7.6 KiB
State transition bends: source, gates and Windows native evidence
This archive records branch-specific work at b8685416f56e53e30bd11ecf93a54cea10ac8458, branch codex/state-bends-ytr83r-20260923. It is not an integrated release or Linux acceptance; Linux work remains paused. Original receipts and preparation documents are preserved byte-for-byte, including historical pending wording and failed outcomes. The results below and finalized reviews explain the completed scope.
Results and limits
Frozen final-02 gates recorded 1,835 passed tests, zero failures and seven ignored across 61 suites; strict workspace/all-target Clippy with warnings denied and desktop/two evaluator-example builds passed. See independent gate review and raw final summary. The three candidate-built executables were preserved before native packaging; raw relocation receipts distinguish historical shared-target locations from admitted preserved artifact hashes.
The parent operator exercised Windows State view03 Activate: Actions → Add bend; drag the authored square; square context menu → Remove bend; Clear bends; Undo/Redo for each; untouched reference view04; Save, normal close and fresh-process reopen. All17 complete-Project comparisons passed. There are25 capture metadata records and26 images (including a16×16 chooser overlay); repeated-state image bytes are retained. All three processes closed with exit0, empty stdout/stderr and unchanged immutable input checks. See operator report, independent native review, and raw runs.
The baseline is the small original attachment fixture, with IDs prefixed ea202922-0033-4000-8000-. Only view200 / transition41 waypoint vectors vary:
| State | Exact waypoints |
|---|---|
| B | [(500,310)] |
| N1 | [(500,310),(526,310)] |
| N2 | [(500,310),(520,370)] |
| N3 | [] |
Before native input, candidate core geometry plus independent Python arithmetic selected Add's midpoint(526,310), insertion index1. The saved-source evaluator compared entire typed Projects with these predeclared states. SourceRight0.5, targetLeft0.6, manual label fraction0.5/offset(8,-16), semantics, placements and other saved views remained exact. Run2 reopened cleared N3. An additive predeclared Run3 supplement reopened the actual saved nonempty N2 checkpoint, displayed both authored squares, and saved with exact source bytes preserved. Generated route elbows/neighbor geometry can settle without changing stored Project presentation; equality claims concern the full stored Project.
Native evidence does not cover a held intermediate preview, mid-drag cancellation, direct segment-drag insertion, off-grid return, outside-release cancellation, all projection/symbol cases, exact camera persistence or general small-window/accessibility/label readability. Real-canvas automated tests supply separate preview/cancellation/direct-insertion coverage. Reference labels were crowded; no general layout-quality acceptance is inferred. Review source, acceptance pan repair and integration boundary review alongside the final gate results.
Nonempty-reopen supplement binding
Supplement plan was declared before Run3 and identifies an exact copy of Run1 checkpoint04-move. The operator opened that new path through the native chooser. Checkpoint16 uses the frozen checker with explicit --actual and compares the whole saved Project to existing N2. Its immutable saved source matches the predeclared input SHA170c3e59e189223838247b37a8153472ea27444f8d216e25f78af591be5c6fb4. The mutable supplemental copy is omitted; immutable checkpoints preserve the same bytes.
The frozen Run3 launch/close and capture helper's default working-source hashes still refer to original N3. They do not identify the opened supplemental file. The additive plan, observed chooser path suffix (the left prefix is clipped) and explicit-actual checkpoint provide that N2 binding. Capture23 includes both the chooser and a small provider overlay; capture24 shows the saved reopened route. A failed initial typing/set_value attempt is disclosed in the raw observation/report; coordinate refocus then succeeded. No old input manifest or expected catalog was rewritten.
Preserved failures and source coverage
gates/failed-targeted-02 preserves an earlier dirty diagnostic failure. Exact historical contents of two untracked tests were not retained: app/state_bend_tests.rs SHA97dd304bf6a1fffe14d4f0780ea66f6df9f14120fbb16a4869f02c75063f7e58 and behavior/state_bend_tests.rs SHA77fe84cfd420de45e8e2bc389b776e49662fb83e893f8b9d5d9fced5b8468d47. Their raw failed receipts/logs are included, but those source versions remain explicitly unavailable; no substitute or guessed reconstruction is presented.
gates/failed-final-01 preserves the identified26979c1 failure, including the read-only pan regression and overconstrained self-loop source assertion. Final-02 at b868541 repairs those issues and independently passes. Both clean snapshots have complete captured-source maps. All8,259 available captured-source records across the three snapshots match available Git bytes exactly; no CRLF override was required. For the dirty targeted snapshot, matching later Git bytes serve solely as byte storage, not a claim that the later commit was tested then.
source-map.json maps every available captured source record to exact Git bytes. The verification schema supports raw overrides, but this archive includes none: all8,259 mapped records match their Git blobs exactly. No wholesale copied source tree is required. Historical source gaps and external reference inventory enumerate the exceptions. Apart from the two approved historical test gaps, unavailable references are intentionally omitted executables/runtime. Final successful and native nonbinary evidence has no unresolved reference.
Verify on another machine
Fetch the source repository with commits b8685416f56e53e30bd11ecf93a54cea10ac8458,26979c109384add3e2a4bfd0092b8da1464e37a2 and53d01a3d4451b94792986f2dfac56f6f04e111a4. From this archive run:
python verify-build.py verify --archive . --repo /path/to/ARCADIA-Studio
The verifier uses archive-relative paths and Git objects, never original machine paths. archive-manifest.json maps original absolute paths to copied relative paths; embedded absolute paths in raw receipts are historical. It verifies recorded byte lengths/hashes, source maps and receipt-reference availability, and reports omitted/unavailable references honestly. Python3 and Git are sufficient for integrity verification; no executable package is required. Do not interpret its integrity result as re-executed tests or native behavior acceptance.
Reproduction of native behavior is separate: build the three tools at the source pin with the recorded environment, preserve artifacts, adapt machine-local paths in copies of the recipes, create a fresh stage/profile, derive expected geometry before input, then follow the plan. Recipes require Windows and the admitted runtime for the recorded build. Do not edit this archive's frozen manifests/expected states to rerun it. Binary/runtime/PDB files, build directories, private profiles and mutable working files are intentionally omitted; immutable saved checkpoints and package hashes remain included.