ReVolt tmpfs manual-race attempt — 2026-11-18

The sole fresh unchanged original reached the genuine main menu. Manual race entry, driving/checkpoint consequences and ordinary Quit remain unverified: the independent resource watchdog permanently closed the scope before any guest DOWN. The next Start Race operation was refused locally without creating a command or sending an endpoint request. No scope was reopened or replayed.

Evidence is in scratch/runs/20261118-revolt-tmpfs-menu-to-manual-race/; result.json is published last. The requested revolt-qualified-operator-manual-race-20261118.md did not exist in this checkout or its tracked history. Its sealed predecessor result, RECOVERY, qualified-control hashes and unique operator inputs were read instead. The predecessor's accepted main-menu → Select Race/Single Race transition remains historical evidence only, and its disk-closed runtime received no requests.

Frozen original and reused qualification

The served graph is exact published f6962ba5b1cf2829a66f5cc4d80b293c30719b37, with unchanged compiled/served wasm a6002d5dbcfeb534a0eadd223da616e525aa1542dad2bde381fe54e00ee862f8 (2,418,710 bytes). All 463 media pins match before launch and after cleanup, including the original executable and Mss32.dll. Source manifest and relevant binary diff record excluded worktree changes. Two later published differences (lib/cli-child-observation.js, src/main.watx) are also excluded; no equality with current published HEAD is claimed. The prior exact Git recovery compilation is reused, rather than repeated alongside gameplay. Executing child module and roles remain UNKNOWN because optional child Runtime/Debugger requests were omitted.

The exact qualified ordering/release/readiness operator was reused, including unchanged release.cjs, guard.cjs, operate.cjs and invoke.cjs. Its browser replacement/UNKNOWN/independent-UP controls were not repeated as a setup campaign. The matching frozen native admission tests and local ordering controls passed. One preliminary native test invocation mixed a coworker's in-progress current test with the frozen helper and failed on their new readback case; the mismatch was recorded and corrected by compiling exact frozen test bytes in memory. No original runtime was affected.

Only run-local disposable-state handling changed: published 2a1afa6c4 supplies tmpfs profile/cache/temp/crashes and raw-XWD-in-memory whole captures; 5a451516c supplies the component-update suppression switch. Puppeteer's ordinary background suppression remains in its normal launch arguments. Canonical controller/helper files were not edited. No production code fix or new production build was needed, and no new authored browser campaign was launched. Independent-release child helper loading remains the predecessor's unmeasured scheduling limitation; no release child was actually needed in this original because no DOWN was admitted.

Declaration and actual observation

Before launch: one original, startup 900 s, total 1800 s, Quit reserve 180 s, terminal observation 60 s, cleanup 20 s; transport/capture 3 s, independent fallback 5 s, gesture at most 1200 ms, menu freshness 30 s and driving freshness 5 s. Nothing was extended. A separate prelaunch amendment declared the early resource reserve as 1,520,000,000 bytes, replacing the predecessor's 1,550,000,000-byte reserve because disposable browser writes and raw captures now use tmpfs. The required floor remained 1,500,000,000 bytes, with 100 ms independent sampling and per-request floor checks. There was no RAM gate.

Personally reviewed whole frame original/natural-menu-1.png shows the ReVolt logo and main menu, Start Race selected orange, then Best Times, Options and Quit. It is neither Demo attract nor manual gameplay. At 11:58:59.207 UTC, the independent watchdog sealed the shared scope UNKNOWN; sampled free space was 1,518,329,856 bytes. The gate closed at 11:58:59.215. All 38 ordinary requests precede shared closure. There were zero guest DOWNs, zero UPs and zero later endpoint requests. The attempted Start Race command encountered the existing UNKNOWN file and was refused before writing a command. No race, speed, steering, checkpoint, guest consumption, ordinary Quit or native guest exit is inferred.

No config, save, clock, guest state, media, solver, autoplay, forced-focus or guest-polling changes were made. UT installation remains open; its ISO/partials and other held work were not touched. Shared disk-writer attribution is UNKNOWN; no foreign cleanup was performed. The lifetime sampler records actual disk observations through final publication; its minimum and any floor breach are reported in the last result rather than inferred from the runtime watchdog alone.

Cleanup and recovery

Seven original child/enclosing waits have actual close and both pipe EOF flags: browser, window manager, Xvfb, native supervisor, watchdog, enclosing launcher and invoker. Native supervision records driver exit 1, no leaks, and all 35 recorded owned runtime PIDs are absent. Separate direct-driver EOF is UNKNOWN because its pipes were inherited; the surrounding supervisor EOF is actual. The outer exec tool returned exit 1; its separate EOF flags are unavailable. Browser exit 0 is host cleanup, not game Quit/native guest exit. The tmpfs profile was removed only after browser wait/drain verification; no downloads needed retention. Private display, socket, lock and auth are gone. The independent lifetime board watcher has a separate actual wait/EOF receipt joined before sealing.

Git is the source archive. RECOVERY.md, source/media manifests, relevant binary diff, owned new-file patches and retained tested-module reference describe recovery. No source-tree snapshots or media copies were made. Selected whole/current-display screenshots remain; unselected black startup images have discard hashes, and closed verbose logs have verified lossless gzip receipts. All evidence precedes result.json.

No agents, workers, boxes, Python, Wine, public deployment or foreign edits were used. Box ID null; state NOT_CREATED; expiry null; stop receipt null. Ownership is released after publication and sealing; this worker stops.