Jedi Knight original demo: accepted licence, blocked in setup
On 2026-10-20, the original installer displayed its demo licence and an ordinary mouse click on Yes advanced to Welcome. Setup then reached Select Components, where an ordinary click on the visible Next button was delivered to the underlying progress dialog instead. Installation did not finish. The original game was not launched: its advertised installed executable was absent. No playable, browser, audio or rendering qualification is claimed.
Identity and complete fresh closure
Runtime: bx_6czhvg6m, Linux x64, Node v24.18.1. Source checkout:
/home/user/midi-remote-only-qualified-20261020-1447/source, HEAD
6b96535d00442d4581b65d54e04abc31ae65febe. All 6,915 candidate tracked blobs
match production candidate 14792a36a6cf1aa21f17f64615d5932d07e8d94f;
the qualification-only tools/qualify-midi-native.js is absent from that
candidate tree and is not part of this runtime route. The preserved pre/post
scripts check every candidate blob but report rows.length - 1, incorrectly
subtracting an exception that is absent; their 6,914 summary is off by one.
Canonical module SHA-256:
11aee57cb158ffd734534b97aef9c36e3152aec660c8371d609a771e4eee9a5f.
No build, production fix, new stub or source modification was made.
Original jedidemo.exe SHA-256:
8aae38165e08b68bb63cfd21b8c28ed1d4d8f1bb400a0162588d5b86acdc827e.
All 39 extracted files were rehashed before execution and after retirement.
The recovery recipe and inventory are in commit
57db4c964fb82552b7c014a1ffcc58e1bf7815b9,
jedi-original-recovery-20261020.md.
Fresh ordinary setup.exe bootstrap used the original directory as --vfs-tree
and the existing --capture-launch, --save-vfs and --reg-export mechanisms.
The host's 34 disk-backed system/font inputs were checked before execution;
its existing generated system identities were left unchanged. Bootstrap exited
0 after 602 batches, with the known B0071612 child handoff and caught thread
cleanup warning. Its new capture contains 96 files plus launch.json, with
all original aliases and all 48 host support paths materialized. It is not the
historical nine-file partial capture. The complete capture was sealed before
child execution and checked unchanged afterward.
The actual receipt names c:\windows\temp\_ins0432._mp, empty arguments and
C:\ cwd. Child SHA-256:
12a2929a615924e212490f4ef3a1cbb36ecebddafea0f358991a0fc57b7cab84.
Replay used this receipt, the complete fresh capture and genuine bootstrap
registry/INI export. That export is end-of-bootstrap state, not a launch-time
registry snapshot. No registry edits or guessed command-line options occurred.
Observed ordinary route and blocker
Authorization is the separate user record
scratch/license-acceptance-authorization-20261018.json. That record's
acceptedInGuest=false was not edited. This run independently establishes
acceptedInGuest=true: reviewed license.png, the successful click:415:395
receipt, and after-accept-corrected.png showing Welcome.
The observed sequence was licence Yes → Welcome Next → Install Jedi Knight Demo Next → default 320×200 Next → default destination Next → Start Copying Files Next → Select Components. All inputs used the renderer's ordinary mouse down/up path. The first child run reached the copy confirmation, then its 120-second guard ended it normally. One fresh bounded continuation replayed these observed choices with screenshots before every click; it reached Select Components after briefly showing “Copying files...”.
At Select Components the visible Next button is around (415,395), in top-level
window 0x1006b at (80,59), 480×362. The ordinary click was logged at batch
1126100, but mouse hit-test/down/up messages explicitly targeted 0x1000f,
the still-visible underlying child progress dialog at (155,304), 323×139.
The Select Components page remained visible and lost its active caption.
One ordinary keypress:13 produced WM_CHAR 13 targeting focus 0x10001 and
also did not advance it. This was a character-input attempt, not a full
keydown/keyup Enter sequence. Snapshot and before/after images corroborate the
stalled page. No forced WM_COMMAND, guest memory changes or alternate launch
route was used. The exact deeper cause of hit testing/focus is not diagnosed.
The normal installer wrote registry entries for
C:\Program Files\LucasArts\Jedi Knight Demo\JkDemo.exe, Jedi.EXE,
SysCheck.exe and Resource\Res1low.gob, but the final VFS contains only
deisl1.isu (2,940 bytes) beneath that target directory. No installed executable
or game assets exist there. The UI had also advertised “0 MB” before copying;
why no payload was copied remains unqualified. Registry values alone do not
establish a successful installation. This concrete setup/input and missing
payload boundary prevents an original installed-game launch within this task.
The continuation was stopped through the existing control quit operation;
terminal exit 0 is a controller stop, not installer completion. All guest
processes were absent at postflight. Runtime was released at 15:35:37 UTC,
before the 15:47 cutoff; the sandbox was neither stopped nor extended.
Preserved evidence and checks
Local run: scratch/runs/20261020-jedi-original-accept-launch/.
evidence.tgz is 1,439,552 bytes, SHA-256
1620acbcca8938ed0a83898d1199cc65f974025bd1ed10367719d58a057a1deb,
recovered and verified at 15:35 UTC. It contains launch argv, logs, terminal
receipts, per-stage VFS seals, registry exports, ordinary input receipts,
screenshots, postflight, and the fresh capture's generated/support bytes.
Large original payloads and duplicate VFS trees are omitted; restore originals
with the pinned recovery recipe and validate against capture-seal.json.
Do not mistake this evidence archive for an immediately mountable full tree.
Remote evidence was /home/user/jedi-original-accept-launch-oct20; TTL was
15:53:12 UTC, so local archive plus Git recipes are the durable recovery lane.
Some input receipt filenames for the continuation were written under child/
by the external observation script, although timestamps, current port, log and
screenshots identify the active install/ process. Both are retained. Early
harness mistakes (a mistyped expected module hash, a nonexistent inventory
filename, an invalid async wrapper/control URL and rejected click-object form)
were corrected before their dependent action; receipts remain preserved. The
rejected click did not constitute acceptance.
Validation checks: original/source/module identity before and after; 97-entry fresh capture seal; 39 capture-original aliases; three exit-0 terminal receipts; actual acceptance-to-Welcome evidence; observed input routing; final target payload absence; archive SHA-256; exact report-only Git scope and diff check. No native suite or browser run was repeated. Candidate is unregistered, so this is investigation evidence rather than an unrelated dashboard attachment.
Shared HEAD/index/config, unrelated WIP, laptop ownership and automated-review holds were preserved. The report uses an exact sparse task worktree after a projected checkout admission; no main merge. Local cap is 4 MiB, with the 2,210,398,208-byte floor plus 32 MiB reserve and current concurrent reservations. The next task must be assigned by the coordinator; this worker stops here.
Independent report integration review
Reviewed candidate c0de2a2040954dc1ff58668f19e7b619fe442855 against fresh
origin/main at 71a60db4004e5df74cfa232f7a59edc109ffc660. Independent
checks and receipt are in
scratch/runs/20261020-jedi-acceptance-evidence-integrate/.
All recorded evidence hashes were rechecked, including the archive; 57 retained
capture members were hashed from the archive, and 39 original aliases matched
the recovery inventory. Local Git objects confirm all 6,915 candidate blob
identities in the recorded checkout. Remote source/module identity rests on
the preserved successful pre/post commands and their inspected verifier code;
this review did not re-read the remote filesystem or execute the module.
Five screenshots were independently inspected alongside the acceptance, mouse-routing, keyboard-focus and final VFS records. Yes acceptance and the stalled Select Components page are supported. Routing causality remains under separate source investigation; neither a source fix nor a causal explanation for the missing payload is established here. The final target contains only the uninstall log: no installed-game or playable claim is justified. The source count and recovery link were corrected; original evidence was left intact. No runtime, browser, build or download was performed.