Forsaken early look/navigation attempt, 8 November 2026

The fresh original stopped at its first UNKNOWN, before a player HUD. Current look and navigation beyond the Volcano corridor remain unverified. No movement or fire command was sent; accepted movement and PULS 2000→1998 evidence from the preceding run is unchanged. No emulator mismatch was proved and no production implementation was changed.

Evidence: scratch/runs/20261108-forsaken-early-look-navigation/.

Local qualification and reused evidence

Read the current-native-progress handoff and original installed readme.txt §VI: A/Z forward/reverse, arrows rotate, Alt slides, Ctrl/Space fire. The installed Volcano text describes destroying enemies to open a hidden door; reading that text does not establish a visible objective in this attempt.

The preceding qualified combined-before observation is reused unchanged: old composition 21.879 s UNKNOWN, combined 17.030 s exact ACK, with controlled 4,800 ms guest-service receipt delay. These remain controlled helper observations, not a cause proved for the original game. Separate live guest/module, renderer owner/context/module and page-reference checks before and after every image, independent admission, two complete stable readiness reads, 120-second freshness, exact ACK and sticky UNKNOWN remain in place. No saved identity is reused to admit a later capture or input.

local-look-worker/look-route-qualified.json qualifies an actual production guest/render Worker pair and native CDP right-arrow delivery using the production renderer input mixin and production RPC broker. The generated-only fixture receives trusted DOWN/UP; page physical state is 32768→0 and the actual guest Worker host input import reads 32768 seven times during the hold, then 0 after release. Release occurs after 744 ms for a 750 ms request. The numbered client returns exact success ID and number; the saved reply has exact ok/id. This is input-route qualification, not original-game rotation. The fixture's initial verifier omitted the client's number field; that assertion was corrected before the passing final fixture, without changing transport, input, production Workers or deadlines.

Independent release tests with a slow sampler and injected UNKNOWN pass. Production Worker focus tests, 14 production DirectInput keyboard checks and all eight existing capture-deadline tests pass. An unrelated existing test-worker-api-batching.js:518 exact statistics-object assertion fails because its expected object omits current submission-reason counters. This failure is retained in known-test-limitation.json; it is not a look defect or a passing suite claim. An initially mistyped capture-test filename was corrected before the eight passing tests. All three local fixture browsers exited zero and their PIDs, profiles and ports drained. No new locked build is claimed for this documentation-only change.

Personally reviewed original and exact limit

One browser on one fresh box used index.html?candidates&no-log&threads&app=forsaken_demo&d3d-renderer=webgl. Chrome is 151.0.7922.71. Actual guest and renderer module SHA-256 is 7ec49cbbae4cb177909fd8fd9d3599713ee5f2a5ffb7f7e8c890848a540052bf. The owning endpoint is legacy WebGL, with a 640×480 WebGL 2.0 context and reported ANGLE / Intel UHD Graphics 620 / Mesa 23.2.1. Physical GPU is unknown. All six completed bound checkpoints retain zero renderer errors, fallbacks, inline software draws and fallback software draws.

Images in review/, also retained in retrieved/original/, show:

Number 5, ID open-singleplayer, requests a bounded 2,000 ms Enter plus 1,000 ms settle from the reviewed main menu. Its independent release receipt is complete: DOWN interval begins 06:40:34.132 UTC, release begins 06:40:36.132 UTC. The hold's passive sampler drains at 06:40:36.512. There are five exact success ACKs, six independent-release receipts and six completed bound captures. Number 5 has no reply or completed image.

At 06:40:46.610 UTC, the unchanged work deadline fires: “command work deadline; never resend.” Admission's owning guest identity evaluation took 8.085 s (full admission including hold 10.870 s). After settle, the capture-before guest identity evaluation was still pending after 9.092 s when the deadline fired. Browser closure rejects that already-started read at 10.989 s; it is not a completed service-response measurement. The command span drains after 22.862 s, without ACK. No screenshot/after proof for this command exists, and no menu-selection result can be inferred from key delivery.

Input stopped at this first UNKNOWN. No replay, later capture command, look, thrust, fire, private game handler, guest patch, config/scene/clock/seed change or second original was used. WORK=20 s plus declared settle, CDP=20 s, 50-second client ACK minimum and 30-second cleanup reserve were not increased. The fixed browser cap was 06:47:25.162 UTC, well after this failure; this was a command-work limit, not exhaustion of the browser navigation window. Next investigate the original owning-guest observation latency at this native menu boundary; then obtain early reviewed look and distinct doorway/junction navigation. Full 27, parity, audio, unmeasured FPS and enemy interaction remain open. No enemy hit or objective completion is claimed.

Recovery, supervision and stop

Staged source identity is main 585d24b1b804f168b1c1bf7fb973142f2711bd8d plus relevant-source.patch, new-lib_wininit-boot.js.patch and owned helper patches. Actual loaded module bytes are retained separately. Git is the source archive; no source/media archive was made. All 1,488 staged pins, 873 original media references and 640 successful served responses match their recorded bytes. EXE and manifest hashes remain 366ec5d0943daca839836296f9fb52724e94dc155630c6b0c30d7181106b5e24 and 587ce145ab73a3fed2c7136425bb6a7bcf28def312d430c8bf7d1be802b2260b.

The initial observable 25-minute watchdog was armed before UUID-idempotent allocation. Its nohup supervisor/children disappeared without terminal records shortly after allocation; continuous initial supervision is not claimed. This was a local lifecycle defect in the task setup. Initial records are preserved in initial-supervisor/. On discovery at 06:38 UTC, supervision was relaunched in an independent session against the same original deadline, without changing any browser or command deadline. PID/PPID/session checks and fresh heartbeats confirmed the replacement remained live through normal stop. Future workers must detach supervision with start_new_session=True before allocation, rather than rely on a background nohup shell job.

An early evidence read raced startup completion and returned an explicit missing file error, with no runtime action. A later completed inventory/read recovered the startup evidence. All 103 remote references / 96 unique contents, 10,223,468 reference bytes, were retrieved and independently size/SHA-256 verified before requesting stop. PNG CRC/row, capture identity, ACK/release and media checks pass. Chrome exited 0; controller/browser PIDs, all processes naming the owned runtime, HTTP/CDP ports, profile and writer lock are absent. Local free disk stayed above 5 GB; no RAM gate was used.

Box bx_r6m9729w is archived, current expiry null, original expiry 2026-11-08T07:01:11.441Z. Normal stop stop_7a8025fc73623406 completed 2026-11-08T06:41:28.146Z, forced:false, error null. Both lifecycle delegates exited zero and the supervisor has no children. No box was resumed or deleted, and no deployment occurred. This single-task worker releases ownership and stops.