TR2 lower-alley route reviewed — 2026-11-08

One fresh original produced a personally reviewed new lower-alley route traversal away from the original brick slope. Six separately released right turns revealed a flat cobbled lane, with shutters on the left, brickwork on the right and a dark opening ahead. Four separately reviewed forward holds advanced Lara along that lane. The settled final image preserves the changed position and view. Crossing the far opening, an airborne jump and a landing are not established.

The requested evidence is incomplete: I omitted the ordinary observe command that arms publication observation. Every new publication JSON, including publication-final.json, is null. There are no new raw retained/discarded counts to report. The first final-audit failure is retained in initial-audit-failure.json; the scoped audit explicitly records completeRequestedEvidence=false. The box was already normally stopped when the omission was found; no replay, second allocation or old-box resume was attempted.

Evidence is scratch/runs/20261108-tr2-route-ready/ (R). Compare personally reviewed review/webgl/native-start.png, route-right-8.png, route-downhill-current.png, route-forward-4.png and route-final-settled.png. The last PNG SHA-256 is 9207294326c4b514ad606cbaefe9cb78562e43ff0edb516a35f47114fc667e36. This is visible route travel, not a claim to have reached another plaza or crossed a threshold.

Closure, inputs and authority

Starting Git SHA: 104f1ea79ae504971532303ea45888f3a1127210. The production-only driver from d781e6338 was relocated as 29 physical run helpers, preserving the 614-provider closure. The current checker, including nested target preservation from 34c92d72c, passed before allocation. Its SHA-256 is 43f27bb473bf7f12ad11143ea3232fde111a2e063cdae31082313eedde1f2d40. Shared 16-key admission/execution and the actual physical tools/native-key-controls.cjs dependency remained bound. The fixture environment was empty, with no generated-fixture imports or selectors.

All 615 uploads joined before remote content verification published the barrier. Publication, queue admission and parent consumption receipts agree on manifest SHA-256 9c14456d199855d1257e62c65a14d8917f711e34f3c52fb377ddbc374135be2e: 614 providers / 57,915,707 bytes. The parent held the remote runtime lock and subreaper custody before provisioning through descendant drain. Only remote runtime ran; the local wine-runtime.lock was never held.

The current Graphics dialog was personally reviewed before ordinary Enter. After reviewing the steep slope, I turned to inspect an alternative route rather than advancing into it. The newly revealed lane supplied the rationale for forward movement. Each successful forward hold followed a reviewed changed view; no unchanged no-progress Up or Alt sequence was repeated. There was no Alt, firing, compound input, private input handler, guest write, forced scene/configuration, clock or seed change.

Twenty commands ACKed, including three stale-scene refusals before dispatch with zero effects. Each refusal was followed by a new capture and review. Eleven actual holds comprise Enter once, Right six times and Up four times. Their independent release receipts are retained; the 2-second requests released within 1999–2001 ms. All 22 native DOM events are trusted and balanced, and the final read-only check reports all 16 admitted keys released. There was no runtime UNKNOWN. Three premature local image reads occurred while their original asynchronous command downloads were still running; after waiting, the original files were inspected without replay. Those diagnostics are retained.

All five actual parsed JavaScript providers match their pins. The actual CPU and owning renderer module remain d8facd0658b7f158c6e60bef2de0936471f438f8bf41e9240fd75635f1d319dc, with owning native WebGL2 on ANGLE Intel UHD Graphics 620. The accepted unchanged native pixel-authority evidence was reused without another pixel logpoint. This run does not extend its publication qualification. Historical raw accounting remains exactly 670 = 53 qualified retained + 617 discarded, with raw disposition labels coalesced-before-publication (602) and run-normal-shutdown (15). Those counts belong to the predecessor, not this new runtime. No FPS is inferred.

Validation and closed lifecycle

python3 R/verify-prelaunch.py passed exact provider, syntax, helper/read/upload, shared-key and preallocation checks, independently rereading 226 accepted references, 23 media references and 49 repair controls. Accepted build/key/barrier/pixel controls were reused without rerunning unchanged fixtures. The published WNOWAIT reaper and its tested custodian integration were reused: exact PID/start/PPID registration is durable before consuming child exit. No production code changed, so no new full build was required.

python3 R/audit-route.py passes its explicitly scoped audit: actual barrier generation, parsed sources/modules, native events/releases, 17 PNG CRC checks, retrieval hashes, remote drain, local lifecycle and preserved historical dispositions. Its first publication-null failure and a later historical-disposition label assertion failure are retained. The corrected audit uses the actual raw disposition labels and reports the missing new journal explicitly.

Before requesting normal box stop, 269 artifact references / 229 unique hashes / 127,797,075 bytes were retrieved and independently reread. Drain passed for all 28 recorded remote PID/start identities, including custodian, observed/adopted descendants and bootstrap; controller and Chrome exited, both ports closed, profile and writer disappeared, held inputs released, all 614 provider hashes remained unchanged and the remote lock was free. The controller was passively observed exiting before the fresh drain, avoiding the predecessor's early-drain race.

Custody scope is a 50 ms descendant census plus subreaper adoption and durable WNOWAIT registration/reaping. It does not establish exhaustive kernel fork history or identities of intermediates already reaped by their own parents. Remote diagnostic subprocesses were synchronous and parent-awaited. Initial local lifecycle start ticks were not recorded; all three exact supervisor/normal/watchdog PIDs are absent, both supervised children exited zero and the supervisor ended empty. No global kernel/FD-history claim is made.

The detached, observable independent 25-minute watchdog was armed before allocation, with immutable browser, cleanup and overall deadlines. Box bx_nyh7e5hx was created once with a fresh UUID idempotency key; initial expiry 2026-11-08 15:53:46.305 UTC was recorded before runtime. It is now archived, expiry null. Normal stop stop_be8e5e24fb8970af completed 15:34:02.538 UTC, forced=false, error=null. No task box remains billing.

Recovery uses exact Git SHA, empty relevant binary diff, small owned helper patches, individual logs/captures/modules and existing media path/hash references. In particular, test/binaries/candidates/tomb-raider-2-demo/game/DATA/BOAT.TR2 remains at its existing path: 5,920,318 bytes, SHA-256 e710716fcaf039fb3462e26e33f0ead8eb45bd01c8bbfa036c455a1d19823d9e. No source/media archive, peer WIP, public deployment, old-box resume/delete or subworker was used.

Stop here after the reviewed route consequence and closed lifecycle. A separately assigned continuation must arm publication observation before gameplay, then seek a far-opening threshold or jump/landing consequence. Full27, parity, audio and FPS remain open.