Publish packaged Windows native acceptance evidence

This commit is contained in:
2026-09-24 15:25:47 -07:00
parent c82eca1eca
commit 8a82c0da1f
5 changed files with 5234 additions and 0 deletions
+1
View File
@@ -89,3 +89,4 @@ harness/prerelease37-windows-gates-20260924/state-picker-a9cd/** -text -whitespa
harness/prerelease37-windows-gates-20260924/final04-tests/** -text -whitespace
harness/prerelease37-windows-gates-20260924/final04-complete/** -text -whitespace
harness/prerelease37-windows-gates-20260924/package-server/** -text -whitespace
harness/prerelease37-windows-gates-20260924/native-gui/** -text -whitespace
@@ -0,0 +1,20 @@
# Packaged Windows GUI acceptance
The packaged Home/example/guide, State Clear bends/Undo/Redo, picker dismissal, immediate attachment cancellation, Save and fresh-process reopen workflows passed on Windows. This is bounded prerelease acceptance, not completion of the full application specification. The package is the unchanged release ZIP `5554e32533e3cb2c33fb4143a5a0902733d5a8d47577303cb2842d268e7bc76d` (34,168,661 bytes), tested by launching its extracted executable with a private profile, restricted PATH and unrelated working directory.
The evidence archive is `native-gui-715036e.zip`, SHA-256 `f18c21d262ca78bc9c2d8a339c3389fa31c50a81187cc53cc472f3253cda2815` (10,357,627 bytes, 351 members). Extract it to inspect `reviews/native-observations.md`, the predeclared workflow in `recipes/PLAN.md`, and raw results under `native/`. Absolute paths in raw receipts identify this builder machine; peers consume the archived bytes and published commit references.
| Check | Result |
| --- | --- |
| Complete saved Project comparisons | Ten passes: B/C/B/C/P/P/P/P/C/P |
| Clear bends, Undo, Redo | Authored square disappears/restores/disappears; exact saved C/B/C |
| Show-unplaced and Add existing type popups | First Escape dismisses only popup; dialog query/filter retained; outside dismissal and second Escape cancellation work |
| Source/target attachment cancellation | Immediate Escape restores Select/move; exact P and saved-source bytes retained |
| Fresh-process reopening | Exact C and P; saved-source bytes match earlier accepted checkpoints |
| Two launched GUI runs | Normal Alt+F4 observed; exit0, empty logs, unchanged package |
The original restart harness refused before launching run02 because its reused working directory contained cache files. The original failed attempt and file inventory are preserved in `reviews/cwd-incident.json`. Independently reviewed launcher v2 used a fresh working directory for run03 with the same package/profile/environment. Cache attribution remains unproven; no product crash or silent failed-test replacement is claimed.
State label congestion (#36) remains visible. This trial does not establish broad accessibility/DPI/performance, all OA/SA workflows, held-drag/reconnect coverage, physically disconnected operation, or Linux support. Independent downloaded-package QA and release publication are separate coordination steps in Gitea #37.
Source `715036e132677dd2bc2fb37ed316d93dc9987aec`. Every launched GUI run has a terminal close receipt. Failed prelaunch run directories remain identified separately, with original cwd cache metadata/digests in the incident. Original staging manifest SHA-256 `c90371bee1c2b2ae850b5d8e64ab9f1368f541ef51104e3b888a721e05d49c13` is preserved verbatim in the ZIP. All saved-source snapshots, complete typed expected Projects, actual comparator outputs, captures and metadata, launch/close logs and immutable evaluator recipes are included. Individual failures remain labelled as recorded; mechanical acceptance is reported in manifest.json, not inferred from screenshots. Package/helper provenance is bound to `c82eca1eca204dd28994f5c2e8595d8acd2e148e` package-server evidence. Executables, private profile and unrelated-cwd contents are omitted. No broader GUI or release completeness is claimed. Each ZIP member was reread for exact bytes, SHA-256 and CRC. Normal-close cause and visual acceptance need the parent review.
@@ -0,0 +1,31 @@
# Independent native acceptance review — 2026-09-24
**Ready to publish as bounded Windows prerelease native evidence; no new actionable findings.** The recorded saved-model, interaction, reopen and process evidence supports the current `native-gui/README.md` claims within its stated limits. State label congestion remains visible and explicitly unaccepted as fixed.
This review independently read the completed archive, collector, provenance, observations and representative full-size screenshots. Read-only Python audits recomputed archive/content checks and comparisons; no collector, evaluator, GUI, test process, compiler, network operation, model/source edit or recipe change was performed.
## Verified outcomes
- **Archive:** `native-gui-715036e.zip`, **10,357,627 bytes / 351 members**, SHA-256 `f18c21d262ca78bc9c2d8a339c3389fa31c50a81187cc53cc472f3253cda2815`. All members independently passed exact ordered/unique membership, byte length, SHA-256 and CRC checks, plus ZIP CRC readback. The collector's archived bytes equal the actual collector, SHA-256 `336d012656a6848ffdf6d9de9017b540b4d5d847fd21997f718d5726b37bb9f2`.
- **Independent B/C/P oracles:** B/P baseline source hashes match the predeclared fixtures. Reconstructed C from a deep copy of complete B by removing only Activate's unique `(500,310)` waypoint; reconstructed P by removing only the specified view placement of Shutdown complete. Each reconstructed full object equals its frozen expectation, and all typed-validation outputs equal their expectations. The stage manifest binds these expected bytes before either admitted launch.
- **All ten checkpoints:** independently compared complete actual Projects with strict canonical JSON equality in sequence **B/C/B/C/P/P/P/P/C/P**. Every checkpoint is accepted with matching pin/manifest, unchanged original inventory, exact copied-source fingerprints, successful bound converter/snapshot receipts and empty helper stderr. All five saved P inventories equal the first saved P at checkpoint04; all three C inventories are equal, including reopen08. Final working State/P bundles exactly match reopen08/09. No expected model was reconstructed from an edited GUI result.
- **Provenance:** 98 archived immutable file bindings verified directly. Complete package membership/identities agree with the separately audited package-server archive `cd303a…`; its actual hash matches the native provenance link. Source inventory agrees with final-04 pin `715036e132677dd2bc2fb37ed316d93dc9987aec`. GUI `2c03bd6c…`, snapshot `6a1f438a…`, runtime `aa282fa4…`, and supplemental converter `c318ea38…` agree with their exact gate/helper identities. The package remains ZIP `5554e325…`, not a substituted combined-build artifact.
- **Actual process attribution:** run01 PID6516 and run03 PID2044 each bind the admitted manifest, extracted GUI, argv and reviewed launcher, and end exit0 with empty stdout/stderr and package_unchanged. Their child environments/private-profile paths are exactly equal; run03 uses the declared fresh cwd. Its old-cwd inventory equals the incident before/after. Run02 is preserved distinctly as the empty prelaunch-refusal directory/incident, not a successful run. The parent's recorded Alt+F4 sequences supply normal-close cause; process exit alone was not treated as that observation.
- **Visual evidence:** all 44 capture metadata records identify the admitted extracted GUI and include a full-size image. Closed local capture bytes equal archived bytes. Viewed **27 full-size images**: 03–06, 08, 10, 12, 14, 19, 22, 26–27, 29–38, and 40–44. They corroborate guide/example/return, the authored square disappearing/restoring/disappearing with Clear/Undo/Redo, source/target armed-tool cancellation restoring Select/move, Show-unplaced offering Shutdown complete and dismissing, Dispatch query/filter retention while the popup closes, distinct dialog cancellation and recovery, saved recent-model entries, and reopened cleared State/P views. Final capture44 shows all changes saved. The viewed screenshots and observation narrative are consistent; screenshots are not used as substitutes for full saved-state comparison or proof of unrecorded gestures.
## Limits and publication claim
The current README correctly describes a bounded native result. It does not close #36 or establish broad readability, accessibility, DPI, performance, all OA/SA or Traceability workflows, held-drag/reconnect behavior, physically disconnected networking, Linux behavior, or full application completion. Cache-writer attribution remains unproven. The recovery retained the original failure and used a reviewed new launcher rather than editing an admitted recipe. No further run is needed for this bounded evidence review; independent downloaded-package QA and release coordination remain separate parent/peer work.
Executables, private profile and unrelated-cwd contents are intentionally omitted; their recorded identities/inventories and published provenance are retained. Raw absolute receipt paths identify the builder machine and are not claimed to exist on peer machines. This review did not independently fetch the remote provenance commit or publish anything.
## Reviewed final records
| Record | SHA-256 |
| --- | --- |
| External native archive manifest | `acc5a1f1add9bc8be0711372224d8941ac14dfdc6018f7f6cc38b5bb798cc0b1` |
| Original staged manifest | `c90371bee1c2b2ae850b5d8e64ab9f1368f541ef51104e3b888a721e05d49c13` |
| Final parent observation note | `b4791b582d7028262b82b0742cde0d26b2439c770ff7b86ce6e6b19712053710` |
| Updated external `native-gui/README.md` | `a51c6455238e63b350c8f198c831e2c459de920a56f615fa9e606867e263a786` |
Archive home: `.local/behavior-anchor-plan-ytr83r-20260912/harness/prerelease37-windows-gates-20260924/native-gui`.
File diff suppressed because it is too large Load Diff