Publish independent Route contribution evidence review

This commit is contained in:
2026-09-24 02:46:54 -07:00
parent 2f9ce68a9e
commit 960b75a9ea
2 changed files with 48 additions and 0 deletions
+3
View File
@@ -18,6 +18,9 @@ owns current build and release coordination.
The Route test helper's one-line qualification correction is included; the
earlier peer compile failed before tests ran. The candidate's 66 package
documents/examples pass [closure and source-identity verification](reviews/prerelease37-windows-documents-20260924.md).
The separate e68a839 Route contribution's raw test/lint evidence passed an
[independent evidence review](reviews/prerelease37-route-evidence-20260924.md);
its results do not establish combined-candidate execution.
## Current integrated candidate and evidence
@@ -0,0 +1,45 @@
# Issue 37 Route contribution evidence review
2026-09-24. Independent read-only review of evidence publication `93d2a6f9af40061bf92cd81ba33d5648d0181cb1`, tested contribution `e68a83959900e556768d8470640e28a7df7d8cb8`, and integrated candidate `897d9f049406d5be597543eb0a6755b796259302`. Published by the parent integration owner after reading the report. Live candidate corrections and release status remain in issue37.
## Result
No material contradiction found in the reported narrow gate results, retained failure, or regression-module identity. The evidence supports the contribution-scoped claims below. It does not establish execution of the integrated candidate, full workspace acceptance, GUI acceptance, or packaged runtime acceptance.
The five final gates consistently record e68a839, unchanged inputs, a 756-file map, Windows GNU LLVM toolchain, Rust/Cargo 1.98.1 and incremental compilation disabled. Tests/Clippy record one build job; formatting records CARGO_BUILD_JOBS as null. If reproducing environment verbatim, retain that small formatting distinction rather than claiming every environment field is identical.
| Captured run | Independently inspected raw evidence |
| --- | --- |
| route-regressions-3 | Exit 0; stdout: 31 passed, 0 failed; all six `group::route_lifetime` tests individually say ok. |
| allocation-regressions-2 | Exit 0; stdout: 51 passed, 0 failed; all three Connection and six Route regression cases individually say ok. |
| route-attachments-1 | Exit 0; stdout: 28 passed, 0 failed. |
| route-format-2 | Exact `cargo fmt --all -- --check`; exit 0; empty stdout/stderr. |
| route-lint-1 | Exact `cargo clippy -p oa-desktop --all-targets --message-format=json -- -D warnings`; exit 0; stderr reports completed dev profile. |
Filters overlap; these counts are not a unique-test total. The retained `route-regressions-2` raw stream contains E0425, exit 101, no test-result summary and compile-failure stderr. Its captured HEAD is c655c5ee, while its source proof names 6f247af; the admission note explicitly explains this distinction. Earlier route-format-1 is a separate successful 6f247af run. Seven completed runs and seven proof receipts are present; the separately described interrupted run is not in this archive or counted as a success.
## Source identity and integration boundary
Read-only local Git comparison found no difference in either regression module between e68a839 and 897d. Both final working-tree module hashes also equal their entries in the final captured input map:
- `crates/oa-desktop/src/app/allocation_drag_tests/group/connection_lifetime.rs`: `d5242e2c530e433649a061af6f0cf09e0532cf3922705b509c3a159b04715cf4`.
- `crates/oa-desktop/src/app/allocation_drag_tests/group/route_lifetime.rs`: `7110c79b3fe8c2679e0ec6bd59bbe307399359e89d2d9d6120c546de756a580b`.
The 6f247af → e68a839 crates diff is exactly the fixture correction from `project_view(...)` to `projection::project_view(...)` at route_lifetime.rs:40. Final candidate registration is present in allocation_drag_tests/group.rs:5-8. The captured Connection/Route early guards remain in app.rs:1353-1374 ahead of shared retirement at app.rs:1577, including both live-context/Connection-session checks and Route selected-edge checks.
The surrounding integrated application differs substantially from the tested contribution, including saved-Scenario handling in `route_context()`. Identical regression modules and preserved guard text do not transfer these execution results to the complete integrated source. Preserve the note's existing scope restriction in final VALIDATION.md.
## Evidence actually read and integrity checks
Used Gitea MCP at the exact publication commit to read:
- `harness/prerelease37-connection-admission-20260924.md`.
- `harness/prerelease37-connection-evidence-20260924/route-checks/README.md`, `manifest.json`, `files.json`, and directory inventory.
After reading those manifests, downloaded only their named evidence.zip from the exact commit, with a 6 MiB cap. Received 6,094,831 bytes, SHA-256 `501f01c7279bc9d719151abe9b729769dd0078678c702448671356e91965015f`, matching both files.json and the admission note.
An independently authored local PowerShell inspector verified all 49 ZIP member names/counts, lengths and SHA-256 pins against the MCP-read manifest, rejecting duplicates. It read every run's raw result/proof/stdout/stderr; compared commands, exit, HEAD, input count, unchanged-input flag, proof commit, proof input-map hash and result-receipt hash to the manifest/receipts. All five final gates name the same final input map `e8d0acc2de87827cc220143aafcd4963122cfa6529cc679f886f9870954f30f5`, whose decoded inventory has 756 entries. The two historical gates use the separate older map.
The source-proof receipts assert exact captured Git-blob comparison; this review checked their pinned identity/consistency and independently compared the two regression modules, but did not repeat the entire 756-input Git-blob comparison or the published verifier/selfcheck. Nested source capsules and archived helper bytes were hash-checked as members, not executed. No downloaded program/helper, Cargo build/test, GUI launch, remote mutation, fetch, or candidate mutation was performed.
Local review inputs: `route-manifest.json`, `route-evidence.zip`, `inspect-route.ps1` alongside this report. Their presence is review scratch evidence, not a published handoff.