Files
ARCADIA-Studio/harness/scenario-selection-evidence-20260924/native/README.md
T

29 lines
4.0 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# Native Windows selection exercise
Source `ed55b1781d86ec49af5d2588cbd06f7b878d9009`, closed build `selection-build-1`. The launcher checked that receipt, source proof, binary and tracked fixture before copying the executable and opening an isolated fixture through the app. The archive omits the executable; its identity is recorded in `started.json` and the build receipt. This is branch-specific verification, not the #37 candidate or a release package.
Run `python -B verify.py` with Python 3.11+ to verify the archive and exact file comparisons. The verifier does not execute archived helpers or establish visual assertions. `native.zip` contains original screenshot/metadata bytes, launch/close records, empty app logs, fixture snapshots and the original launch/capture helpers. `prepare.py` requires the retained local run and refuses an existing output directory.
## Observations
All numbered files below are under `captures/` inside the archive. Native inputs used the Computer Use skill's `@oai/sky` API.
- 17: selected saved A in a maximized window. 18–21: contextual Add creates a two-occurrence selection; count, both strokes and Properties primary agree.
- 22–23: group menu exposes selection commands and disabled movement/removal. Inspect first occurrence preserves both members and switches primary.
- 24–26: dragging a selected body and pressing Delete refuse the whole group, retain both members and display the unfinished-operation feedback.
- 27: a single blank drag from (1110,730) to (1060,700) clears the group and moves the visible geometry by (-50,-30). No second press was needed.
- 28–31: contextual Add restores the two-member group after panning. 32: a stationary plain click on the first member collapses to that singleton.
- 33: Save all after selection and pan preserves the entire original 6,206-byte file, SHA256 `74cd2c1dd5eed7eb19895a752bb8ff9ce93655f7638a5312c96f56218568ebb3`.
- 34–35: with the first occurrence selected, the first drag of the second body from (675,454) to (675,474) immediately moves it and selects it. Save all changes only saved A's `updated_report` `send_y`, from 252 to 281.8841247558594. The entire expected file is compared, including unchanged semantic order and other representations.
- 36–37: one Ctrl+Z and Save all restore exact baseline bytes.
- 38–44: A → B → Automatic → A displays each correct representation, retains the existing singleton primary, and final Save all preserves baseline. This is not a native group-clearing-on-navigation claim.
- 46: app remains usable after an unsuccessful title-bar close attempt. Alt+F4 then closes normally with exit 0; binary and fixture unchanged; stdout and stderr empty.
## Limits and automation anomalies
Native Shift-click was not exercised because the API cannot hold Shift across a click. Actual modifier/focus/busy/window-loss sequences are covered by the mounted widget tests in the separate Cargo evidence. Native Redo, dirty Properties admission, stale queued events, group navigation and fallback C were not independently exercised here; test coverage must not be described as native coverage.
Early nonmaximized picker targeting selected B while intending A. Maximizing and using the full-window screenshot resolved targeting; no product picker defect is established. One action selected a 16×16 icon screenshot and was rejected for out-of-bounds coordinates before any app action. Early setup observations remain local; accepted captures start at 17. Later indexed title-bar Close minimized the window instead; restoration was observed, then Alt+F4 succeeded. These are automation limitations, not evidence of a product crash. Capture labels record intent; claims use observed state. Capture 45 produced no file because the API reported a minimized window.
The fixture intentionally contains five model checks and an incomplete occurrence. No new checks appeared. Group semantic movement and aggregate removal remain unfinished. Screenshots and successful launch do not imply full usability or release acceptance.