Research-readiness checklist

Use this checklist when adding a new scientific algorithm or end-to-end workflow to CDS2.

Numerical implementation

[ ] Mathematical definition and domain are documented.
[ ] Input validation covers invalid shapes, non-finite values, and unsupported regimes where relevant.
[ ] Numerical stability considerations are documented.
[ ] Deterministic behavior is guaranteed when the algorithm is deterministic.
[ ] Randomized algorithms expose an explicit seed or generator.

Verification

[ ] At least one analytical invariant, closed-form case, or independent oracle is tested when available.
[ ] Degenerate and boundary cases are covered.
[ ] Property-based tests are added for reusable numerical invariants when practical.
[ ] Iterative methods expose or test convergence/residual information.
[ ] Tolerances are justified rather than chosen only to make a test pass.

Reproducibility

[ ] Random state is recordable.
[ ] Version and configuration information needed to reproduce outputs is available.
[ ] External datasets have provenance and immutable identifiers/checksums.
[ ] Generated reports and plots can be recreated from code.

Documentation

[ ] API documentation explains assumptions and failure modes.
[ ] A minimal runnable example exists.
[ ] A realistic case study exists for substantial new workflows.
[ ] Claims about accuracy or speed link to evidence.

Performance

[ ] Benchmarks compare equivalent numerical work.
[ ] Environment and repetition methodology are recorded.
[ ] Correctness is checked before timing results are accepted.
[ ] Performance regressions are reviewed explicitly.

A completed checklist is evidence of engineering discipline, not a substitute for independent scientific review. Domain-critical use still requires validation appropriate to the application.