Thanks for helping make self-deception structurally impossible.
- Every change must make something harder to fake. A feature that cannot be tied to a form of self-deception (overfitting, look-ahead, PIT drift, forgotten tuition, post-hoc rationalization, dirty data) probably does not belong in this organization.
- Zero-dependency is a feature. New hard dependencies need a strong justification in the PR description.
- Fail-closed by default. When in doubt, a check that refuses to run is better than a check that silently passes.
- Honesty about boundaries. If your change sits at a heuristic boundary (e.g. value-dependent semantics, vendor-specific conventions), say so in the docstring and the README. Never claim verifiability where the theory does not allow it.
- Fork the repo and create a branch from
main. - Implement with tests. The project family standard is
pytest, Python 3.11+, zero (or near-zero) dependencies. - Run the full test suite locally before pushing.
- Open a PR. CI runs the suite on Ubuntu / Windows / macOS.
Conventional commits (feat:, fix:, docs:, chore:, test:,
refactor:) — keep history clean and machine-readable.
Issues are handled on weekends. Bug reports that describe a plausible-but-wrong result (silent self-deception) get priority.
Every repository labels beginner-friendly issues good-first-issue. These
are deliberately small and self-contained: a missing test, a docstring
boundary statement, an example fixture. If you are new to the organization,
pick one, mention it in your PR, and maintainers will review quickly.
All repositories in this organization are MIT. By contributing you agree that your contributions are licensed under the repository's MIT license.