Aranna relative input route, 2026-11-07

The single fresh NEW-GAME-ARANNA-DEMO-20261010 original advanced through Single Player and Start New Game to Choose Hero. It stopped on a run-local input-selection assertion before another command was queued. Normal, map loading and world movement were not reached. No emulator defect or repair is established. The predecessor's terminal canvas-acquisition attempt was never resumed.

Evidence is retained in scratch/runs/20261107-aranna-relative-map-load/. This is a findings-only publication; production source, guards, guest, media, configuration and save bytes were not edited.

Qualification before the original

select-action.cjs and the run-local queue wrapper select measured relative movement and balanced native clicks when the fresh capture says lock=screen. They reject canvas acquisition while already locked, absolute coordinates on relative clicks, excessive movement, and uncertified unlocked movement before queue effects. The existing authenticated input adapter and owner/module/target/protocol/point/quota guards remain unchanged.

Both standard and compatibility generated full-shell controls personally passed the exact Single Player → Start New Game → Normal sequence with screen lock already active. Fresh captures, reviewed geometry and measured native points (399,155 standard; 398,155 compatibility) preceded the selections. Each reached visible SEQUENCE COMPLETE, stage 3, four button downs and four ups including initial acquisition, 466 complete native transactions and zero incomplete transactions. Intermediate black GDI captures were recorded as black; subsequent visible stage captures established the results. These generated controls do not prove the original reached Normal.

route-qualification.json, selection-controls.json, route-contract-controls.json and active-route-refusals.json retain the results: 32 contract boundaries, eight invalid selector cases, three actual active-lock CLI refusals with unchanged queue counts, and six ticket mutations per module all passed. Both-module passive observer controls compared ordinary APIs, CPU/flags/FPU/memory and actual Worker execution against baseline; production Worker/DLL broker controls exercised ordered reads, signed seeks, EOF, errors and byte identity. An initial broker-control launch failed before guest execution because its generated wasm input had not been prepared; this is preserved in broker-controls-missing-input/. Preparing that control input and running fresh generated controls passed. This was before the sole original. No new journal-saturation study was performed.

Sole original and terminal operation

The attempt started at 02:00:45.385Z with a 40-minute inclusive deadline. EULA acceptance, loading and an ordinary balanced Escape led to the rendered main menu. Acquisition occurred while unlocked. Measured relative movement then positioned Single Player at (399,155); command 16 completed its native click with authenticated hit proof. After a fresh submenu capture, command 18 moved (0,110) to (399,265), and command 19 completed the balanced Start New Game click using the existing relative route. This passed the predecessor's active-lock obstacle.

Pointer-lock release was observed at 02:08:15.014Z. Capture original/response-20.png at 02:08:29.670Z visibly showed Choose Hero, a female model and an empty name field. Its fresh protocol state was mode=absolute, lock=null, stableUnlocked=false, relativeWanted=true, everLocked=true, protocol epoch 5. After reviewing that capture at 02:08:47.594Z, the worker selected route-move response-20 1,265 without checking the changed stable mode. This was an operator selection error. The run-local selector correctly asserted s.stableUnlocked before queue mutation or input effects.

The assertion itself has no separate clock record: it occurred between that review and the persisted blocker observation at 02:09:09.828916Z. original/terminal-selection-assertion.json records zero effects, no queued command, last completed sequence 20 and terminal status. Per the task's terminal-assertion rule, no movement, reacquisition, replay or second original followed. Only normal quit (21) was queued. All 22 submitted commands completed, with zero UNKNOWN/partial replies; 266 native transactions completed with zero incomplete transactions. No automated safety flag was observed.

What the passive evidence establishes

original-analysis.json contains 27,310 ordered broker completions, zero byte mismatches and zero malformed records. Resource coverage reported zero drops and errors. There were zero map-decision records, zero map completions and zero post-map API records: the map-loading path never ran. The inherited native journal peaked at 1,246 of 8,192 entries.

The only failed file opens were early C:\DungeonSiege.ini and c:\dir.lqd20, each OPEN_EXISTING returning 0xffffffff/error 2. Menu progress followed both. Missing optional startup files alone establish no map-load cause. No failing map read, signed seek, allocation result or decision branch exists to compare against source bytes and API contracts. The documented SetFilePointer, ReadFile and VirtualAlloc contracts therefore justify no production change from this run. There is no failing-before regression or locked full build for this documentation-only publication.

Identity, cleanup and handoff

The original used Git commit da3b92bcdf50091b8a2d6c1cb0bd933729fa5b75, source identity v4. The standard observer wasm SHA-256 was 9c2308db394f3fa5ce791ddfb35e5a4ace49fb015dd80c982e32248dd2851b76; compatibility qualification used 8f41da76ab6e5106b6d7536940853f3e5534ace2a98b3a42d8817a0a80f4612c. source-post-original.json verifies all 1,607 pinned source/helper inputs unchanged at 02:09:26.649Z. Git SHA, binary diffs and relevant new-file patches preserve source identity without source-tree archives. HTTP pins and all 28 retained media hashes matched; media is referenced by existing path and hash, not copied. This proves the retained extracted-media identity, not native installer provenance.

Lords2 explicitly released runtime/source/build at 01:51:31Z; independent process, port, profile and four-lock checks passed before this worker took ownership at 01:52:01Z. After the original, the recorded browser driver and Chrome exited zero at 02:09:10Z. The surrounding exec session reported 143 for an unknown reason; original-supervisor-status.json preserves that discrepancy rather than claiming every supervisor exited normally.

final-cleanup.json independently verifies all retained control/original processes, ports and profiles absent. Runtime holder and board watcher absence were verified. Four locks were acquired/released successfully, with 3,253,547,008 bytes free on /home. Runtime/source/build was explicitly released at 02:09:47Z, about nine minutes after original start; Lords2 took ownership at 02:10:39Z. No test boxes were used (ID/state/expiry: not applicable). No runtime was resumed during findings closeout.

A separately authorized fresh task should qualify the natural locked-to-unlocked transition in generated full-shell controls and choose acquisition versus movement from the actual fresh mode/lock state before any original. This original is terminal and must not be resumed. The first actual map failure and world movement remain unobserved.