TR2 visible startup and one observed Control — 2026-11-08
One fresh original showed readable Graphics startup, reached drawn pistols and a dog ahead, and completed exactly one native Control hold: requested and recorded 2000 ms. The repaired observer witnessed an in-interval successful poll and paired Action successor. Reviewed after-release pixels show Lara upright, the dog close beside her, and health reduced to roughly three-fifths. No clear muzzle flash, effective aim, hit, kill or Control defect is established. Gameplay ended without another hold, restart, replay or hold extension.
Evidence is scratch/runs/20261108-tr2-startup-control-1810/ (R).
Exact starting checkout SHA is f0ccbf82d87758460eb23f261e0ae5a44da7606c;
relevant.diff is empty. Production code did not change. Provider identities
come from the explicit selected manifest, not the checkout HEAD or inherited
manifest's old commit field. Small owned helper/note patches and media
PATH/SHA references support recovery; no source archive or media copy was made.
Actual startup and fresh environment
Personally reviewed review/webgl/startup.png and startup-reviewed.png show
the original Graphics property sheet, Primary Display Driver, Hardware 3D
Acceleration selected, Full Screen 640×480 High Color, and readable OK. The
read-only inspection before Enter records main HWND 65537 visible at zOrder 2,
unowned dialog root 65538 visible at zOrder 10, and visible Graphics child
65545 under 65538 at zOrder 9. The original input list was empty. One ordinary
100 ms Enter accepted the reviewed settings. The normal balanced native-key
preflight occurred on about:blank before original launch, separately retained.
This run used the published 6de923cb8 shell guard, but its actual shell log
says single-app=false, viewport 945×889, screen 1920×1080, DPR 1, coarse
false. Thus this verifies visible original startup in this fresh environment;
it does not reproduce the previous 780×437 single-app failure or independently
attribute this run's visibility to the guard. The authored before/after fixture
remains the causal evidence for the guard's single-app behavior.
Fresh Browser.getVersion records Chrome 151.0.7922.108, V8 15.1.206.16, Linux x86_64. The box reports Ubuntu 24.04.4, kernel 6.8.0-117-generic, DISPLAY=:0, hardwareConcurrency 4 and deviceMemory 8. Full version, OS, SystemInfo, targets, windows, GPU information and custody command lines are retained in startup/post-Graphics authority JSON and remote custody records. No flags, viewport, backend, binding, input policy, guest clock or seed changed.
Before ARM/gameplay, the actual owning D3D Worker target
5900E17E5ECD096969C0E638EE62A7EA had endpoint 1, requested
{api:legacy,backend:webgl}, no endpoint failure, and a freshly observed
webgl2 context, 640×480. Actual CPU and owning renderer module hashes both
equal d8facd0658b7f158c6e60bef2de0936471f438f8bf41e9240fd75635f1d319dc.
The fresh context returned WebGL 2.0 and the ANGLE Intel UHD Graphics 620/Mesa
23.2.1 renderer string. This was observed in this run, not copied from 1717.
There is a material environment discrepancy: fresh Chrome SystemInfo reports ANGLE Vulkan SwiftShader/SwANGLE 5.0.0, and custody command lines contain appended SwiftShader flags. Both surfaces are preserved. Neither appended flags alone nor the owner-returned Intel string certifies physical Intel GPU use. Their disagreement is unresolved; this result qualifies the requested owning WebGL2 API/module and observed input path, not Intel hardware authority. No backend experiment or corrective restart was attempted.
Observation and reviewed consequence
Normal observe established ARM generation 1, owner
process:1000:endpoint:1, endpoint 1, with a non-null live initial raw journal
and no failure. Its initial one raw record and independent file hashes were
confirmed before the first gameplay command. Native Space drew pistols;
reviewed turns and forward holds reached approach-one.png: Lara upright,
full health, pistols visibly drawn, and a brown/black dog ahead in the alley.
Six stale scene admissions refused before dispatch with zero effects. Fresh
captures preceded subsequent new commands; no refused request was dispatched.
The sole actionObserver:true Control interval was
18:18:40.817–18:18:42.817 UTC. Independent native release remained
unchanged. Physical VK17 and VK162 were down wholly inside the interval;
VK163 was clear. The final passive sample records all seven sampled movement
and Control keys clear after release.
input-observed-control-once-action-observer.json records status:witnessed,
qualified:true, 22,368 boundaries, two retained records, no pending pair,
and removed:true. Activation was after native keydown. At 18:18:41.602 UTC,
poll-return PC 0x447206 has HRESULT 0, DIK1d/9d bytes [128,128], and the
owning bitmap reports VK17/162 down and VK163 clear. Binding slot 8 reads
configured 156/default 157/conflict 0. The paired Action successor at
0x44d3d4 records keyResult 1, EBX 0x40, Action set and OR executed.
Correlation confirms one poll and one Action wholly inside the native interval.
These observations establish input/Action, not weapon or aim behavior. Bitmap
reads explicitly report atomicSnapshot:false.
Personally reviewed held and after-release pictures show the dog at close range and Lara still upright. Health is approximately seven-tenths in the held picture and three-fifths after release, versus full before the hold. Small red pixels are visible near Lara/dog but do not establish bullet impact or identify a muzzle flash. The dog remains upright; no kill is established. There was no further gameplay command after the after-release capture, only ordinary quit. No private guest writes or substantive code-repair claim.
Selected closure, audit and shutdown
The explicitly selected input is
scratch/runs/20261108-tr2-startup-presentation-1750/runtime-pins.json, SHA-256
d47d09a5520ce6bc4d2321c90f9e4b45446f5bed182fd3fd3248912bcd4bd72d.
Normal relocation retained 615 providers/29 physical helpers, with all
586 non-helper providers unchanged, including observer 8579c29a2 and the
guarded shell. Historical 614 and predecessor 615 manifests remain unchanged.
The exact resolver and actual driver's preloader passed before allocation;
all 85 coverage references/446,299 bytes were independently rehashed unchanged.
The 32 observer tests, exact-d8fac causal fixture, direct/RPC, helper, HTTP and
locked full-build evidence were reused unchanged. Fixtures are not original
or backend authority. This docs-only task claims no new production build.
python3 R/audit.py passes retrieval reread, staging barrier, startup stack,
fresh owner/module/source pins, ARM-before-gameplay, exactly one Control,
physical and poll/Action correlation, release, PNG CRC and observed drain.
There are 35 commands/35 ACKs, six zero-effect stale refusals, zero
transport UNKNOWN records, and 30 valid PNGs. The generic cleanup receipt's
literal pending:"UNKNOWN; never replay" is retained; no actual transport
UNKNOWN or missing ACK exists. Raw accounting is **528 = 74 qualified retained
- 454 discarded** (451 coalesced, three shutdown), without journal failure. These are not FPS measurements. All five parsed host JS sources match pins.
One persisted UUID idempotency key preceded the single create. Observable
independent normal/watchdog/supervisor PID/start identities were recorded
before allocation. The immutable 25-minute overall deadline and initial box
expiry were approximately 18:34:58.6 UTC. Remote custody held its runtime
lock from startup through drain; no local heavy runtime or runtime lock was
used. All 616 uploads joined before verified staging publication and
parent consumption, manifest SHA
ec0109d0b50b69c4804638216b056916a8944e74aa447bc9a11b485ad90c32bb.
Before delegated normal stop, 421 artifact references/381 unique hashes/ 457,279,589 referenced bytes were retrieved and independently reread. Chrome exited 0, signal null, at 18:19:11.191 UTC. Drain verified 31 observed remote PID/start identities, adopted descendants, bootstrap/custodian, ports 36321/36653, profile, writers, held-input receipts and remote lock. All 615 provider hashes remained unchanged. Custody is a 50 ms census plus subreaper adoption and durable WNOWAIT registration, not exhaustive kernel history or evidence of already reaped descendants. Observer internal WAT reentry and other threads remain unobserved; missing records imply no absence of execution. All three local lifecycle identities are absent, child roles exited zero and supervisor ended empty. Free /home remained above 1.5 GB.
Box bx_9hdma9ab is archived, expiry null. Normal stop
stop_7250aad4c9efedb0 completed 18:20:09.146 UTC, forced=false,
error=null. No task box remains billing. No resume/delete/public deployment
or subworker occurred.
The assigned single hold and reviewed consequence are complete. Further weapon/aim investigation and the contradictory GPU reporting require a separate assignment; no additional original is attempted here. Full27, parity, audio and FPS remain open. This single-task worker stops.