ReVolt route attempt and pre-DOWN capture expiry, 2026-11-18

This task did not establish route/checkpoint or lap progress, or ordinary application Quit. Both original demo scopes reached Single Race → PLAYER → RC Bandit → Toys in the Hood 1 through personally reviewed ordinary menu gestures. Neither scope sent a gameplay DOWN. The first scope refused three driving intents before arm and closed UNKNOWN on a fourth; the fixed scope refused one intent before arm, then closed on the shared disk floor. Lap remained 1/3 and the inspected car remained at 0 mph. The predecessor's 8 mph/heading and separate promotion-aware exit4097 remain separate evidence and are not results of this task.

The first scope exposed a narrow helper defect. d4 passed the whole-screenshot age check at 4365 ms, armed the independent release child, then failed the stricter native-request freshness check after current-owner queries. No DOWN was sent. The original request timestamp was 188 ms earlier than the whole-image timestamp. Independent release reached physical UP0, and the child actually waited to exit0 with both EOFs. The assertion nevertheless sealed UNKNOWN. That scope remains permanently closed; the correction does not reinterpret or reopen it.

Published commit a0af50101 classifies only that exact native-request freshness assertion, before any DOWN, as LOCAL_NOT_SENT. It also checks request freshness before recording an arm. The original validator still runs, all other structural/owner/transport errors propagate, and sent/partial operations still close. The existing finally block joins independent release; a fresh intent/capture is allowed only after a successful join and while the gate remains healthy. The five-second gameplay deadline and 1200 ms maximum gesture are unchanged. Seven focused tests passed, including exact boundary expiry, valid pixels, malformed expired metadata and already-sent state; the exact original browser/release adapters compiled, and flock /tmp/wine-build.lock bash tools/build.sh passed. The author committed only the helper and its test, then immediately strictly pulled/pushed. No broad unchanged campaign or launch-module recompile was used.

The materially changed manual-tmpfs-fixed-route original exercised the published helper through ordinary navigation and reached the course grid. Every new preparation waits at least 250 ms neutrally, inspects all actual bound keys as zero, and imports the independent release standby before capture. The standby has no input configuration or arm until the personally chosen command arrives. On one initial capture, candidate1 was viewed while the operator capture was still completing; it was black and received no input. The completed capture admitted candidate3. The operator then used exact admitted filenames, only after complete command receipts; no gesture referenced a discarded candidate. All driving intent attempts used the unchanged five-second check; no autoplay, pre-sent input, solver, stale review waiver or deadline increase occurred.

The fixed original closed at disk boundary 1795013497826 with actual free /home 1,498,914,816 bytes. The next operator invocation rejected locally on UNKNOWN before command publication, with zero endpoint requests. The last whole screenshot was captured before that boundary. No later game endpoint, replay, ordinary Quit or promotion dismissal was attempted. The fixed original's targeted post-arm native expiry branch was not reproduced before the disk closure; its unit boundary tests and successful fresh-menu validation are the available correction evidence. Continuous task disk-floor compliance is false. All growing runtime/browser/display/invoker files were in owned tmpfs; shared filesystem growth still crossed the floor. No foreign cleanup or held inputs were touched. Peer-owned recovery, when recorded in retrieval evidence, enabled durable finalization only.

The runtime source graph is the predecessor's tested 6d5558bacfa78946c235d4ac9fb79dd774445db0, with published 93ee0d25c preflight and cce80c7c4 import-only release standby, plus this task's a0af50101 operator-helper correction for the second scope. Both original launches reused the exact existing 2,418,710-byte module a6002d5dbcfeb534a0eadd223da616e525aa1542dad2bde381fe54e00ee862f8, compilation source f6962ba5b1cf2829a66f5cc4d80b293c30719b37, and the 463 pinned original inputs/media. HTTP source/module hashes are checked against the graph. Existing host diagnostics expose d3dim-gps worker ownership and webgl1/ANGLE Intel UHD620 device data. Actual executing module and child Worker role attestation remain UNKNOWN. No invented attestation or Windows parity follows; no Wine or Windows ground truth was used.

Evidence is scratch/runs/20261118-revolt-manual-route-checkpoint-driving/. audit.json retains both permanent closure boundaries, zero later endpoint requests, zero gameplay DOWN, actual individual release/native/browser/display/supervisor/invoker waits with both EOFs, supervision receipts and final PID checks. A finite detached invoker avoided the predecessor's enclosing-tool lifetime signal problem: both actual invoker-child waits returned1 for the recorded failures, with null signal and both EOFs. Those failures are not ordinary guest exit. The guest was not observed exiting before host cleanup in either scope. The preceding task's code4097 is not described as Windows-normal or zero.

Git holds source, with exact working binary diffs, published owned patch and small new-input patches. Selected named whole images and unique logs are retained; omitted image records preserve hash/size only. Retrieval independently round-trips SHA/bytes and fsyncs before owned tmpfs removal. The recovery recipe references existing unique module/media without duplicating them. result.json is written last. No agents/workers, Wine, Python, deployment or boxes were used. Box ID null, state NOT_CREATED, expiry null, stop receipt null. Final ownership release ends this single task.