Files
ARCADIA-Studio/harness/prerelease37-independent-9e8r6t-20260924/native-observations.md
T

5.8 KiB
Raw Blame History

Downloaded Windows Studio: direct native observations

Codex-9e8r6t /root, 2026-09-24. Tested the unmodified fresh extraction of release 7 asset 25 on Windows 11 build 26200. Product source is 715036e132677dd2bc2fb37ed316d93dc9987aec; no Cargo/build or Linux operation occurred in this exercise. This is independent of the producer's machine and GUI exercise.

Launch and workflows

native-launch.py launched the packaged executable directly with PATH C:\Windows\System32;C:\Windows, removed development-related environment variables, and used unrelated working directory C:/arcadia-package-qa-9e8r6t. It retained the ordinary Windows user profile, including existing recovery entries; this is not a fresh-OS/profile test. Python only orchestrated the launch and logging. The original extracted bundle was not edited. A whole companion kit and example directory were copied to the working directory before opening the copied State model. Both application sessions closed normally with Alt+F4, exit 0, empty stdout/stderr and unchanged executable/runtime hashes.

Original screenshots and their timestamps/window identity/accessibility trees are in native-2/ and native-3/. Each capture includes a full-window JPEG, plus a tiny icon image in most captures. Native actions used Computer Use, without application-state injection. Visual acceptance below is bounded to the actual observed sequence.

Captures Observed behavior
native-2 01–04 Home launched; bundled water-pressure walkthrough loaded from the unrelated cwd (94 objects, 62 links, 1 model check).
native-2 05–09 Opened copied State fixture; selected System perspective, readiness view 03 and Activate transition; Actions → Add bend displayed an additional square handle.
native-2 10–11 Undo removed the added handle; Redo restored it. These two observations are visual, not saved full-file snapshots.
native-2 12–17 Dragged new square downward; Save, Undo movement/Save, Redo movement/Save. Whole-source checks establish one additional waypoint, original (500,310) retained, added point moving (526,310) → (530,385), and exact movement Undo/Redo.
native-2 18–21 Switched to untouched reference view 04, then returned to edited view 03. Reference layout remained distinct; the authored edit returned. Whole-file comparison also leaves every reference-view byte unchanged.
native-2 22–29 Opened Edit transition and its source-state dropdown without changing fields. First Escape closed the dropdown and kept the form. Second Escape kept the form too. Explicit Cancel closed it; saving preserved the complete edited source exactly.
native-3 01–06 Fresh process; reopened the working model, returned to System/State view 03, selected Activate. Both authored handles and downward bend persisted. Save produced identical complete source bytes.
native-3 07–08 Opened built-in Engineering guide and its States, modes & transitions topic; guide content rendered normally. No external network-disconnection test was performed.

native-check.py is a read-only full-byte evaluator. It starts with the independently downloaded companion baseline and allows only one exact addition in Activate's waypoint list in readiness_view. Consequently semantics, IDs, node positions, endpoint attachments, labels, other routes and all other views retain their original bytes. Its output is native-check.json. The five accepted saved snapshots are retained. A separate source reviewer checks the oracle and key original images in native-source-review.md.

Scope and observations that are not pass claims

The filename native-2/26-second-escape-cancels-form records the intended probe, not its outcome. Captures 26 and 27 show the form still open. Source review in picker-policy-review.md confirms that ordinary State edit drafts use explicit Cancel; #35 retains existing surrounding Escape policies. No second-Escape cancellation or direct inspection of the history stack is claimed. After explicit Cancel, exact source preservation was checked; no fields had been edited in that form.

At 56% Fit, this fixture's transition labels are small and crowded. That limits readability and remains within the existing label/layout follow-up; this run does not establish broad diagram ergonomics, accessibility, label clarity or aggregate editing acceptance. The fixture intentionally reports 12 model checks (14 objects/9 links); the modeling issues were not fixed as part of a presentation-only check. This exercise does not cover all release-checklist cases; producer and other reviewers retain their separate scopes.

The first launcher invocation (native-1) failed before product startup because the copied environment dictionary used the wrong case for SystemRoot. The helper was corrected to read the case-insensitive original environment; the empty native-1 directory contains no run evidence and is excluded. File-dialog automation needed direct visible Filename entry after UIA set_value was unsupported. On reopening, an undeclared JavaScript local caused one tool call to fail before its action; declaration then succeeded. These are harness/tool errors, not product crashes. Raw successful native-2/native-3 captures and receipts are preserved without renaming or overwriting.

No release/tag changes or publication are performed by this QA branch. Codex-956atf owns release admission and publication.

Parent read-only server evidence reconciliation initially used Python's Windows default text decoding for accepted-envelope.json and failed equality on the Unicode comment. Explicit UTF-8 decoding made the saved HTTP/envelope/store comparison pass. No runtime request, product file, or captured source was changed; server-readback-check.py and its result preserve the corrected evaluator. The original packaged-server runtime smoke itself passed on its first run.