Hexen II transfer-fixed fresh original, 2026-11-20

Evidence: scratch/runs/20261120-hexen2-transfer-fixed-level-enemy/.

One fresh original used exact graph 34c794d549c4ad71dfe40c1647e4ffd0c7a4f8af, the recorded 72dc77da0 capture delta and 65a75893c input overlays. The actual owning CPU Worker bytecode was fetched once and matched 10bdae2c56c79eeb5d0b4ebbc602c8a233e36351380da274a70f7f5e25363564 (2,418,815 bytes). The separate actual renderer endpoint attested gl-gps and WebGL1. No earlier endpoint was contacted or resumed.

Preflight compiled the retained exact module and imported all seven pinned helpers, including the independent physical release implementation, before starting the fixed 20-minute runtime budget. All 18 media pins and the 1,448-file Git graph passed. The actual release child was ready and unarmed before input. The prior focused tests and locked build are reused unchanged; no new shared code, module, controls or build qualification is claimed.

Capture metadata used the unchanged tested transfer helper from 20261120-hexen2-cross-filesystem-capture-enemy/transfer.cjs by direct import. Five real capture transfers passed full-byte/SHA verification and file/directory fsync. The previous EXDEV failure did not recur. Its synthetic transfer campaign was not repeated. Selected whole-display screenshots retain the main menu and Apprentice difficulty menu; intermediate screenshot hashes and scene receipts remain without retaining a frame series.

Personally reviewed main-menu, New Game and Paladin frames each authorized one Enter. All three report native USER-return consumption and successful independent physical UP. A fourth Enter was refused locally as NOT_SENT because its image review expired before dispatch; it never armed the release child or sent DOWN. A fresh difficulty frame was then personally reviewed, but disk closure prevented the next command from being published. There was no later runtime request.

At 2026-11-20T08:22:51.811Z the lifecycle sampled 1,499,267,072 bytes free on /home and permanently closed at its sleep checkpoint. The independent monitors also observed the breach. The task-wide minimum was 1,469,018,112 bytes, so the required continuous floor did not hold despite the changed recovery dependency and 1,642,721,280-byte runtime admission sample. Later free-space recovery did not reopen the scope. Playable level, living-enemy consequence and ordinary Quit remain UNPROVED. This is a resource closure, not a demonstrated gameplay defect.

Browser, independent release, display children, supervisor, launcher, monitor, invoker and task watcher have actual parent waits and output EOFs. Supervisor reports leaked:false; owned PID liveness is checked offline. Cancellation host cleanup is not guest Quit. Unique evidence was retained before owned tmpfs removal. Git plus relevant binary diff, owned helper patch, exact module/media references and recovery recipes form the source record; no source/media copies.

Only this appnote is committed. Publication receipts are in the run. No workers, Python, Wine, shared nativeguard/MCM/Q2/held-region edits, or public deployment. Localhost box ID=null, state=NOT_CREATED, expiry=null, stopReceipt=null; root owns shutdown. All worker ownership is released after evidence sealing.