ReVolt release standby lifecycle, 2026-11-20

tools/revolt-release-standby.cjs owns one unarmed, import-only release child. Repeated captures reuse that child without an operation nonce or endpoint. A gesture consumes it once under a fresh name after the existing current-owner and whole-frame checks. Configuration publication is atomic; there is no fallback spawn after consumption. An unused child is signalled and joined without publishing configuration or guest input. A failed independent release overrides a local NOT_SENT refusal and closes the scope.

The supervisor starts the owned browser and its import-only child under a finite setup bound. It creates the unchanged fixed runtime envelope only after actual preparation, child liveness and silent output checks. Callers using the adapted browser must use this supervisor handoff instead of writing an envelope before browser startup. The unarmed import wait covers that runtime; the independent armed fallback remains 5000 ms, gameplay review remains 5000 ms, and holds remain at most 1200 ms. Existing owner, display, physical-neutral and trusted-UP controls are unchanged.

Evidence: scratch/runs/20261120-revolt-release-standby-lifecycle-checkpoint/. Focused lifecycle tests cover repeated unconsumed preparation, one-use consumption and fresh identities, stale/replaced owner and UNKNOWN refusal, silent retirement and unexpected child output. The actual whole browser operator completed two captures using its prebudget standby, then a finite trusted ArrowUp and independent matching UP. A subsequent unused standby was retired with real wait/EOF after an authored query exception. All actual fixture Worker modules/roles were attested once before input; the exception closed the scope with zero later requests. These actual results precede the original run, rather than substituting six local caller checks for runtime proof.

The affected original used graph 6d5558bacfa78946c235d4ac9fb79dd774445db0, retained module a6002d5d… and all 463 unchanged media pins. Two CPU Workers and the d3dim-gps WebGL render Worker were attested once at startup. Ordinary menus entered configuration; prompts were completed deliberately as Left LEFT, Right RIGHT, Accelerate UP, Reverse DOWN, Fire CTRL, Flip SPACE, Reposition Z and Pause ESCAPE. The frame before Pause shows the first seven assignments, and the final Escape returned normally to Controller Settings. This is an ordinary UI configuration, not a claim that these are factory defaults. No Escape was assigned to Reverse.

While returning toward Start Race, the actual disk gate observed 1497059328 bytes and permanently closed at 1795148699725. Later commands were refused locally; no endpoint was retried or successor original launched. No driving, first checkpoint, lap, ordinary Quit, promotion dismissal or native guest exit was established. Those outcomes and exit code remain false/null. Temporary attract-demo threads were not separately attested; no original gameplay input followed. Both actual supervisors report leaked=false, and browser, release, display, enclosing parent and watchdog waits are retained. The direct tool completions have no exposed outer stream-EOF flags; those remain null.

The locked build passed before the floor event. Final focused tests passed 18 groups; that final invocation included the existing actual-Chrome DOM test after the floor event, which is a resource-policy limitation, not an original-runtime recovery or a floor-pass claim. Earlier focused checks and actual operator evidence passed above the floor. The writer responsible for the later disk fall is unknown. No box was created: ID, expiry and stop receipt are null. Publication status is recorded in the run; local commits are not proof of publication.