Aranna menu-phase observation, 2026-11-06
Run: scratch/runs/20261106-aranna-menu-phase-fresh/ (ignored). A run-local observer preserved fresh native menu coordinates within the existing bounds. Both shipped module variants passed generated controls, but the one changed original still lacked a fresh completed Single Player target hit. The click guard refused; balanced Enter left Main Menu unchanged. No setup, mission, gameplay or emulator mismatch was established. Production was untouched.
The predecessor is the pointer-lock lifecycle investigation. Its original retained lock for255.967 seconds and highlighted Single Player, but startup exhausted the shared32 admissions. This task did not repeat the historical timeout investigation or launch an unchanged original.
Observation and qualification
Preparation reused the predecessor2002's22 unexecuted drafts read-only, adapting copies into the owned run. Execution began only after the explicit coordinator20:24:59.277121–21:09:59.277121 UTC grant, following Conan1938 FINAL4fb74d51e and coordinator cleanup verification. An initial additional cleanup check used the wrong predecessor filename; its failure and correction were recorded before any module compilation or browser test. The corrected check verified111 absent PIDs, four refused ports and four absent profiles. No workers or boxes were created.
The observer adds a one-way observer-only arm export. Before arming, kind24 hit/store admissions are withheld; prior records and counters remain intact. After arming, changed coordinate stores can consume only the first eight shared slots, and hit admission is restricted to the native rectangle280,132–517,178. The combined ceiling remains32 and the global ceiling8192. Existing transaction end paths remain active. Arming writes only observer globals and makes no guest API call, guest memory/register write or execution-control change. Production instruction behavior and import contracts remain unchanged; the sole export-contract addition is aranna_menu_observer_arm.
Fresh same-closure baseline and observer modules were compiled for standard and compat. Both baselines exactly matched shipped bytes. Compilation source was8b1a791edb464b1e31e6c514626005b3a2f32306, with closure hash5eec7954f91f1486ced51b92fd895d6c5d4efb33a5225fa3f2e8950c55bd652e. Standard baseline280014368b95a768bd31397971715f5d95089c3a22d018758b5d58f3d4a1650e and observera2e43b1336b58b3fd32da7771b86ce3f39c5a7e23d7baff1b473b8f989bf16a5 were pinned. Compat baseline7d476f1d52fdc4538de94a436df69164eb02df26285c3a50a2b0516ec4996eca and observer084504b5027abe1a2b3c7c67a5ad6b4e90f239a5ec2cd3c9812992f4f5b8dd07 were pinned.
All eight browser passivity cells passed: baseline/observer × standard/compat × ordinary API/owning Worker. Generated startup executes512 iterations of coordinate churn. Each observed cell retained121 earlier records, spent zero shared slots before arming, and later saturated exactly32 admissions with five stores and27 hits. CPU registers, lazy flags, x87 state/payload, public API results, guest results, immediate/buffered mouse behavior and balanced keyboard/button actions matched. Actual native target packets, complete transaction ends and negative endpoint/return guards passed. Results were losslessly compressed from194257554 to4669506 bytes, with verified decompressed SHA-256 and a restoration recipe. A full-shell launch refused a transient disk-floor failure before launching; headroom was restored before proceeding.
The first native compat qualification rejected an observation-order race: the arm export acknowledged at20:31:26.384, trusted movement occurred at.388, and the asynchronous arm packet arrived at.390. The retained failure was corrected in the driver by awaiting the actual marker before permitting movement. Both actual native full-shell controls were repeated with that driver before the original. Standard visibly activated target A at native396,155 using three shared admissions; compat activated it at397,155 using ten. Their owning CPU target/session/module identities were recorded. All423 standard and311 compat transactions completed. Eight negative guards per variant rejected wrong owner, wrong module, stale phase, wrong native point, wrong target, missing native end, movement after review and saturated quota. The repeated standard control required ordinary trusted reacquisition after an observed automatic expiry; no lock guard was disabled. Independent control cleanup passed64 checks before the original, with2742521856 bytes free.
One original and the remaining boundary
The original used the actual instantiated standard observer above, owning slot0/endpoint1, target62347F44DAE62288DE290169C390E723 and session28C016F03C20B44B475D1331E03AABBB. Ordinary EULA Accept and natural intro reached Main Menu. A reviewed trusted click acquired CANVAS lock. Arming at20:43:43.383 followed the retained marker2529 and preserved all32 shared slots; a guarded relative move0,-245 then visibly highlighted Single Player and its campaign hint.
Native store packets2530 and2533 retained manager points399,400 and399,155. Only two shared admissions were used. However, there were zero kind24 hit starts and zero event84 target packets in this armed phase. The click proof therefore refused with fresh native manager point and completed native target identity required. No Single Player click was issued. An ordinary balanced Enter DOWN/UP left the reviewed Main Menu unchanged. The complete retained capture contains228 admitted transactions with228 ends and zero incomplete transactions; its maximum global observation count is2904. The earlier201/2535 checkpoint preceded Enter and cleanup.
This establishes that late arming preserves coordinate evidence and solves the startup quota exhaustion. It does not establish a completed target hit before a proposed click. The generated controls exercised their target predicate during a native click; they did not prove that the original's hover/movement path would execute the same observed predicate before a click. A successor needs a generated pre-gesture point-and-loaded-target proof that succeeds before its first target action, then an appropriately timed bounded observation window. Another unchanged original or relaxed click guard is unsupported. No production repair is justified by this run.
Terminal state and recovery
Queued quit7 completed at20:45:16.171. The driver log records PAGE CLOSED +404.7s; Chrome recorded exit0 at20:45:18.735. The enclosing runner tool returned143 and wrote no terminal.json; its exact exit time/cause and the driver's exit code are unknown. The holder and board-watcher tools also returned143 when awaited. All are now absent, but continuous flock ownership is not claimed. Post-stop app absence proves cleanup only. No original retry occurred.
Final cleanup verifies60 recorded PIDs absent,12 refused ports, six absent profiles and four free locks, with2723319808 bytes free, before the grant deadline. No boxes were assigned: ID/state/expiry null, ownership none. Production was not edited, so no new production focused-test or locked-build claim is made. The generated module compilation/contracts, eight-cell passivity, repeated both-module native controls and retained-packet review are the tests performed here.
Git is the source archive. The run retains exact commit identities, relevant binary Git diffs, owned new-file patches, input hashes, observer module patches/bytes, reviewed captures, assessments and cleanup evidence. Existing game media is referenced in place rather than copied. The compressed control results have a verified restoration recipe. Assessment/report artifacts precede result.json, which is published last.