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.
All repositories in this organization are MIT. By contributing you agree that your contributions are licensed under the repository's MIT license.