Jedi original Play passes prerequisites, then cannot start the game
One ordinary installed Jedi.exe session with the two genuine DirectX files
passes the original prerequisite predicate. The first new blocker is the
ordinary WinExec handoff: the unquoted executable path under C:\Program Files
is split into C:\Program.exe plus parameters. Both launcher attempts return 2,
and its original “Unable to start the Jedi Knight Demo game program” dialog
appears. No child engine starts and gameplay remains unqualified.
Evidence: scratch/runs/20261103-jedi-provisioned-original-play/. This is a
findings-only change; no production source, guest bytes or registry were edited.
Genuine provisioning and identities
The published prerequisite recovery
is reused, with all 41 retained artifacts independently size/SHA-256 verified.
The signed 12,144,784-byte DX81 package retains SHA-256
d1762981f497a623b2f3e725c0916b96e716bd4a8845c5a4cc7f764b1a5cc5f7.
The only added file visibility is through the supported explicit VFS mounts:
| System file | Bytes | SHA-256 |
|---|---|---|
| DSOUND.VXD | 48,170 | 92fd75ba86688d7058a449fb37242f2bfb57bc6057d92fb06d5b127212dea404 |
| DDHELP.EXE | 31,744 | fa21dafc3c4052824861fbca1acb06b225ca51b0de93dd255aedbbce533315cf |
These are file-only prerequisites, not a complete DirectX installation or
native VxD/DDHELP execution. Existing DDRAW/DSOUND builtin execution policy and
their accepted 1,024-byte file entries remain unchanged. Restoring those
accepted entries does not retain a staticModule entry tag; builtin-only
execution is selected separately by lib/dll-registry.js. An initial scratch
verification assertion confused those two layers and was corrected without
changing production code or repeating the guest session.
Alice35407 and GL64819 exact threads each had actual terminal0 and exact
AUTHOR RELEASE before this task built, allocated a boat or ran a guest.
GL released commit c1f211fdc027052994180ca52b0bff873715cd99 at 05:08:26 UTC.
A fresh flock /tmp/wine-build.lock bash tools/build.sh then passed. All
4,156 sealed current source/support inputs remained identical during the
build and after the guest. The 2,081,417-byte module has SHA-256
36a80f073b3bc4f59bb66d8c292684085c111be42cf999523723ff0cfab64788.
current-source-inputs.json.gz has SHA-256
cfdf83f296ed5d5686aec2e859867949fa69e2676336c501e9e7369c114590a9.
actual-module-inputs.jsonl records this module as both the actual compile
input and completed instance. There is one guest process and no child receipt.
The accepted original package recovery again verifies 361 mappings and all 28 installed originals / 39,900,850 bytes. Before and after observations also hash every original in the actual VFS, both added genuine files, and all 219 staged runtime members. All 136 observed VFS entries are unchanged. The accepted installer registry, both live captures and final export compare equal across all 76 keys. No DirectX version or installation metadata is added.
One ordinary Play and the new blocker
The personally reviewed retrieved/session/before-play.png shows the original
launcher with its Play button. One renderer click at (320,284), batch
1,036,084, produces the following observed path:
- All four required names resolve: DDRAW.DLL, DSOUND.DLL, DSOUND.VXD and
DDHELP.EXE. The two newly provisioned files have their exact original sizes.
All four existing optional
CompareFileTimecalls return +1. - At batch 1,038,212 the predicate returns
EAX=0at0x4028af; batch 1,038,213 reaches0x402905. Neither missing-file branch0x402f1anor DirectX-dialog branch0x4028b3is taken. - The original launcher calls
WinExecwith the unquoted commandC:\Program Files\LucasArts\Jedi Knight Demo\Jkdemo.exe, show command 5. The current host parser splits the first whitespace and adds.exetoC:\Program. The CLI therefore looks forC:\Program.exe, which is absent, and returns 2. A second original attempt withJkDemo.exedoes the same. Between them, the launcher's ownFindFirstFileAfinds the actual installedJkDemo.exe, size 1,388,032 bytes. - The personally reviewed
retrieved/session/after-play.png, batch 2,260,976, shows the original “Unable to start” dialog, rather than the prior DirectX requirement warning. No suggested direct JK.EXE launch was followed.
The source-only launch-parser-observation.json reproduces the current parse
and includes a quoted control without executing a guest. Relevant code is
host.js's parseShellLaunchCommand / resolveShellLaunchPath and the ordinary
spawn boundary in test/run.js. This localizes the observed behavior; it does
not establish which Win98 path/short-name or launch semantics should change.
There is no proposed engine repair or old-fail/new-pass claim in this task.
The next bounded work must establish faithful Win98 behavior and process-path
provenance before changing that contract, then retest the same ordinary route.
The session was stopped at this new blocker using harness quit: exit0 after
68.569 seconds. final.png is the reviewed solid teal desktop after quit;
the blocker capture is after-play.png. No gameplay controls, audio, FPS,
engine initialization or native driver execution are qualified.
Retrieval and delegated shutdown
Boat bx_k44k9b4f had finite TTL 1,200 seconds and original expiry
2026-11-03T05:30:07.205Z; its ID, expiry and this worker's delegated shutdown
ownership were recorded before use. The 256,993-byte evidence archive and all
31 members were retrieved and full-SHA verified before the stop request.
A separately retrieved/hash-verified postflight record at 05:13:15 UTC shows
no scoped guest/controller/supervisor processes and no 8143/8144 listeners.
Stop stop_b0beb044b0ea8b96 was accepted at 05:13:15.411Z with error:null.
This worker selected --no-snapshots, which removes the sandbox/stop metadata
after shutdown. Subsequent info and scoped stop-status queries return 404,
the scoped all-state list has no such boat, and its preview returns HTTP410
“preview archived”. Those are the actual retained shutdown observations.
A literal terminal STOPPED / completed-stop receipt is unavailable;
the pending acceptance receipt must not be relabeled as completed. A global
stop query was denied by the API key and was not pursued. Future tests should
retain normal lifecycle metadata through completed-stop verification. No
boat was resumed or left available, and no new boat was allocated to hide
this evidence gap.
The brief pre-runtime local disk shortfall was observed, flagged, and cleared by coordinator-owned lossless cleanup before any build/boat/guest operation. No subworkers, browser run or public deployment occurred. Findings and recovery are published; Alice's next task follows this worker's actual terminal/release.