Native C recovery, Undo/Redo and reopen: b10, Runs 2 and 3
These are two closed native runs of source b10b28cefd1d0794073ba241a0c8de9a83e85f68. They establish the bounded observations and saved-Project comparisons below. They do not complete the native plan or establish overall issue acceptance. The parent operated the application; the archiving reviewer inspected the closed records, representative screenshots and exact file/Project differences without sending native input or rerunning any program.
Results
| Run | Bound process | Close | Saved comparisons |
|---|---|---|---|
| Run 2 | PID 8204, 15:01:36–15:23:35 UTC | Alt+F4 observed; exit 0; stdout/stderr empty | Refused/cancelled: same; C repair: only 300→424; Undo: same as original; Redo: same as repaired |
| Run 3 | PID 4848, 15:24:22–15:26:20 UTC | Alt+F4 observed; exit 0; stdout/stderr empty | Reopened and saved C: same as repaired |
Run 2 requested an A self-Call position below its Reply. The application refused it and reported that the saved arrangement was unchanged; model Undo reported nothing to undo. A separate message drag released outside the canvas left the saved source unchanged. No distinct cancellation feedback is claimed: the previous status remained. Refusal image and comparison.
C initially retained its Saved identity and saved-position notices while displaying complete automatic fallback. Selection and Focus/Restore worked; the attempted creation and spacing edit did not alter the saved model. An unapplied Text change kept fallback in Behavior and showed pending source drafts with Properties unavailable. Explicit Apply all then changed only C's stored reply position and restored the authored reversed-lane geometry. The unfinished-position notice remained. See fallback, pending draft and applied repair.
The repair is exactly one source substitution, send_y 300 → send_y 424. Independently comparing the complete parsed baseline and actual Project gives exactly one changed leaf:
/diagrams/70200000-0000-4000-8000-000000000042/layout/Scenario/messages/0/send_y.
That record addresses occurrence 70200000-0000-4000-8000-000000000023 in C. Every other Project field remains exact, including C's lane x values 900 and −500, its unfinished stored position 2000, A/B, and semantic ordering. Repair comparison and the adjacent baseline/actual/expected Project JSON make this independently inspectable.
One model Undo restored fallback and the original saved file; one Redo restored authored C and the repaired file. Run 3 reopened the exact incident file and explicitly selected C through the arrangement picker, then saved an identical repaired file. Automatic last-view persistence is not claimed. Undo, Redo, Run 3 reopen.
All seven checkpoints are 6,206 bytes. Run 2 checkpoints 01, 02, 03 and 05 have SHA-256 74cd2c1dd5eed7eb19895a752bb8ff9ce93655f7638a5312c96f56218568ebb3. Run 2 checkpoints 04 and 06 and Run 3 checkpoint 01 have SHA-256 fc1e12f18de4f3841140998657a8e117f41711ec80ec0bc6aebf0e9c0d61f827. All five comparator receipts report exit 0, unchanged bound inputs, and complete saved-Project equality against the expected result. Their output sizes/hashes, before/after captures and actual-versus-expected JSON bytes were rechecked during archiving.
Capture and evidence limits
The 14 Run 2 JPEGs and one Run 3 JPEG have 15 distinct SHA-256 values matching their adjacent metadata; JPEG file signatures and declared MIME agree. Four representative captures (refusal, C fallback, applied repair, reopened C) were visually inspected during archive preparation. Distinct hashes alone do not prove every described action. The direct operator notes, screenshots and saved-state comparisons have separate roles.
Run 1's stale capture closure is documented in its published archive. Run 2 used an explicit current-state argument; these images are new Run 2/3 records, not replacements for Run 1. A Run 2 tool call encountered an undefined window binding before issuing application input; the operator refreshed the binding. No input was repeated by the archiver. Original notes and metadata are preserved verbatim, including their compressed wording.
Positive native spacing and the remaining service Details workflows were not completed in these runs. C was repaired before any positive A spacing, so its baseline contains no cumulative A edit. Held-drag trajectories, exact OS packet coordinates, general accessibility and internal runtime/history-stack serialization are not established. Native observations support the reported single Undo/Redo sequence; external comparators establish complete saved Project equality, not internal editor state or camera identity. This archive excludes every other run, current working files, private profile, executables/DLLs and duplicate gate caches.
Prerequisites and integrity
The launch chain binds Run 2 to Run 1's successful close, then Run 3 to Run 2's close. Both launch records identify the same package, child-only environment, source pin, native input manifest and launch-script hash. Normal-close receipts bind immutable inputs and empty logs. Original absolute machine paths in receipts are provenance, not portable commands.
- Published Run 1 preparation and limitations at 708af836 supplies the immutable native fixture/plan copies, corrected timestamp staging recipe, input manifest, launch script and preceding close receipt.
- Published comparator and GUI evidence at 04129a75 supplies the spacing comparator source/build/self-check, exact dependency captures, frozen verifier and accepted GUI build. The old
8bf7607comparator directory name is ancestry; the current bound comparator and native GUI source are b10.
GUI: 116,721,152 bytes, SHA-256 7ece85c6012ae0ae75fc56ed3c3aadb8bc116448ad1782be422ef34a629dc3ce. Spacing comparator: 22,720,512 bytes, SHA-256 ce536924836d62210639a3af76d15793d0bd51a429d8cac0c90edde66955fe62. Runtime DLL: 190,464 bytes, SHA-256 aa282fa47bb2c4e210cea5e470cdbdca58568f0be0bc73e404d86ab9cd95ef6a. These payloads are referenced, not recopied.
archive-manifest.json inventories all 102 originals (2,172,638 bytes), exact checkpoint/comparator/capture bindings and published prerequisite Git blobs. Run 2 contributes 83 files and Run 3 contributes 19. It also hashes this README and the raw-evidence attributes; the manifest does not recursively hash itself. Every original was copied byte for byte and compared again to its closed source. Nine raw .log files, including empty native logs, are intentional evidence and may require explicit forced Git addition. No program execution or source modification was part of archiving.