# State multiselect closed gate evidence tools

Preparation requires an explicit closed run list from `.local/state-group-gates`.
Preserve failed `desktop-1` alongside accepted unchanged runs. Exclude `core-1`
from this package: its false unchanged-input flag and one map/archive mismatch
belong in the separate pinned incident. Neither failures nor incidents are passes.

Example check-only invocation (extend with every subsequent closed run):

```
python -B prepare.py --root CHECKOUT --runs desktop-1 core-2 desktop-2 --check-only
```

Only after all selected captures close, omit `--check-only` and provide
`--output CHECKOUT/harness/experiments/state-multiselect-20260923/checks`.
Historical helper copies may be passed via repeatable `--helper PATH`.
No existing output is overwritten. Run `python -B verify.py PACKAGE` for readback.

The strict verifier is unchanged from the ordinary multiselect package: exact
seven raw files, helper bytes, safe archive inventories, all input-map/archive
bytes equal, receipt relationships and inputsUnchanged exactly true are required.
Failures retain their actual exits and streams. Source capsules preserve local
historical bytes; recorded HEAD alone does not identify those bytes.

The earlier 733-input helper omitted the core compile-time integration fixture
`harness/experiments/edge-attachments-20260922/native-fixture.arcadia` and the CLI
fixture `harness/experiments/textual-authoring-20260911/alarm-response.arcadia`.
Lint-1 failed, and its all-target capture additionally has a source-capsule
scope limitation for the core fixture. Later 735-input captures add both explicit
inputs. Desktop/core
example captures retain their actual tested scope; do not infer that one map
establishes a later map or complete all-target source identity.

Offline pin/receipt consistency is not test execution, authentication, native
acceptance, or proof of usability. Executables and toolchains are not distributed.