Please open an issue before a substantial change. Include a minimal synthetic example, expected and actual behaviour, package version and software environment. Never attach confidential or personally identifying data.
Documentation corrections should identify the page and section. Keep examples small, deterministic where practical and within the documented scientific scope. Separate presentation changes from numerical, model or default changes. Methodological changes require explicit review; a passing smoke test is not evidence of a statistical guarantee or a production-study result.
Use the existing tests and public interfaces. Explain any new dependency and preserve citation, data-source attribution and licensing. Report validation limits rather than making unmeasured performance claims.