Lords II original after the Win16 PROGMAN repair

Evidence: scratch/runs/20261107-lords2-after-progman-gameplay/. Follows the published Win16 PROGMAN repair, commit ea81b0a3baa4d73d80bbe8baaf37c409fb8b5ac3. Its terminal result, focused regressions, mandatory locked build, publication and explicit 04:11:50Z release were checked before this original. Aranna retained the shared local runtime/source window; this task used a separate remote runtime and changed no emulator source.

One fresh box, bx_mcrph8az, was created with a persisted UUID Idempotency-Key. The 620 individually pinned inputs were recovered from Git 7ff557c013c9c6c1b8319f74c432ca73caedfda3, the relevant patch and the retained untracked-file patch, small run inputs and original media paths/hashes. There was no source/media archive. All common runtime pins matched the predecessor; only the modules changed from its pre-repair original. The exact retained repaired standard module was 1c64b1d38256ccc5f2b774f59f144c707b0a9705359249f4d047c7447e5f6e13, compatibility module fe041599cbe9ac7b856bdb0e732f643864af5bdf835ccf0dd6a9ec6f099f38d6. Both modules were retained locally and individually uploaded; the shared mutable build was not substituted.

Before launch, all pins, media, free disk, Chrome and Node were checked. Both modules passed the native Win16 PROGMAN regression, redraw, WinFlags, temporary-drive, LZCopy and armed/disabled observer passivity tests. The production queued-input validator also passed. The predecessor already published the locked build and broader focused DDE regressions for these exact repaired modules; this task made no code change requiring another build.

The unchanged lords2_demo_setup ran once, with the standard module, Chrome 151.0.7922.108, a 1000×800 viewport and cooperative no-threads. I personally inspected all nine full-resolution captures before further progression input. Eight ordinary clicks followed the retained setup sequence through the default destination, file-copy completion and its OK. Matching read-only state was saved for every capture. The WAV-test Yes preserves the predecessor's ordinary configuration choice; audible output was not independently verified. The setup's CD-ROM/minimum-requirement warnings offered ordinary continuation, as in the predecessor.

The original now passes the repaired path: DDEML.7 receives instance 1, explicit PROGMAN/PROGMAN string handles 1, NULL context and returns DX:AX = 0:1, with SP +20. Five connects and five XTYP_EXECUTE transactions succeed; the latter return DX:AX = 0:1. Two genuine links appear under C:\WINDOWS\Start Menu\Programs\Sierra: lords 2 demo.lnk and uninstall lords 2 demo.lnk. The VFS inventory has 172 entries under C:\SIERRA\L2DEMO. Observer errors and dropped API records are zero. This verifies real shell effects, not only a successful connection return.

At the first new blocker, the installer displays WinExec of Setup32.EXE failed. Its actual KERNEL.166 command is setup32.exe C:\SIERRA.INF A, show command 5, returning AX=2, DX=0 and Pascal stack delta 10 including the far return. The runtime logs no registered app for "setup32.exe". Progression input stopped at the 04:17:52Z capture; command 8 was normal harness Quit, without dismissing the error or launching anything else. There was no repaired-original retry, installed launch, menu or gameplay claim.

The retained original proves the next lookup discrepancy. Its earlier KERNEL.166 launched C:\WINDOWS\TEMP\_SETUP.EXE /o C:. That Win16 task calls the bare setup32.exe while the VFS current directory remains C:\. The unchanged helper exists at both c:\windows\temp\setup32.exe and c:\sierra\setup32.exe, each 132608 bytes and SHA-256 b62ffd44fa2c8a0ac5873320f6d6c86827d6e78bb0af3e605cdf7b8f50fa8218. host.js's resolveShellLaunchPath uses only _resolvePath(file) for this bare WinExec name; it produces absent c:\setup32.exe. resolve-diagnostic.cjs confirms that result using the retained inventory without guest execution. The documented WinExec search order starts with the application's load directory before the current directory. The likely repair must carry the calling Win16 task's module directory into host launch lookup and preserve the remaining search/error semantics. Merely registering a game-specific helper, changing guest paths or forcing a launch would not establish that behavior. No such change was made.

All 181 media files and all 620 staged pins passed post-run hashes. Sixty remote evidence files, including captures, matching states, API traces, actual HTTP-response identities, logs and cleanup proof, were uniquely retrieved and SHA-256 verified. Normal Chrome exit was code 0 at 04:18:04.893Z. The harness, server, deadline and Chrome processes were gone, profile absent and HTTP port 36933 closed before shutdown; the Chrome debugging port was not separately socket-probed after the browser process exited. No local heavy runtime, source/build ownership, deployment, archive or nested worker was used. Free local disk remained above 2 GB.

Normal fresh-box stop was explicitly delegated in this task. stop_ed5059ccea7e46d1 completed at 04:18:43.800Z, forced=false, error=null; box state is archived, active expiry null, original expiry 05:45:16.459Z. The next task should repair and regress Win16-task-relative WinExec executable search, then use another separately authorized fresh ordinary install to reach installed launch and player-controlled gameplay. The retained inventory is evidence, not a restorable installed game image.