Lords II recovered original setup — 2026-11-05
Goal NEW-GAME-LORDS2-DEMO-20261015 reached a new, precise setup failure. One original run reached Main Menu, Install, System Test Yes, and Sound Card Test - Wave. Answering No to the voice question crashed child T1. No installed game or gameplay was established. No new review rejection or original rerun.
Canonical media is scratch/game-media/lords2-demo-20261105/original/.
The static-recovery worker verified all 181 files (14,382,323 bytes) against
test/candidate-corpus/lords2-demo-files.json, with no differences. Its findings
and absolute hash manifest are in scratch/runs/20261105-lords2-static-media-recovery/.
Archive SHA-256: afef0f0dd7012c46f483ab09d785431ba5813c2080d8468a52e8ec32c745f49d.
README directs users to unpack with INSTALL.BAT, then run SETUP.EXE. INSTALL.BAT
contains L2demo.EXE -d; recovery already produced the original distribution.
SIERRA.INF asks for a destination, copies FILES into it, creates LORDS2.EXE's
program item, and displays the DirectX 2 requirement. FILES is not evidence of
an installed game. No guest byte or setup decision has been changed.
Run evidence and prepared caller:
scratch/runs/20261105-lords2-recovered-original-gameplay/recipe.md.
After the queue gate, prepare-route.cjs verified all 181 originals, created
one original-directory symlink under the registered candidate fixture, and
called the production browser-manifest writer. It avoided the existing helper's
recursive media copy. Its 180-support-file manifest SHA-256 is
a1749143a2de176c4f3fdf314337a1767f6101eed4a88b6d5b0dee4d59012f8e.
The browser used lords2_demo_setup, the existing input harness and an observer
of exact compiled bytes. The distribution FILES directory was not flattened
into an installed game, and the guest's install copy step was never reached.
Required queue order: stronghold-native-coverage-prerequisite-0950, then d3d7-processvertices-clip-extents-1010, then return-fire-worker-body-transport-1020; each must publish exact FINAL runtime AND file RELEASE ALLBUILDSFINISHED with normal cleanup/free locks. No tests, builds, or runtime before that gate. Deadline is 16:00 UTC. No boxes assigned; box ID/state/expiry N/A. Coordinator owns shutdown and Telegram reporting.
Coordinator sequencing override at 11:33:38 UTC inserts CPU-SMC1120 before RFworker1020 because CPU shared edits were already present. Current order is CPU1120 -> RF1020 -> Lords1105 -> Strong1135. Acknowledged 11:46 UTC. Strong0950 FINAL 11:08:27 and D3Dclip1010 FINAL 11:32:43 were observed; CPU1120 FINAL 12:26:47 and RF1020 FINAL 12:52:32 were then observed. Fresh 12:53:39 verification found named predecessor process identities absent, profiles removed, ports refused, and build/runtime locks free. No test, build, or runtime ran before that gate.
The existing USER230/262 repair bedaaaa7f is prior history and is not a new failure. Further emulator changes require an actual current mismatch.
Actual original run
Runtime main commit cf5e9c1631a8593a9b553bec248c12650e3050c2;
301 relevant physical source files unchanged at postflight, empty
runtime-relevant.binarydiff. Git remains the source archive.
Actual browser-compiled module: 2,171,349 bytes, SHA-256
bc4cabf7745ac4644c3d8cc1f5b70d798277c2a4525e3c665b98d08d6577894b.
Chrome 151.0.7922.108, Node v24.18.1, cooperative backend, fresh profile,
900x720 page and 640x571 guest canvas. Media/module identities and exact argv
are preserved without source archives or repeated media copies.
Personally reviewed startup.png, after-install.png, after-system-test.png,
and after-wave-no.png. Ordinary renderer inputs, with saved replies, were
Install (509,147), System Test Yes (510,235), and wave-question No (130,284).
Audio was not verified in this headless run; no audio-success answer was faked.
The results window's Phantasmagoria template does not prove a completed test.
The wave dialog clips lower text and the No button: read-only enumeration
found dialog 0x1804f at (69,189), 502x103, and child ID7 at client (16,64),
100x28. The last click used its exposed portion through ordinary pointer input.
After No, T1 trapped unreachable at EIP 0x1aa26c, previous2 0x1a9cf8,
ESP 0x1f8550, CS 005f/base 1a0000, SS=DS 0087/base 1f0000,
ES 005f, BX 869c, SI 88a4. crash-detail.json retains all captured
registers, segment bases, 128 stack bytes, and WASM stack. Stack starts
97a25f00a4888700: return 005f:a297, then object pointer 0087:88a4.
The child exited with code1; setup windows disappeared into teal desktop.
The ordinary combined.wat function-index scan was rejected as inaccurate for
this macro-expanded module. A diagnostic name-section compilation reproduced
the actual canonical module SHA exactly; all noncustom sections of named and
canonical modules match. Verified stack: 848 win16_set_sreg (actual trap),
858 win16_far_transfer, 863 th_call_far_mem, 13880 run.
This is an unmapped CS rejection during an indirect far call. Its selector
value was not retained because the route's integer diagnostics were not verbose.
Runtime-generated _setup.exe was preserved from VFS for read-only mapping:
528,080 bytes, SHA-256
7dbf696d4f08464ca1e187686368fe89c5b71b3162c094f463d0e5ec4eaca567.
Original segment4:a26c contains the prologue; a290 loads ES:BX from ES:[SI],
a293 calls far ES:[BX+64h], returning to a297. Segment4:9cf8 calls that method.
These matches imply module arena0x170000 for this run; the actual live module
table was not captured. This is an inference. The initial historical-base
disassembly is retained and superseded by the explicitly corrected mapping.
The registers point to the far-call operand at 005f:8700. Its actual relocated bytes and the object's writer were not captured. Raw original relocation-chain bytes cannot substitute for runtime memory. No emulator repair is justified yet. Next warranted diagnostic: retain that four-byte target, object0087:88a4, its vtable pointer, selector/module ownership table and rejected selector at the existing far-call failure. Distinguish relocation, pointer lifetime or operand translation before changing code; do not relax CS validation or alter the guest target.
Validation and cleanup
All181 media hash/size checks and manifest assertions passed. Diagnostic canonical/compat/named compilation passed, with canonical bytes matching the loaded module. Four captures were reviewed. No shared emulator code or tests changed; no emulator repair/full-build acceptance is claimed.
The caller unexpectedly returned143, observed by12:58:14; cause unknown, normal terminal/cleanup receipts absent. This is separate from the captured guest trap. The surviving owned browser was reconnected only for Browser.close, which resolved12:58:46; its eventual exit code is unavailable. At13:02:15 caller/probe/Chrome PIDs were absent, port46261 refused, profile removed and build/local-runtime locks free. See recovered-browser-close.json and runtime-cleanup.json. No review rejection was received and no original rerun occurred. The fixture symlink/manifest and runnable recipe are retained.