Age of Empires III trial
Original media inventory (2026-10-10)
- Local source:
test/binaries/win98-games-a-d/Age-of-empire-3trial-Win98-SW.exe - Bytes: 383598224
- SHA-256:
4ef69289dfa0817ec14942d85ef597835a9d2b09e1506c60b9938b20daa274ad - PE32 x86 GUI executable; initial bootstrap contains InstallShield strings.
- No entry in the inspected public
DESKTOP_APPSlist or prior qualification in the task/re-note index. This is not a release or redistribution decision.
Original bytes are staged on temporary boat bx_hx8msa33 with verified SHA-256. No installed game path has been established. No installer approval, gameplay, audio or FPS result is claimed.
Task: NEW-GAME-AOE3-DEMO-20261010; owner root Codex, one worker with serialized runtime resources.
Boat staging completed with matching SHA-256. 7-Zip recognizes PE sections and a383337416-byte [0] overlay, not a ready installed tree. Version resources identify Microsoft Game Studios, Age of Empires III Trial and an InstallShield11.0 setup launcher (2005-04-14). Next capture original bootstrap extraction/child launch without accepting installer dialogs.
Bootstrap observation (2026-10-10)
Original bootstrap starts at 0x0041e77c, creates the InstallShield Wizard, and reports OS checking, Windows Installer configuration, then extraction of instmsia.exe and Age of Empires III Trial.msi. At 04:03:58Z controller332371/guest332378 was still running; guard04:07:20Z. This is extraction progress, not proof of complete MSI or gameplay. Output remains /home/user/aoe3-bootstrap-20261010 on the boat; inspect terminal cleanup and captured files before launching a child.
Original launch (04:12Z)
Bootstrap completed04:04:06Z. Captured MSI375136768 bytes yields Data1.cab174 files/569072090 bytes. Host msiextract reconstructed /home/user/aoe3-installed-20261010/Age of Empires III on bx_hx8msa33, without installer approvals. Original age3.exe SHA2565d7fb9a8ae28e3e3758cee1346b2960f20ba7b4feaa511fa7f52ceffeeeca256.
Explicit seeds of original rockalldll.dll, granny2.dll and deformerdll.dll resolve the initial FAST_HEAP import trap; binkw32 auto-loads. Known msvcrt and Explorer98 shlwapi fixtures were staged; native shlwapi resolves PathIsRelativeW. This is launch dependency configuration, not a new emulator API fix.
Current stop is an empty MessageBoxW, return0x486c17. Both UTF16 text buffers begin with NUL. The original data/StringTable.xml (UTF16,2404654 bytes) opens successfully; only the optional .xmb lookup misses. Do not label this missing XML. No-uop reproduces the modal, so that tier is not required for the failure. No button answered. Next trace string-table parsing and error construction before0x486c17. Evidence scratch/runs/20261010T0410Z-new-games-bootstrap-diagnostics, bounded excerpts and original run identities; full logs remain on boat.
MSXML startup resolved to EULA (04:32Z)
Correction to the interim board/STATUS claim: XP did not originally finish MSXML initialization. The full log reported64 exhausted resumes inside the null heap walk; counting neighboring successful DLL lines was misleading. Both tiers reproduce. A read-only initializer probe found an early GetProcAddress(InterlockedExchangeAdd) failure. The API was absent from the table. Adding its atomic add operation with prior-value return permits MSXML heap initialization. Startup probe confirms a nonzero function pointer and subsequent real heap allocation.
MSXML then requests C:\msxml4r.dll through LoadLibraryExA; the original resource-only DLL was missing from the pre-initialization VFS. Seed it from the MSI System:System directory alongside native MSXML4, plus the native dependency configuration and XP profile above. With that assembly the original game reaches its EULA. No Accept/Decline button was answered. Evidence scratch/runs/20261010T0430Z-aoe3-eula (reviewed); this is a held approval boundary, not gameplay.
Validation: canonical build passes; test-wat-locks --interlocked-only10/10, including two real OS workers, old-value return, signed wrap and sparse/unaligned boundaries. Full suite19/20 twice: its unlocked negative control loses no updates on this host. The parent source/test reproduces that same control failure18/19; no new failure. Evidence scratch/runs/20261010T0432Z-interlocked-exchange-add.
Dynamic OLEAUT32 ordinal repair candidate (2026-10-22)
Accepted diagnostic commit 07037a8a2eb82c4a86cd3ee64a098f3e98173e0a records the executed MSXML4 delay lookup GetProcAddress(0x400000, 8) returning NULL, followed by native RaiseException(0xc06d007f) and actual parseURL failure return. The requested module was OLEAUT32; its LoadLibrary result aliased the EXE, so the handle lost the module identity needed for ordinal lookup.
Against remote main 63d563df7a2b901d222a311ceaf55a0823a22261, append OLEAUT32 to the existing static-module identity list and add host ordinal 8 → VariantInit. Wine's pinned oleaut32.spec establishes the ordinal; the existing WAT static resolver and VariantInit handler already implement it. No handler/API numbering changes. The actual repaired dynamic path uses the WAT ordinal resolver before host fallback.
The full-module Calc-fixture regression reproduces the baseline EXE alias/NULL and verifies the candidate's distinct handle, returned thunk, actual VariantInit execution through the guest run loop, VT_EMPTY/union clearing without release, adjacent storage, saved return address and ESP. Unknown ordinals and unrelated DLL/EXE aliases remain unresolved. This fixture uses its own image base 0x1000000; it is not an original-game retest.
Five candidate test files pass (focused OLEAUT32, existing DirectX/DirectSound ordinals and two BSTR suites). Existing test-getprocaddress-sparse-name.js fails identically on baseline and candidate at “real named export resolves”, 0 !== 2105344. Stop at that dependency; no broad repair, canonical build attempt, deployment or gameplay claim. All 373 retained original pins match. Evidence and exact hashes: scratch/runs/20261022-aoe3-oleaut-ordinal-repair/REPORT.md, HANDOFF.json, verify.py. The assigned box remains running under coordinator lifecycle ownership.
Qualified repair and remaining DirectX prerequisite (2026-10-22)
The preceding candidate-stage limitations are historical. Independent review 854f3bdd3d73434b048ee0e6070c00914628d8f4 accepted repair 24220bf65a04f665c847fc4037a732f8e82db87c and fixture correction 2977ba3bd08e5bde5bab5e5323a5de71a14d6b49: the synthetic DLL must publish its separate lifecycle row LIVE; FREE rows reject both export routes without allocating a thunk. Census adjustment 28379f3d7475317b429c4c18f082c4deb5bbd489 changes only the host-file hash and appended rationale, preserving all 407 classifications, pending ownership and review locks.
Exact candidate c7fae5ac7ea9e29dfa3f83a7c2bd87d092e0b3b2 subsequently passed one normal canonical build, all gates and compilation, at 13:19:05–13:19:27 UTC. Module SHA256 6ac6b43475d9217ffffa98aaec3f57243de47c78c0647dcd31abef404e008678; complete 3420-file source closure and 373 original pins were verified. One ordinary Accept restored the English post-Accept dialog body/buttons; the original native ordinal8 lookup returned actual thunk 0x075071a0. This source integration maps the production bytes to that qualified candidate; it does not claim a fresh main build. Build/runtime evidence and recovery.
The actual DirectX prerequisite remains unresolved. Diagnosis 792b99780de32a1c9e1ff59a6b9795886f5a7b96 found missing DxDiag COM registration and an inadequate fallback version. A separate genuine-provider retest 6146c8862d5a30c800e5e32915cd240fceaecb9f repaired class creation with the real DLL and registration overlay only: CoCreateInstance/Initialize/property calls succeeded, but actual properties reported DirectX7.0 from unchanged registry 4.07.00.0700. Native helper 0x00040007 < 0x00040009 returned Config_InitVideoCardInfo0, with the same readable capabilities-error dialog. No provider/registry overlay is integrated here. Actual DX9 availability, GPU sufficiency, gameplay and actual parseURL/primary XML returns remain UNKNOWN; title <unsupported> is separate. No fabricated versions/caps, guest patch, bypass or AttemptToRun. Native-provider evidence and recovery.
Original installer recovered (2026-11-05)
Recovered the recorded original installer to scratch/game-media/aoe3-demo-20261105/aoe3trial.exe from the Internet Archive demo copy. Its 383598224 bytes and SHA-256 4ef69289dfa0817ec14942d85ef597835a9d2b09e1506c60b9938b20daa274ad exactly match the historical installer pin above. The newer 447485608-byte trial is a different fixture.
Evidence: scratch/runs/20261105-aoe3-original-media-recovery/recovery-plan.json, download.headers and recovery-receipt.json. The bounded download completed successfully; the complete file was hashed before renaming the partial download. No extraction, installer execution, agreement action, game run or sandbox action occurred during recovery. Supporting installed assets still need reconstruction from the recorded hashes, and the historical DirectX prerequisite above remains unresolved.
Replacement bx_byzj52ta was previously provisioned and stopped with a snapshot; current command/runtime availability requires preflight. Superseded bx_hx8msa33 stays untouched. Ordinary Accept/Yes authorization from 2026-10-18 persists; this recovery supplies no new acceptance or gameplay observation.