Files
ARCADIA-Studio/harness/experiments/traceability-view-17f1f89-20260923/README.md
T

34 lines
5.5 KiB
Markdown

# Traceability view: Windows verification and handoff
Source: [17f1f89fef4a6bf520f4b09a2116548b967d030e](http://217.177.34.108:3000/sorokin.oleg.edu/ARCADIA-Studio/src/commit/17f1f89fef4a6bf520f4b09a2116548b967d030e), branch `codex/traceability-view-ytr83r-20260923`, issue #31. This is branch-specific evidence, not an integrated release. Linux remains user-paused. Original records retain historical preparation/pending wording; this report and finalized reviews describe the completed outcome.
The change explains mixed OA/SA traceability views, counts actual visible correspondence lines, exposes existing review/visibility/placement actions, and defaults the existing-element chooser to all compatible types in these views. The Show action preserves its captured mixed diagram. Existing semantic traces, ordinary relationships and saved filters keep their established behavior. Design and implementation details are in the source commit's `harness/traceability-view-20260923.md`.
## Verified outcome
The [final gate](gates/final/summary.json) passed 1,216 desktop tests, with zero failures and six existing ignored tests; strict desktop all-target Clippy and matching desktop/evaluator builds passed. Inputs remained unchanged. [Independent gate review](reviews/traceability-view-gate-review-20260923.md) confirms the records. Core, server and Cargo trees equal b8685416f56e53e30bd11ecf93a54cea10ac8458; their prior workspace evidence is published at d6aabb5dec82900ebfac1183abb4036ca06cc0e5. This is not a new workspace or Linux test claim.
The Windows trial exercised the explanation and line count, layer-trace review, missing-target visibility reasons, immediate discovery of an opposite-layer existing element, placement, trace properties, Undo/Redo, Save and fresh-process reopening. Five complete-Project comparisons passed: baseline B, placement P, Undo B, Redo P and reopen P. The only Project change is one node in the selected view; semantic data, other placements, filters and the reference view remain exact.
The predeclared target was (500,350), with at most one world unit error per axis. After full equality passed, the new UUID `4ec01fff-2024-40cf-b01d-e7d3cfbb1cd3` and position (499.8161,349.93808) were bound once. Redo/reopen require those exact values. The visible correspondence count changes from one to two, with ordinary OA and SA exchanges retained. See [independent native review](reviews/traceability-view-native-review-20260923.md), [checkpoints](native/supplement-source-bundle-v2/checkpoints), and [captures](native/runs).
Both processes closed normally with exit 0, empty process logs and unchanged immutable inputs. There are 20 capture metadata records and 22 images (21 distinct byte sequences). Complete eight-file bundle inventories were checked after the first close, before fresh launch and after the second close. These are observations at those times, not continuous or atomic snapshot guarantees.
## Bundle reader and retained failures
Save As creates included source files. The original single-file evaluator cannot load these directly. A [separately reviewed converter](native/supplement-source-bundle-v2) reads copied bundle bytes through the existing core loader/compiler, validates the Project, formats self-contained source and verifies full Project roundtrip equality. The unchanged evaluator and frozen B/P oracle then compare that source. Only accepted outer bundle receipts establish checkpoint acceptance. The converter's new auxiliary build used a Cargo-selected cached core artifact; it is not attributed retroactively to the product gate. [Build review](reviews/traceability-view-bundle-build-review-20260923.md) records that provenance.
The fixture retains 15 engineering completeness warnings: missing system, unallocated leaves, empty port contracts, unreviewed traces and missing behavior context. They are recorded and inspected, distinct from loader errors and process stderr. The fixture is not warning-free.
Failed final-01 compilation and final-02 test results are preserved with their exact source versions. The first native attempt refused a Save As entry name that collided with an included module; after normal operator close, its observer failed because no expected source existed. It is not counted as a passed run or inferred crash. The rejected v1 bundle proposal is retained as unexecuted preparation. No expected Project was changed to accommodate observed output.
## Portable verification
Fetch the source history containing 17f1f89fef4a6bf520f4b09a2116548b967d030e and its ancestors 4fbdf6f385fa268d41074459f7a4728e5e1e9b11 and bfc259854f9f7b8cdd40b8b9a043340033749b75. From this archive:
python verify-build.py verify --archive . --repo /path/to/ARCADIA-Studio
The archive contains 470 original files (59,481,613 bytes) and maps all 8,274 captured source records to exact Git bytes, with no source gaps or reconstructed overrides. Raw absolute paths describe the originating machine; verification uses archive-relative files and Git objects. Executables, runtimes and compiler libraries are omitted with explicit hash references. This is portable evidence, not a runnable package or verification of omitted binary bytes.
The trial supports the exercised workflow, not a novice-user study, every trace/filter combination, narrow-window accessibility, or direct native proof of graphical relation-endpoint trace omission. Automated/source reviews supply separately stated coverage. Integration and release ownership remain with Codex-956atf under #27; the owner must verify the combined result before release.