Files
ARCADIA-Studio/harness/experiments/allocation-drawing-301126f-20260923
..

Direct Allocation drawing — Windows evidence

Verified source: 301126f6098286bf440c1bd1b1ed685753984723, branch codex/allocation-drawing-ytr83r-20260923. It adds the Allocation tool to eligible architecture projections through the existing guarded two-endpoint drawing workflow. See source scope and decisions and Gitea issue30.

Windows gates passed: 1,222 desktop tests, 0 failures, 6 existing ignored, strict desktop all-target Clippy, matching GUI and snapshot evaluator builds. All six new Allocation tests passed. The six gate steps recorded unchanged source inputs. Unchanged core/server workspace evidence is reused with separate attribution to b868541/evidence d6aabb5; this is not a new workspace suite.

Root exercised OA Rescue team → Assess incident and SA Rescue system → Evaluate report using the right-side Allocation tool. Creation, Undo/Redo, occupied OA target refusal and Escape cancellation, Save and fresh reopen produced nine exact complete Project/source-bundle checkpoints. The two relation UUIDs were bound once; every other field was fixed in advance. Both normal closes were observed, with process exit0, empty GUI logs and complete saved bundle/identity continuity. The accepted trial contains26 capture records. Parent observations, independent final review, and explicit coverage distinguish live actions, archived review and automated comparisons.

The native recipe describes the fixture, commands and oracle. Expected fixture-completeness warnings are retained; the final fixture still lacks behavior context. Other projections and stale/draft/invalid-target cases have automated/source coverage, not claimed native coverage. This is branch-specific Windows evidence, not combined integration, release, whole-project, accessibility or every-window-size acceptance. Linux remains user-paused.

Verify the portable evidence

Fetch source commits301126f, cf37c36 and17f1f89 in a repository, then run from this directory with Python3:

python verify-build.py verify --archive . --repo /path/to/ARCADIA-Studio

archive-manifest.json maps unchanged copied originals; source-map.json maps captured source bytes to their exact Git commits (or explicit byte-matching overrides). Raw records retain historical machine paths, but the portable verifier resolves copied evidence and Git objects without those paths. external-reference-inventory.json explicitly classifies omitted binary/runtime and private/mutable references. This verifies evidence bytes and provenance; it does not reproduce omitted executable bytes or re-execute the product. The separately reused source converter remains attributed to17f1f89 and identical core/Cargo objects.

Collection contains604 unchanged originals (56,849,882 bytes) and11,036 source records, all matched exactly to Git with no reconstructed overrides or historical source gaps. The accepted trial has28 image files for26 capture records. All4,457 remaining external references are explicitly omitted binary/runtime records (1,080 unique byte/hash pairs); no unresolved nonbinary reference was admitted.

Failed gate gates/failed-final-01 remains a compilation failure at cf37c36, with no executed tests. partial-interrupted-native/ preserves the first trial's four accepted OA checkpoints separately; its missing close and unknown exit cause are explained in the interruption note. It is not successful close/reopen or crash evidence. The fresh accepted trial is under native/.

This archive contains evidence and reproduction recipes, not a runnable release. Integration/release ownership remains issue27; no other session can consume unpublished local files.