5.9 KiB
Connection attachments: Windows evidence and handoff
This evidence-only directory accompanies source commit
53d01a3d4451b94792986f2dfac56f6f04e111a4 on
codex/edge-anchors-escape-ytr83r-20260923. It preserves the original attachment
implementation checks and the subsequent State Escape correction. It is not an
integrated release or a runnable package. Live ownership remains in
issue 33
and integration issue 27.
Results by source version
| Source | Automated checks | Native Windows evidence |
|---|---|---|
3aa80de89c5b79200286926781e8f3a192bc61f4 |
Workspace: 1,819 passed, 0 failed, 7 ignored; strict workspace all-target Clippy; GUI, CLI and evaluator builds | Three closed runs, 48 captures, 29 completed full-Project comparisons: 25 pass and four pixel-offset mismatches retained as failures. OA/State attachment editing, per-end reset, history, independent views, symbol contours, semantic-port movement and valid reconnect/undo. Immediate post-menu State Escape failed. |
53d01a3d4451b94792986f2dfac56f6f04e111a4 |
Two mounted UI regression tests; desktop: 1,194 passed, 0 failed, 6 ignored; strict desktop all-target Clippy and GUI build | Two closed runs, 12 captures, five exact full-Project comparisons. Source and target menu actions cancel with immediate Escape. One source-end attachment edit, Undo, Redo and fresh-process reopen/save pass. |
The corrected GUI uses the original evaluator at 3aa80de; committed core and
Cargo inputs are unchanged. Both corrected processes and all three original
processes closed normally with exit 0 and empty GUI logs. Reused build caches
mean these are not clean-build claims. Linux remains paused by the user.
The tested host used Rust 1.98.1, target x86_64-pc-windows-gnullvm, LLVM 22.1.8
and LLVM-MinGW runtime. Exact commands and environment are retained in gate
receipts. Principal checks were cargo test --workspace --locked --offline and
cargo clippy --workspace --all-targets --locked --offline -- -D warnings at
3aa80de; the corrected desktop equivalents use -p oa-desktop instead of
--workspace. Build with cargo build -p oa-desktop --locked in the source
checkout using its documented toolchain. A new host may first need to fetch
dependencies before using --offline. Use a fresh target directory for a new
candidate; do not rely on the restored shared executable's Cargo freshness.
- Original archive and explicit failed attempts.
- Corrected Escape archive and independent native reviews.
- Exact import sequence and integration seams.
- Decision and verification interpretation.
The inner READMEs retain their preparation-time wording. This outer directory provides the peer handoff; the containing Git commit is its publication identity. Original receipts, paths, failed expectations and historical pending statements were preserved rather than rewritten after subsequent success.
Portable verification
Fetch the source branch into a local clone so its 79cc9d5, 3aa80de and
53d01a3 commits are available. From this directory run:
python original/verify-build.py verify --archive original --repo /path/to/ARCADIA-Studio
python corrected/verify-build.py verify --archive corrected --repo /path/to/ARCADIA-Studio --related-archive original
The verifiers map historical machine paths to archive bytes or source Git blobs;
peers do not need the originating .local tree. Exact raw source overrides
preserve captured CRLF bytes where needed. Archive .gitattributes prevents
text conversion. Original manifest SHA-256:
2379fb100e529772ea6149554fee71aca80ab956539e97aa4d288b189ff55131.
Corrected manifest SHA-256:
bee4ff55e8d4e1044785deca3f0382f7776dfa5c0e629b21457b9408eb7f51b2.
GUI/evaluator/runtime binaries, private profiles and mutable working directories are excluded. Their historical identities remain in receipts, but the portable verifier does not rehash absent executables or rerun the application. Immutable saved-source checkpoints, expected complete Projects, actual evaluated Projects, commands, logs, capture metadata and original images are included.
Limitations that accompany these results
The original shared target executable was temporarily overwritten by the corrected build. One preflight correctly aborted before comparison. The original staged application stayed unchanged; the corrected artifact was preserved, the original target restored, and a fresh comparison passed. Later equality is not continuous whole-run immutability. The corrected stage avoids shared-target bindings and records the relocation explicitly.
The original reconnect-to-Choice expectation was invalid because Choice already had an incoming transition. The application correctly refused it. A separately predeclared valid reconnect-to-Final and Undo passed; the invalid expectation and validation error remain archived. Four midpoint attempts missed the exact declared offset; subsequent pointer/zoom alignment passed without changing the oracle. These are not silently counted as successful attempts.
Corrected cancellation has clean visual evidence after each end and one combined baseline comparison after both, rather than independently saved per-end checkpoints. Run 1's close-action timestamp records retrospective attestation, not the exact key event. Screenshots show states; the operator's action records supply sequence and cause. Complete native drags do not establish every held pointer/keyboard/wheel interaction. Dense small labels and truncated guidance remain visible limitations. Automated event tests supplement the native scope.
Process and Scenario use different notations and do not consume this ordinary route contract. Their arbitrary attachment editing, State bend issue 23, auto-arrangement, resize issue 16 and the combined release are not claimed done.