Aranna native point, 2026-11-07
The retained target divergence is explained and the correction passed both-module
full-shell controls. The one fresh original ended at a precise run-local command
reader failure before the menu; world movement remains unverified. Evidence is retained in
scratch/runs/20261107-aranna-native-point/. Production guards and guest
media/configuration/save bytes remain unchanged. No emulator defect is inferred
from a host/native coordinate difference.
Retained coordinate cause
Offline examination of the terminal lock-transition original resolves the previous target divergence. Immediately before action28, the native manager point was (395,265), independently measured from owning store records. The driver's requested host (400,530) became browser (340,668) after CSS integer quantization, then logical canvas (399,500), and host/native screen (399,529) through the actual exclusive presentation viewport.
Aranna recentres its host cursor to (400,300) and accumulates displacement into its own UI position. The actual move therefore added (-1,+229), producing (394,494). Action29's down and up each delivered the same nonzero displacement again. The resulting native positions were (393,599), then (392,599), including the observed screen-edge clamp. This exactly accounts for the retained final point and the empty name field. Browser unlock did not change the guest's coordinate protocol. This is a run-driver targeting error; these observations establish no emulator instruction or API discrepancy.
retained-point-cause.json records the actual host events, owning native
stores, source evidence hashes and a checked replay of these three additions.
The prior original remains terminal and was never resumed.
Intended correction
The run-local driver maps to integer CSS positions by evaluating the actual floor/floor/clamp route, recording the representable host point. It plans a displacement from the freshly observed native UI point and actual host recenter, then returns to the exactly representable neutral recenter before ordinary unlocked down/up. A separate capture verifies that neutral movement preserves the native target. Relative movement remains selected only under actual screen lock. Unsettled unlock remains observation-only; explicit reacquisition requires fresh relative intent. Existing exact owner/module, image review, protocol, geometry, movement, native target and quota guards remain in force.
Generated controls preserve recenter-and-accumulate message handling after periodic recentring stops; browser unlock therefore does not turn their native UI into absolute input. Ordinary native character messages record and render the typed string. Neither control mechanism writes original guest state.
Runtime handoff
Lords2-file-copy explicitly released runtime/source/build at 02:46:09Z. At 02:47:42Z the successor independently checked all five recorded processes and the profile absent, both ports refused and all four locks free, with 2,930,786,304 bytes available. No boxes were assigned or created.
Qualification
Both standard and compatibility full-shell generated controls passed the composed
route. Each observed real lock expiry, an actually stale ticket refused with
effectsAttempted=0, fresh stable unlock, measured displacement to native
(410,165), exactly neutral host (400,300), balanced click at unchanged native
(410,165), six native character messages rendering Aranna, an observation-only
unsettled interval, trusted reacquisition, and relative correction to (399,155).
Each completed five down/up pairs with no UNKNOWN or partial receipt. Actual
native hover/store transactions and reviewed captures establish the points;
completion was visibly reviewed in standard response13 and compatibility
response12. Subsequent black GDI frames are recorded as black, not described as
visible completion. Both controls exited normally and their processes, ports
and profiles were independently absent before the original.
Six valid mapping cases and five invalid inputs, nine selector refusals and 32 route-contract boundaries passed. Twelve baseline/disabled/armed observer cells across both modules passed, including actual Worker execution. Both-module broker controls passed signed seek, short read, EOF, errors and immutable byte identity. One broker preparation command used a relative wasm path and failed before guest execution; an absolute-path preparation then passed. The failed preparation log remains retained. Production guards were not relaxed.
Sole fresh original and first new blocker
The original started at 03:00:36Z, with its inclusive 40-minute deadline. EULA acceptance and three distinct, personally reviewed intro scenes completed using one mouse down/up pair and three balanced Escape presses. Response08 shows a black transition with the sword cursor, not a menu. Choose Hero, successful name entry, Normal, map loading and world movement were not reached in this original.
Command9 requested only observation (wait:1200, screenshot). The reader failed
at queue-runtime.cjs:15:22 with SyntaxError: Unexpected end of JSON input.
It did not create 9.admitted.json or 9.done.json. Chrome exited zero at
03:04:09.789Z, and the driver exited one at 03:04:10.029Z. The waiting operator
then reported receipt timeout and ended input without replay or another original.
Nine earlier commands have completed receipts; the final observation command
has an UNKNOWN terminal outcome. It is not represented as a proven zero-effect
stale refusal, even though its requested steps contain no input verb.
The run-local publisher uses writeFileSync(finalPath, JSON, {flag:'wx'}) and
the reader polls existence then immediately parses the file. The exception and
source support a publication/read race in which the pathname becomes visible
before its contents are complete. The bytes seen by the failed read were not
captured, so that exact race remains an inference. The final retained command
file is valid JSON. This is a harness blocker; it establishes no emulator
instruction/API defect. A future task should qualify atomic command publication
before another separately authorized original. No held-risk retry occurred.
Resource and identity evidence
Reused passive observers retained 25,623 ordered broker completions with zero
byte mismatches, zero malformed records, no map decisions, no map completions
and no post-map API records. Captures reported zero drops/errors. The two early
optional opens (C:\DungeonSiege.ini, c:\dir.lqd20) returned error2 before
subsequent intro progress. They are not evidence of a map-load failure. There
was no actual ordered failing map read/API/allocation to attribute. The native
journal contains 247 complete transactions, zero incomplete, peak quota1066 of
8192. Each full-shell point control contains 545 complete transactions and zero
incomplete. No quota or production guard was reset or disabled.
Source/helper pin v5 records Git commit
d96dd8ab7bf776350dc2ec1a73d708a8cefe543f, relevant tracked binary diffs and
28 relevant untracked source patches. Source/module checks passed after the
original and before lane release. The actual owning Worker instantiated the
2,206,837-byte standard observer, SHA-256
e7d6617cecdf3275db05454763c07cad7e6eb557afc2a14ca3413d960328fd31.
The standard baseline SHA is
07f1cc970038f8ab31313b8b200edf26caa63473f293f5fc937f5cfe5c5c7f6e.
Compatibility baseline/observer SHAs are
1d4bd4330ea16947959e3f4472339d5a32a61216cd8b5246198d11046b652c3e and
6b5edb202b5e03d84056b8663869f8485de2b5a7d592ece0e350dc43a88a005d.
Actual Worker target/session/endpoint and served identities are retained in
original-module-reviewed.json, captures and original-http-identity.json.
All 28 existing media paths passed post-run length/SHA-256 checks. 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 identify existing extracted bytes, not installer provenance. Media was
referenced without copying. Guest media, configuration and save bytes were not
edited. No source/media archive, deployment, nested worker or test box was used.
Cleanup and handoff
final-cleanup.json at 03:05:24.524Z independently verified 50 tracked PIDs
absent, ten ports refusing connections, five profiles absent, all four locks
free and 2,819,215,360 available bytes. Cleanup finished within five minutes of
original start. Runtime/source/build was explicitly released at 03:05:33Z.
Chrome and browser resources closed normally; the driver failure is retained
as nonzero, not described as a normal supervisor exit. The surrounding runtime
lock-holder exec session reported143 after its process disappeared; all locks
were independently free. No boxes were assigned or created, so box ID/state/
expiry and shutdown handoff are not applicable.
The >2GB operational disk guard stayed enabled and did not end this original. No automated cybersecurity flag was observed. Findings only are published; production source was not changed, and no new full build was needed for this document-only publication. Next is a separately assigned atomic-publication qualification, followed by any newly authorized fresh original. The retained original is terminal.