Conan: preserve genuine installation, then original game launch

Task NEW-GAME-CONAN-DEMO-20261010, continuation from main a5c90bfd0ccc40c96ad457c617e898ae9064bda3. Evidence: scratch/runs/20261105-conan-installed-state-gameplay/. Unique installer-produced state destination: scratch/game-media/conan-installed-20261105/.

The genuine installation is now preserved, including its final INSTALL.LOG and registry. Ordinary Finish completed and the identified original game launched, rendered its menu, and accepted New Game and Warrior difficulty. Gameplay is not established. Its first recorded emulator failure is QueueError: D3D9 draw: texture not resident; it remained on the Kordava loading screen. This is committed investigation evidence, not a tested renderer repair.

Genuine installed state

Reused the working recipe in conan-recovered-installer-20261105.md, with one recreated original installer session and no archive/media discovery. Existing media remains at scratch/game-media/conan-demo-20261105/ConanPC-Demo-from-Herni-Vyber-23.exe, 126741363 bytes, SHA-256 73038ac99299abc8d5e0cef28d83324e6b493fecdbd21fd65964919376fa3007. Historical magazine-demo equivalence remains unknown. The installer was not copied into evidence or the installed-state directory.

Ordinary trusted clicks selected U.S. English, the default destination, I Accept, and Next. The acceptance screenshot shows the radio selection and enabled Next; the following screenshot shows Start Installation. Installation reached its genuine success page.

The observer connects with defaultViewport:null. Before/after attachment measurements remain 1000×800 at DPR 1. Installer screenshots and actual input both use a 660×602 canvas/CSS rectangle at (10,160). The successful Finish click used page (402.7,630.162), inside the reviewed button. It completed by 18:00:35.415Z, well before the actual queue deadline 18:04:51.764Z. Guest ExitProcess recorded code 0 and the shell published the Conan desktop shortcut.

Read-only remote VFS references were held through Finish. Chunked recovery before Finish retained 31 installed files, 156126585 bytes. Recovery after Finish found the additional genuine 4134-byte INSTALL.LOG:

Canonical state is retained in the unique directory above: manifest.json, files/, registry.json, and registry-changes.json. Treat this specifically requested preserved state as a retained task input. preservation-verified.json rereads all 37 files and verifies their sizes and SHA-256 values, the registry against its actual observation, and every route mount. All 31 predecessor-installed files match the predecessor inventory exactly. Final manifest SHA-256: adf7c78e9ab56519edd53479b2151f04e18496668dec450ba2af12cc0b90cd38; registry SHA-256: 64969ff6f94d58b54431e18f937af3374bda5755e233ad62517840e1611d8de3.

The real game remains 1630208 bytes, SHA-256 c343ef6a51a88c5b01a3c3e1511035359d25336fbe010123bd2e02962b77df56. prepare-game.cjs reconstructs only a private app route from the final manifest: that executable, its actual installed working directory, 36 companion mounts, and the exact observed registry. It does not alter installed bytes. The recipes and receipts reference the canonical files; no second media tree or source archive was made.

One original game session and first failure

One original game session started at 18:03:00Z using the Worker backend and SwiftShader/WebGL in Chrome 151 on Linux. Its canvas is 640×480, displayed in the same input/presentation rectangle at (10,194), 660×568 CSS pixels. Passive observer attachment preserved the 1000×800 outer viewport.

Personally reviewed all 20 saved PNGs across installation and game execution. The game screenshots establish NVIDIA splash → normal Conan menu → New Game → Warrior difficulty → “Kordava - Like a Thief in the Night...” loading. Trusted Escape and Enter presses/releases reached the guest and produced these normal menu transitions.

After difficulty confirmation, the first recorded emulator error is D3D9 draw: texture not resident, at line 61 of game/console.log, after the 18:05:14.684Z Enter release. The log does not timestamp that error itself. game/loading-state.json at 18:07:33.072Z independently records the owning bridge's QueueError: D3D9 draw: texture not resident:

The error corresponds to the D3D9 executor's texture residency validation; lib/d3d9-backend.js checks for missing/wrong-sized resident RGBA in Device.levelPixels. The precise failing texture key, dimensions, and first causal command are unresolved. Two bounded render-Worker prototype wrappers delegated to the original method unchanged and collected no matching failures. Their empty records do not prove that no residency failure occurred, nor that the wrapper reached the failing device implementation. No cause or repair is inferred from those empty records.

A trusted ArrowUp hold at 18:09:00.038–18:09:02.171Z reached guest keydown/up (wParam=0x26). Personally inspected before/held/after captures still show the loading screen, without a player character, HUD, or visible movement. Normal menu response is proved; player-controlled gameplay is not.

No source/guest patches, forced scenes, fabricated DirectX state, second installer recreation, second game session, or speculative API fix. The next technical step is to isolate the failing D3D9 draw/residency transaction with its texture keys, pixel extents, releases and ordering, then derive a meaningful generated renderer regression. This run does not justify changing a Win32/x86 handler, residency budgets, or scene selection.

Build identity, checks, and cleanup

Source-only preparation lasted until Aranna1850 FINAL 17:33:51Z, legacy1915 FINAL 17:43:35Z, and AoE3DXSETUP1925 FINAL 17:55:42Z. Independent cleanup verified 326 recorded PIDs absent, 133 ports refused, 33 profiles absent, four locks free, no heavy runtime, and more than 5 GB free before execution.

Installer source commit: fb256f9f6456e1d790e9fe0813adaec28cdadc35. Game source commit: 26353de28d1132ef6630a5cfae870c985496f05f. Each session records all 301 relevant tracked-source hashes; final rereads match every hash. Relevant tracked binary diffs are empty. The preexisting untracked lib/wininit-boot.js is recorded as a small patch and hash, never edited. Owned new documentation is also retained as a patch.

Both actual browser sessions compiled the same 2173559-byte module, SHA-256 2e4280f8d513b99d57273cb70838871e03088c84d49d64357a66ddb318c8f8b3; one unique copy is retained as tested-wine-assembly.wasm.

Current production boot helpers, DLL graph, system files, bitmap/subset fonts, all genuine game route file pins, production queued-input validator controls, and private script syntax passed. Static audit finds live handlers for all 70 installer and 130 game direct imports; dynamic behavior is not thereby qualified. flock /tmp/wine-build.lock bash tools/build.sh passed before each original run. No production repair or new emulator regression was committed.

After the installer's queued quit and final readout, its outer tool caller returned status 143; caller/probe were absent but Chrome remained. The cause is unknown, with no terminal/Chrome-exit receipt. Cleanup-only CDP Browser.close on its exact owned endpoint at 18:01:41.506Z returned successfully; all three PIDs, both ports and the profile were then absent. Do not infer a normal probe exit from that recovery. The game caller used a separate process session and its queued quit completed before its deadline: probe code 0, Chrome code 0, normal caller cleanup at 18:09:25Z. This is browser cleanup, not normal guest game exit. Game PIDs/ports/profile and four locks independently passed cleanup.

Evidence remains under this run ID; no unrelated candidate-corpus entry is attached. No boxes, subworkers, public deployment, explicit platform flag, or laptop Heroes II work. Box ID/state/expiry N/A; coordinator lifecycle unchanged.