ReVolt resource-restored manual course attempt, 2026-11-18

This fresh original-demo scope made visible manual progress beyond the grid: RC Bandit passed beneath the start line, started the lap timer, passed the manhole approach and turned toward the curb before the first junction. It then stopped against the curb. This does not establish meaningful checkpoint progress, a completed lap or ordinary application Quit. Lap remained 1/3 and position 8th. The exact remaining terminal cause was the shared disk guard: at 1795014958934, actual free /home was 1,498,550,272 bytes. The scope permanently closed UNKNOWN; the next capture rejected locally before command publication, and the audit finds zero later endpoint requests. No attempt to Quit or dismiss promotion followed closure. Earlier UNKNOWN scopes were never contacted.

The owner authorized this materially changed resource-restored scope after peer recovery restored about 1.7 GB. Preparation verified 463 original input/media pins, the reused 2,418,710-byte module SHA-256 a6002d5dbcfeb534a0eadd223da616e525aa1542dad2bde381fe54e00ee862f8, and the exact browser/release adapters before launch. The served graph was 6d5558bacfa78946c235d4ac9fb79dd774445db0, module compiler source f6962ba5b1cf2829a66f5cc4d80b293c30719b37, with published 93ee0d25c preflight, cce80c7c4 import-only release standby and a0af50101 exact pre-DOWN native-expiry classification. There was no product edit, launch-module recompile, media copy or broad unchanged test campaign. No new source defect was proved, so no code fix or locked build is claimed in this task.

Ordinary Options → Controller Settings → Configure Controller displayed Left=LEFT, Right=RIGHT and Accelerate=UP. An Escape intended to leave configuration was instead recorded as Reverse=ESCAPE. The operator acknowledged the observed binding and completed Fire=ENTER, Flip Car=SPACE, Reposition Car=Z and Pause=CONTROL. Escape later opened the real race pause menu; it was not a successful reverse recovery. Resume returned to the same curb position. This configuration choice and the early right turn are operator limitations, not proved emulator defects.

Personally reviewed ordinary menu input selected Single Race → PLAYER → RC Bandit → Toys in the Hood 1. Before every prepared capture, the operator waited at least 250 ms neutrally, inspected all actual bound keys as zero and imported the independent release standby. Captures during loading or black output received no input; later complete admitted captures used their exact filenames. All driving choices were made after personally inspecting a whole display. Five driving gestures and one Escape+Left pause gesture reached DOWN; their actual native-request and whole-image ages were within the unchanged five-second deadline, and durations were at most 1200 ms. The captured lap clock and changing road landmarks establish motion, not a completed checkpoint or lap. Five driving DOWN operations were d6, d7, d10, d12 and d13; the actual pause was d15.

Nineteen intents were healthy NOT_SENT. The narrow published helper classified native-request expiry both before and after arm, with successful independent release joins. These refusals did not extend the review deadline or authorize replay: each later intent used a fresh capture and new operation. Some short menu presses produced no visible change; longer separately reviewed finite presses advanced the menus. The later reposition and pause intents expired before DOWN. No autoplay, solver, program board, timestamp recycling or stale-review waiver was used. The manually observed advance to the junction approach is retained, but the requested route/checkpoint/lap and Quit objective remains unmet.

The audit verifies 203 successful HTTP source records and one module record against the pinned graph. Existing diagnostics expose a d3dim-gps Worker-owned WebGL1 device and ANGLE/Mesa Intel UHD620 renderer. Actual executing module and child Worker-role attestation remain UNKNOWN; HTTP bytes and host identities do not substitute for it. No Wine or Windows reference run was used, and the predecessor's separate exit4097 is not an exit or Windows-normal result of this scope.

The finite detached invoker retained an actual child wait; nested run/launcher/native-supervisor/browser/display/watchdog and independent-release waits are preserved. The audit checks 63 actual wait records with stdout and stderr EOF and finds all 163 recorded runtime PIDs absent, with native supervision leaked=false. The invoker-child and enclosing launcher failures returned code1 with null signal; browser/display/watchdog cleanup returned their retained statuses. Guest exit was not observed before host cleanup. An actual parent wait for the detached invoker itself is unavailable and remains UNKNOWN; PID absence is not substituted for that receipt. The board watcher was explicitly stopped and its actual SIGTERM/EOF receipt is retained.

Evidence is scratch/runs/20261118-revolt-restored-space-manual-course-progress/. Runtime/browser/display/log growth was in owned tmpfs; shared disk growth nevertheless crossed the floor, so continuous floor compliance is false. No foreign files or security holds were cleaned. Retrieval verified SHA-256 and size and fsynced retained outputs before removing the owned tmpfs directory. Seven selected named whole screenshots remain; omitted image records retain hashes and sizes only. Git, relevant binary diffs, owned input patches and a recovery recipe preserve the exact source context without source snapshots, frame series or media copies. result.json is published last. No agents/workers, boxes, Wine, Python or deployment were used. Box ID null, state NOT_CREATED, expiry null, stop receipt null. Ownership release ends this single task.