Drakan current-browser control attempt, November 8

Player movement, flight, turning and attack consequence remain unverified. The original registered demo reached its native beta warning, but the driver required a graphics producer before admitting the warning's OK input. At that point the guest had created no render endpoint or renderer Worker. The unbound readiness loop therefore exhausted its twelve observations without opening the input queue. This is an observed driver bootstrap/admission gap, not evidence of a Direct3D rendering or game-input regression.

Evidence: scratch/runs/20261108-drakan-current-browser-action/. The personally reviewed retrieved/software/unbound-startup.png shows “Drakan Beta Demo Warning” and its OK button. It is a diagnostic image, never an input-admission reference. At 03:42:25.626 UTC the matching state records drakan_demo, PID 1000, running guest Worker endpoint 1 / slot 0, focus HWND 65539, empty d3dStats, no render endpoints, no owning renderer/context and no input events. The twelve subsequent page observations retain that absence; readiness fails at 03:42:31.676 UTC. No commands or ACKs exist. No beta OK, menu selection, level, movement or attack was attempted. The WebGL arm was not launched.

The production URL selected both gl-renderer=software and d3d-renderer=software, with Threads enabled. These selectors do not prove an actual software D3DIM draw: no graphics backend was instantiated. The staged D3DIM-specific producer projection was never reached because the existing prebinding no-endpoint guard returned pending first. Its positive behavior is unqualified and must not be treated as a passing D3DIM readiness helper. There was no CDP timeout or ambiguous guest-input delivery; the terminal failure was the explicit readiness exhaustion. The driver closed and did not replay or start another arm.

Before allocation, one focused generated setup used trusted native Chromium T/G key events, the production DOM/Win98Renderer input bridge, production GuestWorkerLink/RPC, the current full module and generated x86 calls to public GetAsyncKeyState. Both 400 ms holds produced high-bit samples wholly inside the hold and zero high bits after release. Four trusted DOWN/UP events were balanced. The existing independent release timer was reused, adding only T/G VK and DOM-code mappings to the held-key helper. This proves that keyboard path only; it does not prove Drakan DirectInput polling, relative mouse, mouse-button delivery, flight or attack. The local screenshot shows the host desktop/readme, not gameplay. No unrelated suites or emulator changes were made.

All 64 existing fixture files were freshly SHA-256 hashed; all 62 manifest entries match their recorded sizes. No retained per-file historical game digests were located in the available October 7 handoff or cited run paths (the cited historical run directories are absent here). Historical hash equality is therefore explicitly unknown. Current file hashes, paths and sizes are retained in media-preflight.json; resource bytes, working directory and the registered startupRegistry were not changed. The original readme COMMANDS and October 7 a1710870 acceptance were read. That acceptance establishes rainy-level rendering in both thread modes, not player action.

The tested source is e450426c216fbf13d3a711e26d486160241dd17e, with an empty tracked runtime diff and a retained patch for the relevant untracked lib/wininit-boot.js. All run helper inputs also have new-file patches. Recovered actual guest module SHA-256 is f9650607181301b5d2a9990420f437bd9f1ca10d40ac73067339f07499d53121 (2,182,500 bytes). Chrome is 151.0.7922.108 / V8 15.1.206.16. The 677 staged pins were unchanged at terminal audit; 160 complete HTTP responses matched their staged digests, including the original executable and media manifest. No renderer module/context proof exists because no renderer Worker existed. Both retained PNGs passed CRC/decode checks. No source/media archives were made.

All 31 remote artifact references (30 unique digests) were retrieved and independently reread/size/SHA-256 verified at 03:42:51.772 UTC before normal stop. Chrome exited 0 at 03:42:32.043 UTC; driver PID 6327 and Chrome PID 6339 were absent, HTTP/CDP ports 35595/34607 closed, and profile/writer lock absent. The independently prearmed 25-minute watchdog and normal-stop child both exited 0, with an empty supervisor. Box bx_2gqa2txn is archived, stopped 03:42:56.036 UTC, expiry cleared (initial expiry 04:06:22.163 UTC). Stop stop_af6b78f0f57dd2e1 completed at 03:42:56.055 UTC, forced:false. Local and remote disk checks stayed above 5 GB. No resume, delete, old-box reuse, RAM gate, public deployment or nested worker occurred.

Next: qualify a separate native-dialog bootstrap against its actual owning guest/module and presented GDI dialog, allowing reviewed beta OK before a graphics endpoint exists. Then require actual D3DIM producer/presentation and transport readiness before the first graphics binding and menu/gameplay input. Keep exact postbinding ownership, module, freshness, independent-release and UNKNOWN closure controls. A fresh original run must still establish visible terrain-relative movement/heading and a reviewed outgoing attack consequence on each viable backend. FPS, audio, parity and full-game support remain separate.