ReVolt startup DOM and manual race — 2026-11-18
The sole original reached a manually selected Single Race, with RC Bandit on Toys in the Hood 1. The reviewed whole race frame shows lap 1/3, position 8th and 2 mph. No acceleration or steering was delivered: the shared disk boundary permanently closed immediately after that capture. Driving/checkpoint consequence and ordinary Quit remain unverified. Initial car motion is not evidence of user control.
Evidence: scratch/runs/20261118-revolt-startup-dom-obstruction-manual-race/; result.json is sealed last. Read the predecessor's sealed result and recovery before this attempt. No prior scope was reopened, no second original was launched, and no endpoint request followed closure.
Startup observation and exact limits
This run did not reproduce the earlier unnamed overlapping DIV. Its exact class/text/ancestry and lifecycle remain UNKNOWN. The earlier black screenshot predates the failed observation and cannot identify that element. No causal visibility defect or production repair is established.
The new run's passive startup observation recorded the shell dialog DIV#wine-launch-window, class wa-launch wa-launch-windowed, hidden with zero geometry. Its text included “Starting Re-Volt Demo”, “Waiting for the program's first window (0:01 elapsed).” and “Downloaded:7 files, 1.7 MB”. The contemporaneous full output visibility predicate passed. This identifies a shell dialog in this new run, not the predecessor's obstructing DIV. A single hidden observation does not establish the dialog's preceding visible lifecycle.
Source lib/launch-progress.js documents a 500 ms reveal delay, progress presentation and ordinary hide behavior. The run-local operator added class, text, HTML and ancestry to refusal diagnostics without changing the visibility predicate. Before admitting its first scene, a bounded startup observation could wait only on a recognized launch progress dialog with visible progress and no error. That wait branch was not exercised or qualified here. Its exact-class recognition is narrower than the observed windowed class; do not promote it as a general readiness fix. No overlay was hidden, removed or restyled. Unexpected obstructions still close UNKNOWN.
Personally reviewed whole frames show black initial output progressing naturally to Re-Volt's loading screen and main menu. A short Enter was followed by explicitly labeled Demo Mode; neither attract motion nor transport acknowledgment was counted as manual gameplay. A longer Escape returned to the menu. Reviewed manual choices then opened Start Race → Single Race → default PLAYER name → RC Bandit → Toys in the Hood 1 → Summary → race. Nine bounded guest DOWNs have independent release-before-DOWN receipts and physical UP state 0. All were menu/attract-navigation operations; no driving command was created.
At 12:27:56.315 UTC, the shared disk scope closed at its existing 1,520,000,000-byte early reserve. The runtime watchdog observed minimum 1,518,948,352 bytes. UNKNOWN was recorded at 12:27:56.323. race-ready.png completed at 12:27:56.269, before closure. The subsequent acceleration operator call refused locally on UNKNOWN before writing a command file. There were 155 ordinary requests and zero after shared closure or UNKNOWN. No replay, timeout extension, forced focus, guest-polling, clock, config, save, state or media change was used. The 5-second driving freshness policy remains unchanged and no driving gesture reached its gate.
Graph, recovery and cleanup
The frozen current published JS graph is 80c86f1fb89b09e4d88c49d9e7181c5a029cd300, including 06f27181f renderer/admission changes. The retained served module remains a6002d5dbcfeb534a0eadd223da616e525aa1542dad2bde381fe54e00ee862f8, 2,418,710 bytes, with original compiler-source provenance f6962ba5b1cf2829a66f5cc4d80b293c30719b37. These are deliberately separate identities: newer src/main.watx includes the CLI command-line fragment, and this task did not claim to compile current HEAD or execute its module. Optional child Runtime/Debugger attestation was omitted; actual executing module/roles remain UNKNOWN. All 463 existing media pins matched before and after. Published tmpfs/component switches and qualified release/final-wire ordering were reused. Foreign WIP was excluded and recorded as a binary diff.
Offline audit passed: nine DOWNs/nine independent neutral releases, unchanged media, zero requests after closure, 15 actual runtime/release waits with both pipe EOF flags, 61 recorded owned PIDs absent, native supervisor leak=false, profile/display/auth removed. Browser exit 0 is host cleanup; driver exit 1 is not ordinary Quit. The invoking tool returned 143 without the invoker-child wait receipt; that enclosing wait and its separate EOF are UNKNOWN. Direct-driver separate EOF is UNKNOWN because its pipes were inherited by the native supervisor.
Shared disk was already below the 1.5 GB floor when this task began (1,007,947,776 bytes), so startup was withheld until recovery above the existing reserve. Runtime samples stayed above the floor. The separate lifetime sampler also returned tool status 143 without its disk summary or board-child wait; both recorded PIDs are absent, but full lifetime disk compliance and those individual waits are UNKNOWN. Later offline readings are observations only. Only this run's logs were losslessly compressed and its redundant images discarded; selected whole decision frames remain. No foreign cleanup or new RAM/resource gate was introduced.
RECOVERY.md, exact Git manifest, binary diff, owned new-file patch, selected named images, compressed logs and actual waits preserve the evidence. No production code changed, so no code build/commit or causal-fix acceptance is claimed; the focused offline audit and original browser observations are the validation. Only this report is committed and immediately strictly pulled/pushed. No agents, workers, boxes, Python, Wine or deployment were used. Box ID null, state NOT_CREATED, expiry null, stop receipt null. Ownership is released and the worker stops after sealing.