ARCADIA-Studio/harness/experiments/compact-help-20260925/candidate-5657/review.md

Independent native successor review

Reviewed source 5657a996ed2cefb1fa84ef967b655eaf8f9f6817; executable SHA-256 83CCEFF8B713491AFEE92FFCAFE62B02F5EEA6A3D899F6F334848FDC7BB0D88D independently rechecked after the run; runtime SHA-256 F45604A12E8BCD839FD86F667C1BE1CC7C4F9DA2B3C69FF381AC708F0E1AE7D4 from build.json. Used Windows native Computer Use, the isolated model/project.arcadia, normal 1440x900 and minimum 1100x700 client sizes (1442x932 and 1102x732 captures). The prior candidate report remains historical under ../candidate-f830/review.md.

Result

Both prior blocking native findings are corrected in the exercised workflows. No new blocking finding was observed in this bounded retest. This is not complete product, accessibility, multiselection or first-release acceptance.

Help keyboard reading and caret preservation

1. In normal Allocation, focused inspector Name, selected its text, typed LEFT RIGHT, and pressed Ctrl+Left to put the caret before RIGHT. Opened Help with F1. PageDown visibly scrolled to lower sections; PageUp returned to the top; End reached the bottom; Home returned to the top; Down moved one increment and Up reversed it. Evidence: 02-help-pagedown.png, 03-help-end.png.
2. Pressed Escape, then typed X without refocusing. Name became exactly LEFT XRIGHT. Reopened with F1, clicked visible Close, then typed Y without refocusing. Name became exactly LEFT XYRIGHT. Thus keyboard reading did not move the background caret, and both dismissal paths preserved input continuation. Evidence: 04-caret-after-escape.png, 05-caret-after-close.png. Revert restored the accepted property.
3. Opened Help again and resized the native window to its minimum width and height through the native system menu. Help reflowed with Close reachable; End still reached the bottom and Escape returned to the same diagram. Evidence: 06-help-minimum-resized.png.
4. At minimum size, opened System functional chain Handle a pressure alarm, chose W, clicked blank canvas, and entered LEFT RIGHT in the pending contribution Name. Ctrl+Left, F1, End, Escape, then X produced exactly LEFT XRIGHT without refocusing. Create remained disabled with its missing-source explanation and Cancel reachable. Cancel preserved the accepted model. Evidence: 15-process-caret-restored.png.

Compact controls and actionable feedback

• Normal and minimum Allocation show projection, Help and Arrange together, with a separate full-width Destination slot. At minimum size the canvas begins near y=350 versus baseline y=414, approximately 64 pixels recovered. Changing from no selection to a selected work card and exercising refused drags did not move the canvas boundary. Evidence: 01-allocation-normal.png, 07-allocation-minimum.png.
• Clicking Arrange displays Arrange ✓; the tooltip explains one-drag scope, Alt+A and Escape. Escape removes the check; Alt+A arms it again. Evidence: 08-arrange-tooltip-armed.png.
• With Arrange armed, dragged Assess pressure alarm from Control room toward Field technician. The operation was refused with the visible footer message Make room or adjust the view, then drag again. Zoom in to read the work and destination. The accepted model remained saved and the work stayed allocated to Control room. Evidence: 09-arrange-refusal.png.
• Repeated the drag without Arrange. Core validation refused the reassignment with exchange_allocation_performer: Actual activity allocations must agree with one permitted communication direction; review the means or the activity allocations. The shortened footer remained visible, and pointing at it exposed the complete readable explanation. The selected activity and accepted model remained unchanged. Evidence: 10-assignment-refusal.png, 11-refusal-tooltip.png.
• Changed View content to Traceability. Count, Add existing elements, Review layer traces, View visibility and Help all remain reachable at minimum width. Canvas starts near y=356 versus baseline y=470, approximately 114 pixels recovered. Undid this temporary projection change. Evidence: 12-trace-minimum.png.
• Opened saved State view 5 · SA pressure supervision: the generic idle canvas instruction is removed. Choosing +Region still shows its active creation prompt. F1 opens the States topic; Escape returns with +Region still selected, without creating an object. Evidence: 13-state-idle.png, 14-state-active-prompt.png.

Limitations and minor observations

• The existing State active prompt still truncates at minimum width (Create Region ... machine or composite s...). This retest confirms it was retained, not that every pre-existing State prompt has complete narrow-window readability.
• The compact control says Arrange, while the guide still says Arrange next drag; its tooltip explains that wording. Minor consistency cleanup was reported to the integration owner; no user workflow failed because of it.
• Modifier-click/native multiselection and long-name group toolbar layout were not independently operated because the supported Computer Use API has no modifier-click/key-hold operation. Do not infer them from the single-selection screenshots. The parent owns mounted-widget verification separately. Source-editor caret behavior and full Scenario workflows were not repeated in this bounded successor retest.
• The approved native launch API does not expose stdout/stderr or process exit status. After Alt+F4, refreshed native window enumeration confirmed both successor application and console windows disappeared. No crash dialog was observed; this is not an asserted exit code or log inspection.

No save was performed. A post-run hash comparison against baseline-proof.json found no changed files among all 37 copied fixture files. The UI returned to the saved baseline with 94 objects, 62 links and one model check before closure. Screenshots are supporting evidence; the results above describe actual input and observed responses.