Admit targeted Windows State picker correction evidence

2026-09-24 03:44:58 -07:00
parent a5496cd41d
commit 29e4f67206
2 changed files with 28 additions and 1 deletions
@@ -25,7 +25,9 @@ The full Windows test run at `8ddfadf4` failed a State picker fixture whose
 canvas-relative coordinates hit the added selection toolbar. The current
 source includes peer `a9cd364e`'s measured picker slot and capture assertion;
 the [independent source review](reviews/prerelease37-state-picker-20260924.md)
-retains its bounded assumptions. Corrected combined gates remain pending.
+retains its bounded assumptions. [Reviewed targeted evidence](reviews/prerelease37-state-picker-evidence-20260924.md)
+at `a9cd364e` passes formatting and all26 State tests, including the original
+failure. Corrected combined gates and packaged acceptance remain separate.

 ## Current integrated candidate and evidence

@@ -0,0 +1,25 @@
+# State picker targeted evidence admission
+
+2026-09-24. Reviewed publication `be845878b4d5ca0f987941672d0a0a0756fb4e7a`, `harness/prerelease37-windows-gates-20260924/state-picker-a9cd`, for exact tested source `a9cd364eacbbffa0693bab05b5a2b65f9d105535`.
+
+## Result
+
+No material discrepancy found in the narrow reported results. Raw command receipts and logs support formatting exit0 and **26 passed, zero failed, zero ignored, 1,509 filtered out** for `app::state_diagram_tests::`. Both receipts record the exact a9cd source, exit0, failure:null and unchanged inputs. Each before/after map contains 3,709 entries; all four map byte hashes agree (`6adb9b24d2de6a64addae65bbc443789b377b161bdfce04c74568ab4065d1781`).
+
+The raw stdout individually records the previously failing `app::state_diagram_tests::picker_escape::mounted_state_missing_picker_escape_preserves_exact_document_then_reopens_and_places ... ok`. The original combined 8dd final-03 failure remains preserved in the earlier attempts archive and is not relabeled a pass by this narrower successor result.
+
+Commands are exactly `cargo fmt --all -- --check` and `cargo test --locked --offline -p oa-desktop --bin arcadia-studio app::state_diagram_tests:: -- --test-threads=1`. Raw stderr identifies the current source checkout, a newly completed unoptimized test-profile compilation and execution of the desktop test harness. This is execution of the filtered desktop test suite, not a launched Studio product/native acceptance workflow.
+
+## Runner and refusal boundaries
+
+Receipts bind the previously reviewed v2 run.py hash `698ed7182726c77fa4a1c85fc6dd08187ed20689941f57e749e3f0bc845f3252` and the packaged environment receipt `8c4d049cee43ba723c68a5c636dbc3583144053912475cb3db2e6f6e88d6a79f`. Captured environment identifies Windows GNU LLVM, jobs1, incremental0 and the owner's explicit target directory. The targeted wrapper inherits and verifies the frozen prior toolchain, refreshes candidate/ancestor Cargo config observations, adds its own recipe hash, then uses the unchanged checked command runner.
+
+`admission-refusal/admission-failure.txt` retains the earlier reported pre-command refusal: CARGO_TARGET_DIR had forward-slash spelling differing from the admitted backslash string. It explicitly says no commands launched. The retained wrapper performs environment verification before either command; its later recipe differs only in selecting a fresh `-02` output directory. No admission check was removed. The original file is a concise refusal note, not a captured Python traceback or command-result receipt, so the exact root-cause explanation is a publisher-recorded statement corroborated by recipe ordering, rather than independently reproduced here. It is not a product-test failure or a passed gate.
+
+Unlike full final-gate inventory runs, this targeted command supplies no test-artifact inventory option or Cargo artifact JSON stream. It does not independently preserve/hash the executed test binary or produce a full native test-name inventory. That limitation is consistent with its targeted source/command/log scope; do not substitute these records for the final packaged binary or full workspace evidence.
+
+## Evidence and acceptance limits
+
+Read the manifest, README and directory inventory through Gitea MCP before the bounded 1,200,000-byte download. Received exactly 1,152,040 bytes, SHA-256 `08fcb8e316423b5ab60afd84dd7a103dcca05da0dd1f22aa9cabfe6fc58b7599`. Independently checked exact ZIP member inventory, all 23 member lengths/hashes and duplicate-name absence. Read raw format/test results, stdout/stderr, summary, source-map counts/byte equality, runner/environment bindings, targeted recipe and original refusal note/recipe. Local inspector: `inspect_picker_targeted.py`; it only reads/hashes archive data.
+
+Admission is limited to targeted correction evidence at a9cd. The integrated 715036e freeze and active final-04 require their own source-attributed complete gates; identical product code plus harness-document additions does not imply those executions passed. Strict workspace Clippy, desktop/server product builds, packaging, clean-machine/native GUI and server persistence acceptance are not established here. Source/compiled-code authenticity is not independently proven by internally consistent receipts. No archived code, build, test executable or product was run; no candidate or Gitea write occurred.