TR2 first exact linked original shot return — 2026-11-08
The one fresh remote original qualified its first exact linked shot RETURN.
The sole dispatched Control hold requested and measured 1900ms, independently
released. The unchanged observer retained26 records: four completed weapon
raising cycles, then a linked type1 shot entry at 42f270 and return at
42d59a. Raw EAX was ffffffff (−1), target0, pistol ammo1000→999 and Lara
HP1000 throughout the recorded state. This establishes an original attempted
shot through its exact return. It does not qualify effective aim, effective
firing, a hit, damage or a kill. No second hold or further gameplay followed.
Evidence R: scratch/runs/20261108-tr2-repaired-startup-linked-shot-2116/.
python3 R/audit.py independently checks the recovered streams, exact links,
actual custody waits, lifecycle and source recovery. final-audit.json records
the result; first-shot-qualified.json records inspection before normal Quit.
Selection, staging and actual startup
The 2027 attempt failed before DevTools on the old kernel; no late attachment or historical retry was performed here. The owner separately authorized this fresh original after the actual Q2 startup and normal close. That single Q2 success is not a universal startup reliability claim.
The regenerated closure has607 exact providers,57,948,302 bytes and17 physical
helpers. It preserves the accepted2019 browser-run.cjs, shot-options.cjs,
launch.py, inventory.py and retrieve.py byte-for-byte. The selection
replaces kernel e36d3499… with published Git
bcf8c7bd788f2c63668406b11f72652006cde8dc, SHA256
5371233b12d3a63e13a9768529af43e5a61b7ff1682b2d3c09e4db6a976e12d0.
The old hash fails the new exact selection. Journal and command compression
are both explicitly gzip. Observer d018d806…, logger 979b3a8b…, readiness
9ba9ae8b… and close 1f4fe08b… remain selected, with unchanged flags and caps.
Exact source providers resolve from their published Git blobs. Both CPU modules
resolve from existing1810 paths and hashes, avoiding peer WIP and module copies.
The actual driver's preflight verifies all607 providers and exits before Chrome.
All609 uploads, including manifest/lifecycle inputs, joined before the remote
HASH barrier; the outer launch requires that same barrier. Manifest SHA256:
02f8c65de4268128f6277e8324ac1940ad91aa01347cce6caf13e526c1db2542.
DevTools became ready in 11,971ms, within the unchanged30-second startup
cap. This original actually executed and drew. The reviewed Graphics sheet
shows Primary Display Driver, Hardware3D Acceleration, Full Screen and
640×480 High Color; one ordinary Enter100ms accepted its current settings.
Nine HTTP resolver pins returned200 and matching hashes. Owning guest and
renderer executed CPU SHA256
d8facd0658b7f158c6e60bef2de0936471f438f8bf41e9240fd75635f1d319dc.
The actual owning context is WebGL2,640×480, renderer target
A5869C39F1C8766571B3BF4D695FF772, endpoint1, legacy/WebGL.
The observed Intel UHD620/Mesa string versus SystemInfo SwiftShader/SwANGLE
remains unresolved; no physical GPU qualification follows.
Accepted2019 evidence89 references independently hashes correctly. Its27 Node, observer57, weapon150, inventory3, two actual launch controls and locked build are reused by hash without identical offline reruns. The repaired kernel's155 accepted references, including33 focused controls, eight-thread control and locked build, also independently hash correctly. No new production build or source change is claimed. New closure, actual driver preflight, remote HASH, original execution and recovered-evidence audit are this task's checks.
Current pixels and sole native hold
The publication observer armed generation1 for process:1000:endpoint:1.
The immediate arm capture had zero records; a later passive capture retained41
records, live=true and failure=null before Space or Control. One released
Space1500ms drew pistols. Current pistols-fresh.png,21:21:14.932Z, was personally
reviewed: upright stationary Lara, drawn pistols, full nonempty health bar and
no visible dog in the original alley. There was no forward, orientation, combo
or dog-facing input.
Two stale-scene requests were explicitly refused before dispatch, effects0: one Space request and the earlier Control request. Each was followed only by the required fresh capture/review before a newly admitted command. They did not deliver gameplay input or consume an observation attempt. No UNKNOWN was replayed. Exactly three key commands dispatched: Enter, Space and Control.
The sole Control command explicitly supplied actionObserver:true,
weaponObserver:true, untilShot:true, holdMs:1900. The accepted2019 dispatch
forwarded weapon:true, untilShot:true; the old1940 omission was not used.
The independent native release receipt spans 21:21:27.002–21:21:28.902Z,
1900ms, within2000ms. KeyUp completed successfully, and both passive bindings
around the after-release capture show Control/left/right Control and arrows0.
No further gameplay input was sent after this attempt.
| First linked final-cycle boundary | Original observation |
|---|---|
| 42078,447206 | Poll HRESULT0, DIK1d/9d128/128; VK17/162 down, VK163 clear; publication GEN40982. |
| 42251,44d3d4 | Action row21 links poll20; keyResult1, OR executed, EBX40. |
| 42386,41427f | Full input100040; publication ESP equals saved input caller ESP+6c. |
| 42946,42e640 | Same full mask, linked weapon entry, exact saved caller/stack accepted. |
| 43022,42f270 | Shot row24, type1, caller42d59a, target0, source7ed1a454; shot ESP is weapon-entry ESP−50. |
| 43291,42d59a | Shot row25 links shot24 and Action21; return ESP entry+4; EAXffffffff, ammo999; STOP/removal. |
Earlier cycles retain five rows each and arm frames0→1→2→3→4. Each cycle has
its own successful poll, Action, full input publication and saved weapon
caller/ESP link; generation progresses40973,40979,40980,40981,40982. Target and
locks remain0. State is gunStatus4, type/request/lastType1, water0,
waterDistance32512, target angles0/0, Lara animation103/bodyState2/yaw−32768.
The high mapped Lara pointer is 7ed1a454; observed health remains1000.
The report has43291 JS boundaries, four completed weapon cycles and one shot
return. It removes about199.6ms after installation, within the unchanged2500ms
deadline and32-record cap. Correlation qualifies five in-hold polls, five
Actions, four completed weapon returns and one shot return, with a physical
Control sample. At this boundary pendingWeapon:true is expected because the
outer weapon invocation has not returned; pendingShot:false is required and
present. Actual-retired debit and caller-remainder forwarding remain unchanged.
The26-row type2/zero-ammo authored fixture is not production evidence for this
type1 original. Historical1940 frame0→1 remains early readiness, not ineffective
firing. Raw nonzero return, ammo decrement and bright after-hold pixels do not
establish target damage. Shot receipt review preceded the next and final action,
ordinary Quit.
Drain, recovery and delegated stop
Normal Quit received exact ACK11. Chrome terminal was code0/signal-null, with closeRequest ACK, no signals and 12,217ms close duration inside its30-second cap. Both compressed pipes reached EOF before profile removal and cleanup ACK. The parent actually waits for supervisor6354/start8132, exit0 and stdio EOF. Kernel custody has350 owned PID/start identities and350 matching actual terminal reaps, complete=true, and empty remaining/unreaped/unresolved/cancellation sets. Independent audit matches each reap to its before-reap and terminal journal sequences. Command-stream EOF and original hashes match producer receipts. All13 browser log producer receipts independently match their originals.
Both ports39471/44491 are closed, writer.lock absent, runtime lock free, profile removed, and observable owned writable FDs empty. The supplemental global FD census retains972 unreadable entries across49 processes, including a disappearing FD; global visibility is limited. CLONE_UNTRACED and nonleader-exec limitations remain. Absence is not substituted for a terminal wait. Detached runtime and local lifecycle supervisor exits are not promoted to external parent waits.
All172 regular unique remote artifacts,20,453,193 original bytes, were recovered
without an extension filter through158 distinct gzip transports/3,848,290 bytes.
Every compressed and original SHA/size, nested gzip CRC and original-stream
receipt was independently reread, including gzip -dc. One executing module
uses its existing path/hash instead of another copy; selected helper sources
are omitted from unique transport only after exact hash checks. No symlinks or
special files remain. The inventory covers every regular file under the owned
run; selected sources and the module retain explicit reference authority.
A durable new UUID preceded the single idempotent create. Independent observable watchdog/normal roles and their supervisor were prearmed before create, with immutable25-minute deadline 21:42:38.344Z. Box ID and initial expiry 21:42:37.680Z preceded runtime. Recovery and independent HASH/CRC verification preceded delegated normal stop. Both lifecycle roles have actual parent waits0, and the supervisor reports no remaining children; the watchdog did not fire.
Box bx_d9smvggm is ARCHIVED, expiry NULL. Stop receipt stop_e2d77a0f73cb3f52 completed 21:23:37.687Z, forced=false, error=null; stoppedAt21:23:37.666Z. No old box was resumed, deleted or touched.
Recovery records exact source Git SHA, relevant diff, small owned-input patches
with reverse checks and the inherited boot-provider recovery patch. Preparation
and local verifier corrections are retained in local-preparation-failures.json;
the gzip reread fix used already downloaded bytes, without remote replay. One
used passive review JSON was losslessly compressed afterward, explicitly marked
as metadata compression; all unique logs were gzip from their first stored byte.
After recovery,48 used review-file duplicates were replaced with explicit
path/SHA references to their independently verified lossless transports;
review-recovery-refs.json retains reconstruction recipes for all of them.
Sampled /home free space stayed above1.5GB; no continuous-floor claim is made.
No source archive, media/module copy, peer-WIP capture/edit, local Chrome,
private guest operation, configuration/backend/binding/clock/seed change,
subworker, RAM gate or public deployment occurred.
The assigned original is finished. Effective targeting/hit/kill qualification would require a separately assigned fresh task; this result authorizes no new hold. Full27, parity, audio, FPS and renderer qualification remain open.