an event, Practice, Credits and Quit. The later Practice click attempt
failed opening the vanished stdin fd before any bytes were written. Thus two
native keys were sent, zero Practice clicks; no Practice, race or gameplay claim.
The runtime ended SIGKILL after58.272 seconds wall, user CPU60.327 seconds, peak group RSS341213184 bytes. CPU60 was enforced; supervisor reason was null, so the exact kill source was not independently observed. No second execution, forced unlock/state patch, audio suppression or production source edit occurred. No final guest state/exit trace exists. The retained title snapshot is labeled as title-only; startup exit traces are not runtime terminal evidence.
After the runtime, the exact pinned production runner parser and control handler were extracted and tested without a guest: three valid native commands queue, three malformed controls reject. The corrected controller validates native cmd forms and performs this schema preflight before any future guest. Its local validator has six positives and five rejecting controls. Final controller changes are not runtime-retested. Executed and corrected supervisor files/hashes are separate. This corrects the actual harness defect without changing production.