Kingpin: Life of Crime Demo (Xatrix / Interplay, 1999)
App kingpin_demo (candidate kingpin-demo, localOnly), alpha demo v0.22n.
Quake II engine: kingpin.exe imports only system DLLs; ref_gl.dll (or
3dfxgl.dll as the GL driver) and main/gamex86.dll load at runtime; the
data is main/Pak0.pak plus loose main/ files. Registered 8373af6cd.
The readme licence forbids decompiling or disassembly: debug it from traces and runtime behaviour only.
Source and licence
The archive.org item win-98-games-k-o.-7z is one 5.9 GB 7z, but the
download service serves single members of it:
.../Win98_Games-K-O.7z/Kingpin-DEMO-OpenGL.EXE (110023168 bytes, SHA-1
551cd726…), a collector's WinZip self-extractor of the demo tree that
7z x opens. readme.txt: personal noncommercial home use only; no
copying, distribution, decompiling or disassembly. Hence localOnly.
Running
- Args
+set vid_ref gl +set gl_driver opengl32: ref_gl through opengl32. On the CLI GL needs--headless-gl(or--gl-renderer=software). - Slow first-level load and the loopback timeouts. After GAME → EASY
the LOADING plaque stays up from about b500 to b3000 at the 200 ms/batch
headless clock, with almost no API calls (CPU in ref_gl/gamex86). That is
more than Quake II's 120 s
cl_timeout/timeout: the console shows "Loading plaque timed out", "Can't find pic: net" (the lag icon) and then "Server connection timed out", and Escape lands in the main menu. Either--tick-ms-per-batch=20or+set cl_timeout 100000 +set timeout 100000gets into the level; the registry uses the cvars, since they also apply in the browser. The real fix is a faster load: worth a profile. - The intro cut-scene runs to about b6000, then first person in the alley.
Route (headless, --batch-size=50000)
Enter at 300/400/500 (GAME → EASY), space at 5000, Up from 6200 walks into
the wall ahead, Right at 6550 turns toward a corrugated-metal shed. Evidence:
scratch/runs/20261011T0800Z-kingpin_demo-gameplay-claude1863.
This historical headless route is separate from the browser evidence below.
Original browser gameplay reviewed, 2026-10-31
Player-controlled browser gameplay reached. No emulator repair was needed.
Evidence is local and ignored:
scratch/runs/20261031T1914Z-kingpin-native-browser/; the controlled session
is in its browser/ subdirectory. result.json, reviewed-scenes.json,
browser/verification-summary.json and full-release-final.json record
the observations, screenshot hashes, identity checks and cleanup.
Provenance and tested identity
Before launch, downloaded the collector repack named in
test/candidate-corpus/manifest.json and verified its full SHA-1
551cd7267945aa13ff9b5cfdd6991d56279e2738 (110023168 bytes; SHA-256
2a5511a7f570889938165606ced3f24de0daa718ba1bf7335744122eb0aeb10a).
All 800 installed files matched the corresponding ZIP members by SHA-256,
including the executable
f95dff8f4f97890679e4db4179e8dc39a43d0c1d2871dd4bca901af6aa122f04.
All 799 manifest entries were present and size matched; the executable is
additional to those entries. provenance.json preserves each member hash.
The temporary downloaded repack was removed after comparison. Recovery uses
the existing corpus recipe (node tools/fetch-candidate-corpus.js --id=kingpin-demo) on an absent fixture; do not overwrite this preserved tree.
Published main was reread and matched the origin remote as
2380f286559c166ac1161405fd2d3eea87bfe5b9 immediately before runtime
(input e875d370578dee0318b75bceb51b0238fca21e28 is its ancestor).
Origin used HTTPS for that check; setting GIT_SSH_COMMAND alone did not
change its transport. prelaunch-transport-correction.json records this
correction to the harness metadata. Final publication explicitly rewrites
origin to SSH for the strict-known-host pull, push and remote-main check.
Executed module SHA-256:
d39f362187ce89abda47fa53bca6bb6576badd8d21b78e44c6fbad973d1636d8.
The previously tested 362 source and 866 compiler files matched before launch.
The controlled session pinned 528 source/UI files, 866 compiler files and
821 fixture inputs (800 installed files, manifest, shared DLLs and typelib);
all hashes and the module matched again after shutdown. No build was rerun
and no emulator source changed. Fresh public lib/apps.js remained SHA-256
63bb9815e404dd7e446d3748e195c621782a04c8358580bd2cf1a864cb57e209,
with kingpin_demo absent from public DESKTOP_APPS.
Visible native route and input response
On a private loopback HTTP server with COOP/COEP isolation, open
index.html?debug&no-log, select Kingpin Demo, retain WebGL and
default Threads, then click Launch. The existing registration,
manifest and arguments above were preserved. Chrome used a native
1024×768 content viewport with defaultViewport: null; the original
gl_mode 0 game surface remained 640×480. Audio and clock stayed normal;
assets, INI/config, saves and guest state were not edited or injected.
From the original splash, Escape → Enter (GAME) → Enter (EASY) starts
the game. GAME directly presents EASY/MEDIUM/HARD, with no separate New
Game row. Reviewed captures 02/04 show the menu and difficulty;
05 shows LOADING, 06 the scripted third-person alley intro, and 07/08
the settled first-person alley with crosshair and health 68. The intro alone
is not counted as player control.
Balanced trusted browser input then produced these personally reviewed results:
09-native-forward.png: Up moves from the open alley toward the brick wall.10-native-right-turn.png: Right turns toward the burning barrel.14raises the initially holstered pipe;15-native-pipe-swing-held.pngshows Mouse1 swinging it. This is melee attack, not firearm discharge.17-native-jump-held.pngand18-native-landed.png:araises the viewpoint and it returns after landing.19-native-crouch-held.png:clowers the viewpoint.21-mouse-look-settled.png: trusted mouse movement changes yaw and pitch.23faces the open alley again;24-native-strafe-left.pngshows comma translating left toward the fence without a turn key.25-native-pause-menu.png: Escape returns to the native menu.
The original Pak0.pak/default.cfg was read as configuration data to verify
bindings: Up/Down forward/back, Left/Right turn, comma/period strafe,
Mouse1 attack, a jump, c crouch, f activate, / next weapon, Escape
menu. Space is backward, not jump. Reading the original config and
readme involved no guest disassembly or decompilation.
Bounds and limits
Runtime waited for the current Slave Zero filtered-audio NEWFULLRELEASE
(19:17:58 UTC), then independently checked the entire Chrome/global crashpad/
profile process set while holding inherited lifetime FD9 on
/tmp/wine-browser.lock. Python Popen.wait retained the lock through all
runtime cleanup. An initial 19:19 session was closed before guest input:
the scratch HTTP allowlist had omitted optional stdole2.tlb and UI assets.
Its complete cleanup was verified before the corrected 19:20–19:24 session;
only one browser ran at a time, within the initial 19:36 target deadline.
harness-correction.json preserves this limitation and correction.
Corrected-session checks passed: 285 HTTP responses all isolated, 283 served hash joins, 52 trusted events with 3 mouse and 13 key down/up pairs balanced, and all inputs released. Disk minimum was 2415022080 bytes, above the 2 GB floor and 2.1 GB early-stop threshold. No runtime signals were needed. Independent final release at 19:25:02 verified both sessions' 38 recorded PID/start identities and all own helpers absent, the global Chrome/crashpad/ profile set empty, all four HTTP/debug ports closed, both profiles removed, and the lock free. Boat ID/state/expiry: N/A; runtime was correctness-local.
No outgoing hit, kill, objective completion, audio quality, FPS, touch or
sustained-combat result is established. No emulator failure was encountered,
so there is no defined repair boundary or oldFAIL/newPASS claim. Next concrete
boundary: navigate original sr1 to a native interactable/NPC and verify an
activate/dialog response or targeted melee hit; listening and performance
need separate qualification. Original demo files and captures remain private;
no assets or runtime evidence were committed or deployed.