Contributing

Contributions to the Skill, schemas, validators, tests, and evidence records are welcome.

For every molecule or interaction record:

1. Use a stable record ID and normalized molecule identifier where available.
2. Cite the exact source through references/source-ledger.jsonl.
3. State product, matrix, concentration or threshold context, method, and limitations when the source reports them.
4. Distinguish occurrence, correlation, contribution, reinforcement, synergy, suppression, and masking.
5. Keep model-generated or incompletely checked records in proposed review state with HYP or UNK status.
6. Do not copy article bodies, database exports, or other material that cannot be redistributed.

VER is sample- and claim-specific. Detection alone does not verify sensory contribution. Co-occurrence or correlation does not verify synergy.

• Keep validation and graph-building scripts offline and deterministic by default.
• Do not add credentials to code, fixtures, logs, or workflow files.
• Document all network, shell, file-write, and API behavior in the pull request.
• Pin third-party GitHub Actions to immutable commit SHAs.
• API-assisted imports must create reviewable proposals; they must not overwrite accepted records automatically.

Run:

python scripts/validate_knowledge_base.py
python -m unittest discover -s tests -v

Also run the Codex Skill Creator quick_validate.py against the repository root with UTF-8 mode enabled on Windows.