# STEP20C targeted closure evidence

EXPERIMENTAL / PARTIALLY_ADVANCED. STEP20C remains NEXT and production
multi-output remains NOT YET SUPPORTED. This directory records the consumed
targeted batch separately from its immutable pre-execution fixtures.

`RESULTS.json` separates measurements, raw runtime/runner status, scientific
cleanup and architecture verdict for all fourteen attempts. `ARTIFACTS.json`
binds the 460 retained run files, including 125 PNGs, by SHA-256. Images, logs,
SHM, binaries, builds and the future packaging recipe remain in the private
laboratory `/home/fy59/swayfx-mosaic-targeted-7gge6h8b`, outside Git. Historical
evidence in `/home/fy59/swayfx-mosaic-overnight-mcolx0gq` was read only.

The [closure report](../../../docs/design/STEP20C_TARGETED_CLOSURE_DECISION.md)
and [decision draft](../../../docs/design/STEP20C_PRODUCTION_INTEGRATION_DECISION_DRAFT.md)
describe the observations, risks and remaining gates. Positive sampled results
do not qualify all frames, GPU lifetime, atomic generations or a future ABI
pair. Whole-H1 A-return inequality is an observed failure in pristine/A/G;
its INCONCLUSIVE marker concerns causal diagnosis and integration readiness.

The original [core fixture](../step20c-targeted/README.md) and
[diagnostic fixture](../step20c-targeted-diagnostic/README.md) retain their
reviewed bytes, including known limitations exposed by consumed attempts:

- Saved A/G requested an IPC workspace gaps field absent from this binary.
- Input A/G refreshed geometry before clicking, but focus changed the camera
  during the third click. The wire coordinates match the later applied origin.
- C3 rejected legitimate reuse of a native node address. Its retained native
  creation/render/teardown observations do not replace the rejected runtime
  predicate or prove natural timeout removal.
- All fourteen raw terminal IPC exit responses are failures despite recorded
  process exits zero. Cleanup and positive measurements are recorded separately.

Do not launch, repair and retry a consumed ID, reset a ledger, edit a historic
receipt, or treat a synthetic offline trace as runtime evidence. The diagnostic
binary is distinct from A/G. Its passive listener/log overhead is experimental.

Read-only verification of the compact records and retained files:

```sh
PYTHONDONTWRITEBYTECODE=1 python tooling/probes/step20c-targeted-results/verify_evidence.py
```

This command hashes files only; it launches no compositor and does not write
or reopen a run. Publication commits cannot contain their own future SHA.
The final publication references and independent-review receipt are recorded
in the private closeout manifest and final user report.