Aranna setup input, 2026-11-06

Outcome: blocked in the input harness; player-controlled gameplay was not reached. Evidence and experimental helpers are in scratch/runs/20261106-aranna-setup-input/. No production files changed.

The predecessor stable menu input run reached reviewed Main Menu → Single Player → Start New Game → Choose Hero. Its subsequent setup move refused before dispatch because a sampled input generation changed. That is a harness boundary, not an emulator defect.

Route under qualification

lib/browser-input.js obtains lock in canvas.onmousedown through requestGuestPointerLock. The synchronous optionless request belongs to the trusted browser event. Automatic capture expires through releaseExpiredAutomaticLock when the guest no longer requests relative input. The acquisition down suppresses capture-transition motion until balanced release; subsequent pointer-locked movement uses guest counts. The adapter previously prohibited all reacquisition, although this ordinary browser route exists.

The run-local observer distinguishes actual protocol events (lock/focus/visibility) from fluctuations in the sampled recentring hint. Dispatch retains exact reviewed Worker/session/module, app identity, presentation geometry, trusted movement generation and live focus/visibility checks. Absolute pointer actions require stable unlocked input; relative pointer actions require current screen lock. Keyboard events do not require a stable mouse recentring hint. Reacquisition requires reviewed and live unlocked relative intent, with an ordinary trusted canvas click. Existing first-target native provenance and quota checks remain.

Focused controls and their limit

Both standard and compatibility generated full-shell controls passed ordinary Space expiry/resume → trusted canvas reacquisition → balanced release → native target movement and click. Personally reviewed control-reacquire-standard/response-07.png and control-reacquire-compat/response-06.png show ACTIVATED TARGET A, hit=1 and down/up=3/3. Acquisition itself had down/up=2/2 and hit=0. Observed requests came from requestGuestPointerLock with active browser user activation; expiry came from releaseExpiredAutomaticLock. No injected lock request was used.

route-qualification.json records these two passes and 18 negative checks on actual owner/module, protocol, native point/provenance and unchanged quotas. route-contract-controls.json records 32 pure contract checks for the initial qualified adapter version. Prior successful native controls, 112 negatives and eight-cell passive observer equivalence were reused. These results do not qualify every later experimental adapter revision.

The original encountered a different ordering: lock expired after the native target proof was armed. Additional generated controls tested acquisition at that point. control-target-standard2/action-05.json completed one balanced trusted down/up and acquired lock, but the native cursor moved from (398,155) to (396,-22), with hit=0. A subsequent ordinary relative movement reached (399,156) and produced a fresh hover transaction. The unchanged eight-store coordinate journal, shared by DRAW and MANAGER writes, was exhausted; the latest recorded manager point remained (396,-22). Strict proof therefore rejected with no later point change. See target-acquire-boundary.json and reviewed control-target-standard2/response-06.png. No subsequent target click was sent.

This route did not qualify, and its experimental Node adapter reload was never applied to the original. An earlier generated attempt also terminated on missing harness pointer bookkeeping after a Node-only adapter reload; that command had zero attempted effects. The evidence preserves that failure rather than treating it as a pass or replaying it. No native checks or observer quotas were relaxed to admit the original.

Sole original and reviewed responses

The registered original ran from 23:30:50.198Z to normal exit at 23:56:19.422Z on November 6, within its 40-minute limit. The actual owning standard-module Worker/session remained bound throughout. Its reviewed startup/EULA, title and Main Menu captures are retained; original/response-06.png shows Main Menu and original/response-10.png shows the highlighted Single Player target with the completed native (398,155) proof.

original/action-11.json refused the attempted first target click before input, effectsAttempted=0: reviewed relative screen lock had actually become unlocked absolute mode. This was an incompatible protocol transition, not merely a harmless generation increment. An early review sentence incorrectly called that click completed; original/response-11.review-correction.json explicitly corrects it. No original target mouse-down was dispatched.

Freshly reviewed ordinary Enter, Space and ArrowDown events reached the original without a visible menu response. Enter also demonstrates admission across a harmless hint-generation change while actual protocol stayed fixed. Absolute movement attempts were refused with zero effects because the live recentring hint did not satisfy the adapter's stable-unlocked condition. Each new attempt followed a fresh review and a known zero-effect receipt; no uncertain input was replayed. original/response-19.png is the personally reviewed final Main Menu. This run did not reach Single Player, Choose Hero, setup or the world. The predecessor's Choose Hero result remains valid evidence from that separate run.

The concrete remaining work is to qualify an ordinary route for lock expiry after arming, with enough bounded native evidence to certify subsequent targeting. The successful pre-arm acquisition controls alone do not establish that route. There is no demonstrated emulator defect, no guest-memory write, guest patch, heavy benchmark or production repair in this investigation.

Reproducibility and cleanup

source-identity-v*.json records exact Git revisions, relevant binary diffs and run-local helper patches; retained modules and game media are referenced by existing path and SHA-256. No source/media archive was made. The original's 596 HTTP identity checks passed before launch. No full build was needed for this documentation-only publication; no production change is proposed.

All five owned browser sessions are closed. cleanup-runtime.json verifies runner, driver, Chrome and recorded descendants absent, server/debug ports refused and temporary profiles absent. Four sessions exited normally; the bookkeeping-failure control exited 1 as described above. runtime-release.json records the absent lock holder and independent acquisition/release of all three local runtime locks. Explicit release was posted for Lords2 after actual cleanup. No test boxes were allocated or changed. Free /home space remained above 2 GB.