92 lines
5.3 KiB
Markdown
92 lines
5.3 KiB
Markdown
# Native multiselect evidence package
|
|
|
|
This package preserves native run-1 and run-2, both closed with exit 0, empty app
|
|
stdout/stderr and all three GUI/checker/fixture unchanged flags true. Each run
|
|
contains nine positive complete-Project checker snapshots, nine source files per
|
|
snapshot and three exact source-history equality groups. Only
|
|
`views/diagram_oa_view.arcadia` differs from baseline in its move and hidden
|
|
states. These artifacts do not establish native acceptance on their own; consult
|
|
the separate [saved-view report](../../../multiselect-saved-view-20260923.md) for
|
|
action observations, usability findings and the implemented subset of issue 29.
|
|
|
|
| Run | Source commit | Captures | Layout scope |
|
|
| --- | --- | ---: | --- |
|
|
| run-1 | `12a7352ddef5f69a144ead8219a50916233d63a9` | 27 | Before selection/status toolbar correction |
|
|
| run-2 | `f6a6010baed9e6164a1fcc7ecd58256a562827d7` | 24 | Corrected toolbar layout build |
|
|
|
|
Run-1 GUI/checker build receipts refer to the 729-file input map
|
|
`55a5f354bb5049a8c77ba9bafa833438ebf754649b0cf5d175d0c8dcfcedd7bb`;
|
|
run-2 refers to the distinct 729-file map
|
|
`99dc46645236e674244e7c4867595a6000d3d85b54bcc9ab2b0c601923ecf66c`.
|
|
The [source identity evidence](../source/README.md) ties those maps to the commits
|
|
above. The GUI binary pins differ; the recorded checker binary pins match.
|
|
Run-1 is retained as evidence of the earlier layout, not as verification of the
|
|
corrected build.
|
|
|
|
The first run-2 baseline checker attempt failed before the successful retry after
|
|
the toolchain environment was initialized. Its unchanged raw snapshot and retry
|
|
comparison are preserved separately in the [environment incident](../incidents/README.md).
|
|
The successful retry is the baseline snapshot included in this native package;
|
|
the failed invocation is not silently reclassified or presented as a GUI failure.
|
|
|
|
These Python 3.10+ standard-library tools never operate the GUI or rerun the model
|
|
checker. From the checkout root, validate both raw runs without writing a package:
|
|
|
|
```text
|
|
python -B harness/experiments/multiselect-20260923/native/prepare.py --root . --runs run-1 run-2 --check-only
|
|
```
|
|
|
|
Omit `--check-only` to package; optionally choose `--output DIRECTORY`. The default
|
|
is this native directory. Existing package outputs are not overwritten. A missing
|
|
or unfinished run, changed recorded inputs, missing/extra fixture or snapshot
|
|
files, unmatched helper bytes, or inconsistent receipts fails preparation.
|
|
|
|
Verify a completed package independently of the original machine:
|
|
|
|
```text
|
|
python -B PATH_TO_PACKAGE/verify.py
|
|
```
|
|
|
|
Each run includes all original capture JSON/images, process/start/finish records,
|
|
generation and app streams, all nine source snapshots and checker streams/receipts,
|
|
and the history-bytes record. Its fixture includes baseline JSON, the manifest,
|
|
and canonical source files. Exact helpers are found by their recorded hashes among
|
|
the native directory's `.py`/`.js` files, so original and suffixed run-2 helpers can
|
|
coexist. Helpers are archived as data and never executed by these tools.
|
|
|
|
Referenced GUI/checker build results, exact gate helpers and input maps are
|
|
included. The checker example is read from its pinned historical input ZIP, not
|
|
from current source.
|
|
Full build logs and large source ZIPs belong in the sibling `checks` package;
|
|
verify that package separately. This package validates build receipt, input map,
|
|
example-source and executable-pin relationships without duplicating those ZIPs.
|
|
It does not contain or execute the GUI/checker binaries.
|
|
|
|
The verifier checks every archived byte pin and exact safe membership, clean
|
|
recorded exit and unchanged flags, process and capture lifetimes, each capture's
|
|
normalized executable path against the launched GUI, one consistent positive
|
|
window ID per run, fixture/source inventories, nine positive complete-Project
|
|
checker reports, baseline/binary/helper
|
|
relationships, and the actual archived bytes in three history equality groups.
|
|
It verifies that only `views/diagram_oa_view.arcadia` differs from baseline in the
|
|
move and hidden source states, and that baseline source bytes match the fixture.
|
|
Window identity checks join recorded metadata; the receipts do not independently
|
|
map that window handle to the recorded PID. Source models are not recompiled by
|
|
this offline Python verifier. Screenshot dimensions are recorded metadata, not a
|
|
minimum-window usability evaluation.
|
|
|
|
`manifest.json` maps each run's original historical paths to archived run,
|
|
fixture, helper, build and checker-source members and pins their bytes.
|
|
`files.json` pins the packager, verifier, README, manifest and archive. Raw bytes
|
|
are preserved; ZIP ordering and timestamps are fixed. Historic absolute paths are
|
|
labels and are not accessed by offline verification.
|
|
|
|
GUI actions remain separately documented parent observations. Screenshots do
|
|
not prove which inputs occurred, and the checker infers a bounded common delta
|
|
from saved geometry; it cannot validate the exact pointer destination. Source
|
|
history byte equality supports the stated saved-state relationships but does not
|
|
independently authenticate button presses, focus behavior or Undo stack structure.
|
|
Run-1 evidence does not establish the corrected toolbar's usability or any later
|
|
binary's behavior. Consult the parent report for the evaluated version, observed
|
|
limitations, run differences and the remaining full-issue acceptance work.
|