ReVolt fresh review/release ordering — 2026-11-18
The run-local operator repair passed its authored controls. The sole registered original reached the real ReVolt menu and delivered the reviewed 1200 ms Start Race Enter, followed by independent UP and passive physical state 0. Manual racing, visible speed/steering, checkpoint progress and ordinary original Quit remain unverified. The independently declared disk watchdog closed the scope at its 1.6 GB early reserve. No gameplay/runtime defect is established.
Evidence: scratch/runs/20261118-revolt-fresh-review-release-order-driving/; result.json is sealed last. The previous readiness run remains permanently closed, with no reused endpoint or failed operation.
Operator and controls
No shared runtime, common capture helper or disk-guard implementation changed. Published helper revision 30e35805fb7d263c91fb12ee80bd5ae628c6a24b is loaded from Git in memory, including its unchanged cf132e191 disk-guard body. This excludes concurrent shared-tree changes. The later graphics/helper owner was acknowledged on the board; their changes are not adopted.
The controller assembles gesture code before execution, completes served-source/media/module pin checks before launch, obtains a new whole capture, requires its explicit review, and only then starts independent release. The final checks use ordinary host lifecycle links, exact page target/URL/document object tokens, browser PID/start token, process/renderer/window/instance/output tokens, visibility, genuine current-display validation, neutral physical state and freshness. There are no child Worker Runtime/Debugger requests. Executing child roles and executing module are UNKNOWN; served bytes and host module-object identity do not prove executing wasm bytes.
The native local ordering test rejects the predecessor's 32,268 ms sequence and admits an 80 ms current-check sequence with unchanged bounds. Real normal-browser authored controls establish:
- A reviewed pair of 1200 ms arrows reaches the trusted browser input route, independently releases both keys to physical 0, and the authored native fixture shows green while held, versus blue before. Escape then exits the actual fixture with native process exit code 0.
- A stale local operation sends no DOWN. A new reviewed operation in that still-healthy scope performs ordinary Quit after the earlier release has joined. Failed operation names cannot be reused.
- An intentionally delayed 5100 ms post-arm control reaches its original 5 s fallback UP, physical 0 and actual child join. DOWN is refused locally. A new capture is allowed in the same healthy scope; the failed arm is never extended.
- Actual navigation replaces the document after arm. The ordinary owner reader fails on
about:blank(crypto.randomUUIDunavailable); independently, release rejects the missing original document epoch before any input dispatch. There was no DOWN and no late UP to the replacement. This is a document-replacement negative, not a proved native process-replacement run. - A deliberate post-DOWN query exception closes UNKNOWN. The independent release still acknowledges matching UP and physical 0. A later operator command is refused locally. Actual ordinary-wire count after UNKNOWN is zero.
One preliminary launcher failed before creating a browser because the Git-backed Node loader lacked a .cjs fallback. It was fixed before the first browser scope. Three actual authored browser scopes followed; none was reopened after UNKNOWN. authored-controls.json and offline-audit.json retain exact distinctions and receipts.
node test/test-revolt-display-admission.cjs and flock /tmp/wine-build.lock bash tools/build.sh passed. The locked build uses the shared dirty tree; its output is not the original's served module. Existing published native/readiness controls are reused by exact revision, with run-local native ordering and normal-browser controls added. No runtime repair or changed build performance is claimed.
Sole unchanged original
original-freeze.json verifies equality between the served runtime graph at 30e35805f and then-current published HEAD 81150aeb2 for src/, lib/, host.js and index.html. All 463 media pins, including real Mss32.dll, match before launch and after cleanup. Compiled/served wasm SHA-256 is a6002d5dbcfeb534a0eadd223da616e525aa1542dad2bde381fe54e00ee862f8 (2,418,710 bytes). Offline Git compilation reproduces it exactly. Actual executing wasm remains unknown.
Limits were declared before launch: 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; driving freshness 5 s. No bounds or deadlines changed. No guest configuration/save/clock/focus/state/asset change, forced guest polling, autoplay or solver was used.
Personally inspected whole frames original/startup-natural-2.png and original/after-start-race.png show the actual ReVolt menu, Start Race selected orange, with Best Times, Options and Quit below. No attract frame is labeled manual gameplay.
The fresh reviewed capture was taken at 10:44:14.756 UTC. Release armed at 10:44:28.170 with unchanged fallback 10:44:33.170. Final host owner checks completed and DOWN was issued 231 ms after arm (wire issue 10:44:28.404). The frame was 13,645 ms old at DOWN, within the menu's 30 s bound. Independent Enter UP acknowledged at 10:44:29.776 and physical Enter was 0. There is transport/trusted-input evidence; guest consumption is not acknowledged. The later whole frame still shows the menu, so race entry is not inferred.
At 10:44:36.284 UTC, admission sealed UNKNOWN after the independent disk watchdog observed 1,599,639,552 bytes free: below its predeclared 1,600,000,000-byte early reserve, above the required 1,500,000,000-byte floor. The controller made 51 ordinary requests, one DOWN, and zero ordinary requests after UNKNOWN. A subsequent inspect command was refused locally without writing a command or contacting the endpoint. The original was not retried or reopened. Cleanup completed at 10:44:37.506.
An offline recovery compilation was launched concurrently during the original and completed after the input. It was not awaited by action admission, and it made no guest/endpoint requests. Its CPU/scheduling contribution is unmeasured. This is a limitation against the requested ordering of all heavy diagnostics before fresh review; the per-capture immutable-diagnostics-finished marker only marks operator preparation, not completion of that asynchronous recovery compilation. The 231 ms post-arm check and independent release are directly evidenced regardless. No causal guest-consumption claim follows.
Cleanup and recovery
The audit retains 38 actual child/enclosing/release/browser/display/watchdog waits with stdout/stderr EOF, and verifies all 171 recorded owned PIDs absent, private displays/auth removed and browser profiles gone. Native supervision reports no leaks and records actual direct-driver exits. The direct driver inherits supervisor pipes: separate per-driver EOF flags were not recorded; enclosing supervisor/launcher/invoker pipe EOF is recorded. The initial long board-watcher tool ended with exit 143 and no child-close receipt; its separate child identity/EOF is unavailable. A distinct final finite board watcher has its own actual close/EOF receipt. Neither that final receipt nor the 171-PID audit reconstructs the initial watcher gap. No missing direct-driver fields are reconstructed.
Git remains the source archive. RECOVERY.md, the source manifest, relevant binary diff, owned new-file patches and exact retained predecessor input/module hashes describe recovery. Only selected named whole screenshots and unique authored inputs/modules are retained; closed logs may be losslessly compressed with hash/roundtrip receipts. Result metadata is written last.
No agents, Codex workers, boxes, Python, Wine or public deployment were used. UT/LTris/Caesar and held BloodII were preserved. Box ID, expiry and stop receipt are null; state is NOT_CREATED. Ownership is released and this worker stops after sealing.