Aranna lock transition, 2026-11-07
The sole fresh NEW-GAME-ARANNA-DEMO-20261010 original reached Choose Hero and passed the predecessor's run-local assertion. It observed natural expiry, refreshed after three proven zero-effect stale refusals, and performed ordinary hover/click from a freshly observed stable unlock. It did not reach Normal, map loading or world movement. The disk safety guard then refused another operation before queuing it; input ended immediately without retry. No emulator defect or source repair is established.
Evidence is retained in scratch/runs/20261107-aranna-lock-transition/. This is a findings-only publication. Production source and guards, guest media, configuration and save bytes were not edited. No workers, test boxes, deployment, source/media archives or predecessor resume were used.
Run-local change and composed qualification
select-action.cjs now returns observation-only when unlocked state has not settled. It selects ordinary absolute hover/click only with stableUnlocked=true, measured relative movement/native click with active screen lock, and explicit acquisition from fresh unlocked relative intent. Acquisition while already locked and invalid coordinates refuse before queue mutation. The authenticated input adapter retains its actual owner, module, protocol, geometry, pointer, native-hit and quota checks. UNKNOWN/partial receipts end input without replay.
The first two generated full-shell controls, standard and compatibility, qualified the whole sequence: active relative lock → natural expiry → actual stale-ticket refusal with zero effects → fresh stable unlock → ordinary hover/click → renewed relative intent → explicit reacquisition → final relative click. Both personally reviewed final images show SEQUENCE COMPLETE. Each had five complete button pairs, 525 complete native transactions and zero incomplete transactions. The unsettled interval selected observation-only. route-qualification.json records the composed checks, not merely active-lock steps.
Nine selector refusal cases and 32 contract boundaries passed. Twelve passive-observer cells compared baseline/disabled/armed execution in both modules, including ordinary APIs and actual Worker CPU/FPU/memory/COM behavior. Both-module broker controls passed signed seeks, short reads, EOF, errors and immutable byte identity. The initial broker setup omitted its generated wasm export and failed before guest execution; broker-controls-missing-input/ preserves that mistake. Explicitly preparing the generated input and running fresh controls passed before the sole original. Two premature generated screenshot descriptions were corrected in control-transition-standard/review-corrections.json.
Original route and new limitation
Lords2-tempdrive explicitly released the local lane at 02:17:27Z. Independent predecessor process, port, profile and four-lock checks passed before acquisition at 02:17:48Z. All generated resources were independently absent before original start, 02:27:58.909Z. The original's inclusive deadline was 03:07:58Z.
EULA acceptance, separate balanced Escape actions for distinct intro scenes, and measured native menu targeting reached Single Player and Start New Game. Choose Hero appeared in response 20. At the predecessor's terminal state (absolute, no lock, unsettled relative intent), command 21 correctly observed without moving. Command 22 reacquired successfully. Subsequent movements 23 and 25 found the lock had expired between capture and dispatch and were refused before input. Acquisition 26 was similarly refused for a stale mode epoch. All three receipts explicitly record effectsAttempted=0; fresh captures followed them. No stale ticket was replayed.
A short continuation script stopped at its assertion after refusal 26; no following action was queued. Fresh observation 27 then established stableUnlocked=true, relativeWanted=false, epoch 23. The script's attempted acquisition selection correctly refused before queue mutation because relative intent was absent. Ordinary route movement 28 and click 29 to host native coordinate (400,530) completed with balanced buttons. Keyboard action 30 typed Aranna once. The personally reviewed response 30 still showed an empty name field. The owning native manager measured (392,599), not the requested host point. This establishes a target-position divergence requiring investigation; it does not establish an emulator defect, successful name entry or a map-loading cause. No typing replay followed.
At approximately 02:39:23–34Z, the disk guard refused the attempted response-30 review/observer-arm operation before either was queued. The recorded check at 02:39:34Z found 1,235,435,520 available bytes, below the 1.5 GB floor. safety-stop.json records the flag and terminal operation. The coordinator was notified on the shared messageboard at 02:39:43.901Z. Only cleanup quit 31 followed, completed at 02:39:43.996Z. There was no retry, rephrasing, delegation or second original after this safety flag.
All 32 submitted commands have terminal receipts: 29 completed, three rejected before input, no UNKNOWN/partial outcomes. Eight recorded mouse-button transactions have completed down/up timestamps. The passive native journal has 600 complete transactions, zero incomplete transactions and peak quota 2,282/8,192.
Resource evidence and identities
original-analysis.json contains 27,310 ordered broker completions, zero byte mismatches, no malformed records, zero map-decision records, zero map completions and zero post-map API records. Resource captures reported zero drops/errors. The only failed opens were early optional C:\DungeonSiege.ini and c:\dir.lqd20, each error 2. Menu progress followed both. Neither is evidence of a map-load failure. No actual ordered failing map read, API or allocation was reached, so no causal map claim or production change is justified.
Source/helper identity v5 is Git commit cb5f15aa7ae1182e68e38efc46c5923140d52e2c plus retained binary diffs and small new-file patches. The post-original source/module check passed before lane release. The original's actual owning Worker instantiated the 2,206,820-byte observer module SHA-256 3b132f07069fda9aa5c961af87122efb6d53caa486649345723d4886c0227dd8; actual target/session/endpoint identities are in the captures and original-module-reviewed.json. Standard baseline SHA is f18229189ae8e50fe938baaa6f2fe2821ee60c297acf7116f5cfeb8529915eaa; compatibility baseline/observer identities are pinned separately.
All 28 existing media paths passed post-run SHA-256/length verification. Actual extracted DSLOA.exe is 3,874,869 bytes, SHA fced0961231248c09c10d86ff51935569db5f83ed3b2ba2739f0501dbc318589. DSLOA/Expansion.dsmap is 2,886,208 bytes, SHA c397bede7fc61ad8650bc835caa6a3fdadd3a3d1c8e3199bc8b1772c2db6dcdc; Expansion.dsres is 226,873,456 bytes, SHA 3ed0688349f33b2dd7b9458c18cfbbc00490353c5888a93cd02c10da7f3c3b10. These are extracted-media identities, not native installer provenance. HTTP-served identities and environment are retained separately.
Cleanup and next step
Chrome exited zero at 02:39:45.460Z after the guest stopped and observations drained. Surrounding exec sessions reported 143; their cause is unknown and is not represented as a normal supervisor exit. Independent final-cleanup.json at 02:40:17.956Z verifies every tracked process/profile absent, twelve ports refusing connections, all four locks free, and 1,851,486,208 bytes available. No boxes were assigned or created (ID/state/expiry: not applicable). Runtime/source/build was explicitly released at 02:40Z; no runtime was resumed during closeout.
A separately authorized task should first resolve and qualify the host-to-native point divergence across the real transition, then attempt a fresh original after verifying disk headroom. This original is terminal. Normal, the first actual map-loading failure, and personally reviewed world movement remain unobserved.