TR2 first-shot consumer repair — 2026-11-08
Completed the assignment's permitted substantive causal tested repair before
allocation. No box, UUID, browser, original game or gameplay input was created.
The first linked original shot return remains unqualified. Evidence R:
scratch/runs/20261108-tr2-first-linked-shot-2019/.
The actual 1940 consumer passed only
weapon to observedHold; even a command requesting untilShot:true silently
selected the default first-weapon-return behavior. The new exact-consumer test
reproduces that omission. The owned consumer now requires explicit
actionObserver:true, weaponObserver:true, untilShot:true for its sole requested
1900ms Control hold, forwards weapon:true, untilShot:true, and refuses missing,
false, numeric or string options before dispatch. Movement/orientation and combo
commands are refused. A consumed observation attempt, including UNKNOWN,
prevents any later input; capture and quit remain available. Tests observe the
actual dispatch block, rather than just testing a duplicate options object.
The inherited browser also used raw JSONL appends, raw Chrome stdout/stderr, and a close path that could acknowledge cleanup without a terminal wait. The owned adaptation directly imports the published gzip logger, readiness and close helpers. Chrome uses pipes; every stored log append is a complete gzip member. Cleanup joins actual Chrome terminal/stdio close and compressed-stream EOF before profile removal or quit ACK. Failed drain retains UNKNOWN and the profile. Concurrent close callers share one completion promise. Startup retains its 30-second absolute subcap and reserves cleanup inside the immutable run cap.
Selected published provider hashes, independently checked against Git:
| Provider | SHA-256 |
|---|---|
| Observer, f1e9c852a | d018d80647104c21538529d421b7e4ce24afd09c61be68470084bf5f90fe4b01 |
| Kernel, 252888061 | e36d34990d738e616611ce8a8f4309548abcb751bd5ef5ec119d20e561c49360 |
| Command logger, 252888061 | 979b3a8bbd8625ef6bffb009650778736e20ade2d781918b83f59d9944c88f5e |
| Readiness | 9ba9ae8b94ebd8a134045608074403d81f68d3cfdf6c0e3f9dbcb875e0632c1b |
| Close | 1f4fe08b501ed27c90f4256e85574a788e7dd8adff60638f2c69d6a3ca2ea9b2 |
The newer published kernel is explicitly selected because d8573b542's
5fb6488e… lacks command-output gzip. No peer kernel/test/logger source was
edited. The launch checks selected provider hashes before importing the kernel
or forking. Positive selection passes; the old kernel hash is refused.
Both kernel journal and command compression are explicitly gzip. The adapted
exhaustive inventory/retrieval has no extension filter and checks transport and
decoded-original hashes, including the kernel command log.
Validation: 27 Node checks (nine own consumer/selection/readiness checks plus
18 shared helper checks), observer57, weapon150, three inventory/retrieval
checks, and flock /tmp/wine-build.lock bash tools/build.sh pass. Two actual
launch controls execute /bin/true and 262,173 bytes of binary stdout/stderr;
each has a prearmed independent observable seven-second PID/start watchdog.
Both retain complete kernel drain, actual supervisor/watchdog terminal waits0,
empty remaining/unreaped sets and an empty outer child set. Neither watchdog
fires. Gzip/EOF/original SHA and byte counts are independently verified; no raw
command log exists. These are local process controls, not Chrome or gameplay.
CLONE_UNTRACED and nonleader-exec limits remain; absence is not substituted for
an actual terminal wait. The locked build retains CPU
d8facd0658b7f158c6e60bef2de0936471f438f8bf41e9240fd75635f1d319dc.
R retains exact Git SHA, empty relevant tracked diff, small owned-input patches and predecessor path/hash references. Media/modules are referenced by existing path/hash. New test/build/browser-substitute/kernel logs are gzip from their first stored bytes. No source-tree archive, media copy, old-run deletion, peer-WIP edit, local Chrome, guest write/private call, configuration/backend/ binding/clock/seed change, public deployment or subworker occurred. Sampled home free space stayed above 1.5GB; no continuous-floor or RAM-gate claim is made.
This is a tested small consumer repair, not a completed remote provider closure
or staging/lifecycle integration. A successor must finish that selection and
prearm the independent immutable 25-minute box watchdog before one idempotent
create, then qualify current graphics, owning WebGL2/module/HTTP authority and
review upright/full-health/drawn-pistol pixels before the sole Control hold.
The unchanged observer bounds remain2500ms/32 records with actual-retired and
caller-remainder guards. Stop at the first linked shot return, inspect its
poll/Action/full-input/caller-ESP/weapon/shot links and raw result/ammo/target/
health/readiness before any further action. pendingWeapon:true is expected
there; pendingShot:false is required. The five-cycle type2/zero-ammo fixture
return0 is not original production proof. The 1940 frame0→1 result remains
early readiness, not ineffective firing. No shot, aim, hit, kill or effective
firing claim follows from this repair.
Box NONE / NOT_CREATED, expiry NULL, stop receipt N/A. No old box was resumed/deleted. Intel versus SystemInfo SwiftShader remains unresolved; Full27/parity/audio/FPS remain open. Release owned paths and STOP.