Authentic Drakan native warning and gameplay attempt — 2026-11-18

Evidence: scratch/runs/20261118-drakan-authentic-warning-visible-gameplay/ (ignored).

The first registered original reached the actual Drakan Beta Demo Warning before any main window. The complete 1024×712 frame was personally reviewed (warning.png). This was a native NULL-owner MessageBox, class 16, process 1000/thread 1, with the real focused IDOK child. A finite normal Enter was consumed by exact native modal invocation 6/root 2/depth 1; the independent release acknowledged UP. The application then created main HWND 0x10004. Immediate post-release capture refused the retired warning button as current focus, closing the scope UNKNOWN. No later endpoint, replay, level, movement, attack or ordinary Quit is claimed for that scope. Local cancellation retained actual native/watchdog waits, both EOFs and no leaked processes; exit 0 on cancellation is not ordinary Quit.

Preparation used the coherent published 6308e7a3be6c98ab150d07c2f178d8cd09f4b2e6 source graph with three explicit published runtime overlays from af285e4ce94ae1e76f68663871f9ac0495f2937f, and a unique module compiled from that base WAT closure plus the reviewed modal-key correction. Module SHA256: 27555ac96699ebd15eaaec13d634eb067d56fe14e70adb1b8f34ec5e9abb4e5a. The sealed recipe and compile inputs are retained; Git is the source archive. Actual main/Worker source-readback audit matched 232 reads, including the unique module.

The existing 64-path/62-manifest media closure, genuine COMCTL32 28a41e6a35c4509924a32a99841a815cf49e1d3ee6912ec26a823e45472edfa4, COMMCTRL 39bbfafa8b759ebfcf6a00ee878fa494581a9144891bbd788bded61441de13f9, mandatory boot TLBS/fonts and registered Settings100 value were checked through main and Worker paths. Settings100 is the unchanged 92-byte value, SHA256 207f1bdaccfd50d9e739ec3c812a09806eb119af4638351155548e4f7e290f43. Genuine COMCTL32 attached successfully, DllMain returning EAX 1. There were no original binary or registry edits, fake controls/DLLs, private actor input, forced clock, Wine or synthetic ownership.

The measured native preparation defects were corrected: queue physical keys to real native MessageBoxes, consume those keys through the existing exact modal invocation fence, retain permanent process/screen binding across legitimate focus retirement, and permit explicitly mounted genuine COMCTL32 without installing a default DLL path. A follow-up preparation correction uses the existing public presentation-pending contract under the original three-second operation deadline; pending allocates no capture, lease or input. Refusal/UNKNOWN still closes without retry.

Native authored controls demonstrate old failure and new success for a real pre-main NULL-owner warning, native Enter consumption, independent UP, a visibly distinct second modal, and ordinary final Enter exit. A real-tick 350 ms gap between modals demonstrates the immediate-capture failure and bounded readiness success. Local/Worker modal invocation fence negatives, renderer queue tests, process-context negatives, builtin DLL policy, compiled Win4 and genuine COMCTL/COMMCTRL attach/detach tests passed, as did locked full builds. The genuine thunk test does not prove execution of an FT target. Historical Win4 expected-failure and LS32/FT suites requiring absent test-only exports are retained diagnostics, not claimed current full-suite passes. Historical cross-game admission gates were excluded.

All native original lifetimes use native-supervise, a finite 900-second deadline, independent 920-second watchdog and 180-second ordinary Quit reserve. No box was created: ID null, state NOT_CREATED, expiry null, stop receipt null. No agent/worker, public deployment, source snapshot, frame series or media archive was created.

The authorized affected additional original used the exact native-tested readiness helper published in 789b193ca through ordinary merge fd1e36787. Its genuine warning was personally reviewed again, then dismissed with finite Enter; native modal consumption and independent UP acknowledged. The retained afterRelease/postProgress snapshots show the live Riot Engine main HWND 0x10004 while focus remains retired warning IDOK 0x10003, generation 0/live false/thread 0. Presentation did not become ready within the unchanged three-second deadline. This second scope closed UNKNOWN, with no later endpoint, replay or third original. The exact pending failure list was not retained, so the report does not claim a proven exclusive underlying cause for every pending response. The concrete blocker is the invalid post-warning focus/context and failure to obtain a guarded reviewed post-warning whole frame. Native logs also show the Riot Engine and socket sink creation; no menu/level image or movement/attack consequence was obtained. The warning text is clipped in the rendered modal, although full text is available in its public control snapshot; text layout is not qualified.

Both originals were locally cancelled after closure and retained actual native/watchdog waits, both EOFs, and leaked:false; neither cancellation is ordinary Quit. Both launch authorizations are spent. No runtime remains held for investigation. Final source readbacks, hash inventory, process cleanup and exact publication are retained alongside the last-written result.json.