Aranna stable menu input, 2026-11-06
The run-local adapter passed both shipped-module full-shell controls for locked relative input, naturally expired stale-mode refusal, recoverable queues, and stable unlocked absolute movement/click through presentation mapping. The sole subsequent original advanced through Main Menu → Single Player → Start New Game → Choose Hero. Choose Hero is setup, not verified player-controlled gameplay. An ordinary setup move refused before any input because the input-mode generation changed between review and dispatch. No emulator defect or production repair is established.
Evidence is scratch/runs/20261106-aranna-stable-menu-input/ (ignored). The implementation, generated inputs, raw captures, action receipts, module identities, patches and cleanup records are there. This commit changes this note only. Production expiry semantics, observer quotas and native provenance guards remain unchanged.
Admission and identity
Read CLAUDE.md, the Aranna TODOS entry and the pre-gesture predecessor. Published b919d2d14 already proved pre-action hover in both modules. Its original stopped before mouse-down after automatic lock expiry; that was not an emulator fault.
This worker stayed queued behind graphics881019 and Lords2892904. The limited 22:30:08 preparation grant admitted only tiny pure-JS controls. Main runtime began at 22:41:50.992225Z, after Lords explicitly released and its recorded processes, ports and locks were independently checked. The fixed deadline was 23:26:50.992225Z including cleanup. Coordinator 23:00:34 explicitly authorized one registered original conditional on completed generated qualification; admission occurred at 23:04:14 after both qualifications passed. The immutable original-attempt.json guard admits no second original.
Compiled baseline standard SHA-256 is 6cd858302895de91df5af17e3a3742281e31f369ddb6528d7b96252b3423caac, compat 41b0c704c91ba6fc7e4a6645d9421bf7fd3a23139b0b538b576e1e1b85527dd6. Both exactly match the shipped modules. Observer standard is 3d68547dec8af48b7f0a09dc2d855542ce7e1019bdfa9ca35bec2c523f8349e8, compat d7c61f0179bb0500ce82e0947b88fbe28551fa2e29b288f1db66dfa6b233b90e. The unchanged predecessor transform adds passive instrumentation and the one-way observer-arm export; it does not write guest state. Actual CDP owning-Worker bytecode hashes match the selected modules in each route.
Source versions record exact Git commits, binary diffs and new helper patches. The original launched from 9fa90b9394e80636ec5f6a0c2702bedcb0a99e50, with the landed Lords a16845982 repair and only subsequent documentation updates in main. Original HTTP identity passed 572 shell/source/helper/module/media checks. Existing original media paths and 28 SHA-256 references are retained without media copies. Git remains the source archive; no source tree or repeated media archive was created.
Adapter and generated qualification
Every ordinary move/click receives the same image/review hash, exact Worker target/session/endpoint/module authentication as nativeclick. Tickets bind actual mode and mode generation, trusted movement generation, focus/visibility and presentation geometry. Target clicks additionally require the unchanged complete native proof: current manager point, receiver, manager/node, actual loaded target and vtable slot, balanced native transaction end and serial ordering. The shared32/global8192 guards remain enforced at dispatch. Ordinary absolute click performs no implicit movement; its browser position must match the separately reviewed mapped point. A narrowly qualified initial ordinary browser click can acquire automatic lock only before any prior lock, never reacquire it.
The queue persists an exclusive action receipt before dispatch. Only the typed refusal with a prepared receipt and zero attempted effects becomes rejected-before-input; the next distinct reviewed command may proceed. Any attempted effect with uncertain completion is terminal UNKNOWN and is never replayed. The actual queue controls cover rejection followed by a new completed command, unknown consumer termination without a done receipt, restart refusal and unwrapped input rejection. Pure contract controls total24; four actual-adapter mocked transport scenarios include unknown partial down/no replay. Those tiny controls are explicitly separate from loaded-Worker qualification.
full-shell-qualification.json covers these normally closed routes:
| Module | Locked relative control | Expiry and stable absolute control |
|---|---|---|
| standard | control-relative-standard3 |
control-expiry-standard2 |
| compat | control-relative-compat |
control-expiry-compat2 |
Each locked route proves native 398,155 before the first armed target action. Each expiry route retains that reviewed locked proof, sends ordinary Space to the generated guest to cease recentring, and observes actual production releaseExpiredAutomaticLock plus null DOM lock. A deliberately stale generated target command then refuses with effectsAttempted=0. The queue remains usable. A fresh reviewed absolute movement uses existing presentation mapping; browser scaling measures actual native 398,154 from the requested398,155. A new proof at that measured point precedes an ordinary click with no implicit movement. Every permitted target action produces personally inspected ACTIVATED TARGET A, hit=1, balanced down=up=2, and the measured native down point. There is no lock request after natural expiry.
Each of the four routes rejects28 mutations, 112 total, covering wrong owner/module, stale proof/point/mode, mode away-and-back, intervening movement, changed current native target, missing/reordered transaction end, wrong return PC/SP and saturated shared/global quotas. The actual original later also exercises a zero-effect stale-generation refusal.
The eight-cell controls-both-hover-sites2 matrix compares disabled/observed standard/compat modules through ordinary API and owning Worker routes. GPRs, lazy EFLAGS, x87 live payload/environment, API/COM results, SEH reserve/commit/resume and immediate/buffered native input remain equivalent. Both hover return sites are exercised, arming retains earlier evidence and all32 shared slots, generated saturation reaches the unchanged combined32, and packets preserve ordered ends. validation-both-hover-sites2.json and passivity-qualification.json pass. All qualifying runners/Chrome exited0; their recorded processes are absent, ports refuse connections and profiles are removed.
Sole original and concrete boundary
Ordinary reviewed EULA acceptance is followed by natural intros and Main Menu. With no lock yet acquired, Main Menu still continually recentres, so a stable absolute action is not admitted. One qualified ordinary neutral center click acquires the host automatic lock and leaves Main Menu visible; it is not a forced or repeated lock request. Observer arm serial2667 starts with zero shared admissions. A one-count relative move measures native398,399; the subsequent -244 Y movement gives398,155. No target mouse-down precedes the strict proof.
original/pre-action-proof.json authenticates actual loaded target006f34f0, current398,155 and five shared admissions. Command16 uses the newly qualified dispatch contract and completes with current live relative screen lock. Trusted down/up and actual native positive hit/return packets are retained. The immediate response16 still shows Main Menu; response17 visibly shows the distinct Single Player screen. It was reviewed before any setup navigation. Reviewed movement18 highlights Start New Game; reviewed click19 completes. Response20 visibly shows Choose Hero, a default character named Hero and Next/Previous controls. These are setup screens, not gameplay.
The host naturally releases its lock after the setup click. Choose Hero then alternates between recentring and briefly stable unlocked input. Reviewed response21 has absolute/null lock, stableUnlocked=false, modeEpoch8. Command22 is one ordinary mapped move toward Next. Its final live snapshot has absolute/null lock but modeEpoch9 and stableUnlocked=true; the unchanged generation guard refuses stale actual input mode before all input, effectsAttempted0. Response22 again has stableUnlocked=false, active recentring and epoch10. The queue acknowledges this proven unstarted refusal. No movement/click retry or alternative input command follows; quit closes normally with driver0 and Chrome0 at23:13:55.
All 262 original native transactions complete; zero incomplete. This establishes menu and setup progress beyond b919d2d14, but neither a gameplay scene nor controlled character movement. The concrete remaining boundary is fresh stable-mode admission across the original's unlocked/recentring oscillation. No production mismatch is established, and no repair, guard relaxation, forced transition, guest input/state write or lock reacquisition was attempted. A future original needs separate authorization and generated qualification of that additional transition boundary first.
Corrections and recovery
Early generated preparation included an over-strict initial-input refusal and an unused timer fixture; retained corrections distinguish those from qualifying Space-controlled runs. A manual compat proof using the standard-default selector correctly refused before input; explicit compat selection passed. Node's ambiguous assert.throws message overload initially stopped negative aggregation; using the explicit message argument preserves every expected rejection.
Two earlier foreground holder/runner lifetimes ended143 or disappeared; causes are unknown. The standard2 target activation is retained but its normal driver exit is UNKNOWN, so it is not used for final qualification. Its owned Chrome orphan was detected, terminated and profile removed; unknown commands were not replayed. The subsequent detached start_new_session runners use file-backed stdio and all qualifying controls close normally. Holder990677 was independently verified live with actual contention on all three flocks. Inaccurate early preview descriptions have adjacent corrections; qualifying activation captures were personally inspected. Black previews are not used to assert absent target pixels or a guest fault.
No boxes were assigned or created: ID/state/expiry are null. Final cleanup verifies all owned processes, ports, profiles and locks; disk stays above2GB. The note is published with explicit-path commit and immediate main pull/push. Final result is written only after evidence, cleanup and publication, followed by FINAL release. Aranna has no exact corpus manifest identity, so candidateId remains null as an investigation record, with the known dashboard schema limitation stated explicitly.