Disciples II current fresh NewQuest, November 16

Evidence: scratch/runs/20261116-disciples2-current-new-quest/. The November 12 sections of disciples2-demo.md supersede the October progression notes. This task starts one fresh ordinary original quest and does not restore any historical save or repeat the earlier three-unit battle. The original demo's demonstrated AI save/read schema mismatch is outside this task; no save edit, timeout enlargement or emulator compatibility patch is used.

The current 127-row support manifest and separate 4,192,256-byte executable were verified before launch. All 128 installed files match the retained November original-media SHA-256 pins; the executable is adba242a0a3563651baffa6ec7d9a6c0978b7ae844a1cccedb2a1f4e8eb7b1f8. The five registered original DLLs resolve. The default route's missing physical OLE32 warning is explicit: its three named imports are implemented by existing native emulator handlers. No unrelated OLE32 image was introduced.

New task-specific controller, supervisor, source-view and preflight helpers reuse the published native-scene and guarded-input protocols. Source and module bytes are sealed into transient memfds; durable recovery is Git plus binary patches, with media referenced by path and hash. No source/media/Wasm archive, box, worker or shared production edit is created. The module is 692766434b357897de3ad9310edaab8432f9438806183f024b478e3faed9a271, qualified by the current full build and exact public native controller fixture. Its original build provenance is retained as build-source.json and patch.

The actual synthetic Solitaire fixture qualified current native process, USER window generation/focus and permanent consumer binding, finite DOWN and independent physical UP. Confirmation was 23.062 ms and UP 174.381 ms; actual native/watchdog/supervisor waits completed with both pipe EOFs. This fixture does not claim a personal review of the original game. Original actions require their own personally inspected complete host image and current native tuple.

The one original software run has a fixed 1,200-second budget, independently prearmed 1,220-second watchdog and irreversible 180-second Quit reserve. Finite confirmation, physical release and observation bounds are 500, 1,750 and 3,000 ms respectively. No queued gameplay, private game-state reads, clock/config/EXE changes, OCR or solver is used. The registered logical frame addresses remain in the sealed app registry, but no FPS/audio/hardware claim is made and no new performance instrumentation is added.

The original loaded through its logo and black transition to the complete native main menu. The first binding attempt used the personally reviewed black startup scene. The scene changed before binding; the actual HTTP result was ok:false, error:"reviewed native paint changed", timedOut:false. This is a known input-admission refusal before any consumer binding or original game input, not a game hang or a recurrence of the old menu-hover defect. test/run.js latches reviewed-scene binding before validating the pixels, so the same runtime cannot rebind. No replay, relaxed predicate or new original was attempted. A subsequent controller call refused because no owner existed; that closed its local admission as well. The separately reviewed complete blocker-current-scene.png shows Single Player and Quit on the native menu.

This is the task's first actual blocker. NewQuest, movement, a quest/town/turn consequence and native Quit were not reached. Cleanup used the harness's owned shutdown route, explicitly not the visible game's Quit control. Actual original, independent watchdog and both supervisor-parent waits exited zero with both pipe EOFs. The invoking tool also completed zero. The module-holder invoking tool exited 143; its PID and descriptor are absent, but there is no independent normal holder wait/close receipt. holder-wait-qualification.json retains that limit. No box was created, so box ID, expiry and stop receipt are null.

The next attempt requires a separately authorized fresh original and a current complete stable-menu review before the one-shot binding. No production emulator defect or fix is claimed by this run. The failed startup bind does not authorize weakening scene ownership, replaying input or importing a historical checkpoint.

The required locked full build passed after the GL owner released the shared outputs. All four build/watchdog/supervisor waits exited zero with both pipe EOFs, and build-source hashes were unchanged across the build. Both resulting module hashes matched the original run's module family. A final synthetic Solitaire qualification on the then-current sealed harness again passed: confirmation 14.530 ms, independent physical UP 135.642 ms, bounded complete observations, and all four actual waits/EOFs. Its separate final module holder closed normally with invoking-tool wait zero. This does not replace the first holder's qualified 143 result or add original gameplay/native-Quit evidence. Original media/DLL hashes remained unchanged after the run. Build outputs were explicitly released to the queued workers immediately after final qualification.