# Build with us

Help grow FabQED Semi Harness from a deployable seed into integrated capability
native to semiconductor engineering. We welcome engineers, researchers, model
and tool developers, and teams operating private semiconductor environments.
Start with one bounded contribution; no proprietary corpus is needed.

## Useful first contributions

- **Domain expertise:** a scoped CMP, patterning, or advanced-packaging check,
  with units, applicability, counterexamples, and an engineering rationale.
- **Reproducible cases:** synthetic or openly licensed evidence that exposes a
  failure, an ambiguity, or a useful next measurement.
- **Integrations:** an adapter for an existing tool or evidence system, with
  explicit contracts and tests for access, identity, and failure handling.
- **Evaluation:** a matched native frontier comparison or independently reviewed
  engineering outcome, including regressions and uncertainty.
- **Usability:** a clearer installation step, documentation fix, or actionable
  bug report showing expected and actual behavior.

See the [roadmap](docs/ROADMAP.md) for priorities and acceptance evidence.

## Get the source and share a change

The [release page](https://semiaifoundry.com/fabqed/#community) provides the complete
MIT source archive, including tests, synthetic fixtures, and the package builder.
A public GitHub collaboration repository is pending; no GitHub repository is
claimed live by this release.

Until that channel is available, use the
[semiAIfoundry contact form](https://semiaifoundry.com/contact/) to share a link
to a public patch, fork, reproducible case, or proposal. Include the release
version, the problem, your change, and relevant validation. The contact form is
an introduction channel, not a place to upload confidential engineering data.

## Check a source change

From the complete source archive, with Python 3.10+:

```sh
python3 -m pip install -e .
python3 tools/release_check.py
python3 tests/test_public.py
python3 -m unittest discover -s tests -p 'test_workflow.py'
python3 -m unittest discover -s tests -p 'test_operating.py'
python3 -m unittest discover -s tests -p 'test_plugin.py'
python3 tools/build_plugin.py --output dist/my-release
```

Use a fresh build output directory. These checks use synthetic fixtures and
mocked transports; they do not run model inference. Run the checks appropriate
to your change, and report what you did. The installed plugin archive omits test
fixtures and build tools; use the complete source archive for development.

## Evidence and contribution terms

Submit material you have the right to share under the project's MIT License.
Keep source citations and compatible third-party license notices. Contributions
submitted for inclusion are offered under that license. Do not include private
process records, customer outcomes, model credentials, or operating stores.

Performance claims need a stated model and host revision, task scope, matched
raw evidence and tools, comparable budgets, independent engineering review, and
results on held-out work. Fixture success, a citation, or an impressive single
answer does not establish frontier-model improvement. Report negative results.

A code contribution and an organization's learned asset have different review
paths. Candidate organizational learning still requires its own trials, review,
and operator approval; merging software does not qualify a model or a procedure.