sorokin.oleg.edu/ARCADIA-Studio
1
0
Fork 0
Code Issues 37 Pull Requests 2 Packages Projects Releases 9 Wiki Activity

Archive exact a31b40d Windows and OA native acceptance evidence

Browse Source
This commit is contained in:
sorokin.oleg.edu
2026-09-12 03:54:36 -07:00
parent f52d81c1b9
commit bc8bc4f5a1
88 changed files with 37915 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
harness/release-behavior-canvas-a31b40d-native-20260911.md
+44
View File
@@ -0,0 +1,44 @@
# Canvas Windows gates and bounded native acceptance — a31b40d
Preserved on 2026-09-12 by `/root/canvas_evidence_archive`. This report records completed 2026-09-11 evidence for source **a31b40d0f8551ed87c8cd66250b9afb4809b3a79**, not the current HEAD or subsequent allocation changes. No tests, checker, build, or native input were rerun during preservation. Computer Use remains paused.
## Immutable evidence and verification
The [archive manifest](experiments/canvas-a31b40d-windows-native-20260911/archive-manifest.json) indexes **84 original files / 3,030,048 payload bytes**, with original paths, sizes, SHA256 identities and original modification timestamps. Each destination was copied without text conversion and compared against the source hash before and after copying. The manifest is 47,655 bytes, SHA256 `a89377d1a0635375e87d9565a220864a4cd0ac071f1425db00cd866133981e46`. Originals retain their historical absolute paths and status text; these were not rewritten to imply later acceptance.
All three [gate input inventories and closed receipts](experiments/canvas-a31b40d-windows-native-20260911/canvas-union-20260911/) identify a31b40d, contain the same 518 unique input paths with matching sizes/hashes, report exit 0, and record no changed inputs. Their inventories and results match the [pinned build manifest](experiments/canvas-a31b40d-windows-native-20260911/native-qa/studio-a31b40d/manifest.json); all three log hashes match that manifest. This preservation verifies recorded identity consistency, not a fresh comparison with Git blobs or current working sources.
| Closed Windows gate | Actual recorded result |
| --- | --- |
| `cargo test --locked -p oa-desktop --bin arcadia-studio -- --test-threads=1` | 729 passed, zero failed/ignored/measured/filtered; 294.66 seconds test execution |
| `cargo clippy --locked -p oa-desktop --all-targets -- -D warnings` | Exit 0 |
| `cargo build --locked -p oa-desktop --bin arcadia-studio` | Exit 0 |
The build records Rust/Cargo 1.98.1 Windows gnullvm, LLVM-MinGW 20260908 and one build job. [Launch identity](experiments/canvas-a31b40d-windows-native-20260911/native-qa/canvas-20260911/launch.json) matches the pinned executable manifest: 113,005,056 bytes, SHA256 `89f9746c42888e189ea58723a0e64c8af610d97c6e73d1f31828fcd94dae8023`. Executable/DLL payloads, private profiles, temporary files, mutable trial working sources and trial stdout/stderr are excluded. Process closure is unknown; no claim is made that these live logs were closed or clean. The archive preserves [preparation](experiments/canvas-a31b40d-windows-native-20260911/native-qa/canvas-20260911/preparation.json), [checkpoint helper](experiments/canvas-a31b40d-windows-native-20260911/canvas-checkpoint.py), and [session helper](experiments/canvas-a31b40d-windows-native-20260911/canvas-native-session.py) as historical reproducibility material; their original workspace-relative placement and pinned prerequisites are required for execution.
The [independent cleanup review](experiments/canvas-a31b40d-windows-native-20260911/canvas-cleanup-review.md) and [source handoff](experiments/canvas-a31b40d-windows-native-20260911/canvas-obsolete-paths-handoff.md) preserve their narrower source-only conclusions. Their original pending-gate language predates the receipts above.
## Native behavior actually recorded
[Sixteen original screenshots and matching observations](experiments/canvas-a31b40d-windows-native-20260911/native-qa/canvas-20260911/evidence/native-observations.json) document the actual pinned GUI. The supported minimum was 1100×700 client / 1102×732 captured window. All 13 palette tools fitted. Fit, Tab after Fit, F6/Right keyboard panning and blank-area left-drag panning were observed. These are recorded native observations, not conclusions inferred from source tests.
| OA creation | Recorded destination and content |
| --- | --- |
| Action | OA native action, OA Inspect ticket, OA A inspector, before OA Root entry |
| State | OA native state, OA Ready, OA A inspector, before OA Inner anchor in OA Inner chosen operand |
| Reuse | OA native reuse at that same chosen operand; OA B verification with explicit B inspector → A inspector and B recorder → A recorder bindings |
| ALT fragment | OA native alternative before OA Root exit; guards `ticket.priority > 2` and `ticket.priority <= 2` |
For each creation, Save, one Undo/Save and one Redo/Save have independent [preserved checkpoints](experiments/canvas-a31b40d-windows-native-20260911/native-qa/canvas-20260911/checkpoints/) named `oa-{action,state,reuse,fragment}-{saved,undone,redone}` and matching `evidence/*-check.json` receipts. Preservation checked all 36 checkpoint files against recorded sizes/hashes and all **20 recorded oracle results**: 12 full-model/two-source checks and eight exact Project/SourceBundle comparisons, each exit 0 with PASS output and empty stderr. Every receipt identifies checker SHA256 `407318d5c7cae6b237f261a21b0fda55f6fef655cb9dfcfa3153a7e372638498`.
Byte comparisons independently confirmed every Redo matches its original saved three-file snapshot and every Undo matches the preceding snapshot; action Undo matches the prepared scenario baseline hashes. The recorded semantic checker results establish the full-model comparisons; preservation did not rerun semantic parsing. Fragment plus its two operand identities were one recorded history step. Final fragment source SHA256 is `335347b279fa8cbd1bc74b0db83644c47f05b47547e508a80412d4430ef21a7c`; included rationale remains `82df210f22f6aa336a3e7e00240977d671c7c3923db7bc58fb29e6d98b67ef49`.
Observed forms kept their clicked destinations and initially focused Name. Clicking Details with the pending action form preserved that form without opening Details. Escape in the completed fragment draft removed input focus while preserving both guards, destination and draft. Reuse content scrolled separately with Create/Cancel remaining reachable.
## Acceptance boundary
**Reopen is unknown and unverified:** the user stopped Computer Use with physical Escape during the final Open-dialog double-click. No later native action or successful reopen is claimed. PID/window identifiers are historical only.
Current SA, process editing, source-child recovery/cancel, caller return and reorder refusal acceptance remain pending. Focused wheel zoom still needs native verification; unfocused wheel input did not pan. No below-minimum palette overflow or physical held-middle/Space gesture acceptance is claimed. Preparation-only process/source logs do not establish native editing acceptance. The [earlier 3c9e6ee native report](release-scenario-native-3c9e6ee-20260911.md) and [review](release-scenario-native-3c9e6ee-review-20260911.md) retain their narrower, earlier-build scope; they do not establish current SA acceptance. See the [pause record](release-user-pause-20260911.md) for interruption context.
No archive discrepancy was found. Root verified all 85 archive files byte-for-byte against their staged Git blobs and independently matched all 518 input records to the tested a31b40d Git blobs. Initial folder staging omitted six globally ignored log files; explicit archive staging included them before the successful verification. No source bytes or historical receipts were changed. This report is evidence preservation, not a release or current-source acceptance claim.