Parity and validation

The initial semantic reference is the supplied gpbiometrics 2.0.0 source release.

Reference precedence

1. R implementation
2. R tests
3. Rd documentation
4. vignettes/examples
5. explanatory repository/site prose

Completion criteria

The frozen 406-name export set is registered and implemented in Python. This is an API and implementation freeze, not a claim that every optional external package will return bit-identical numerical output across all library versions and platforms.

At the stable 0.1.0 freeze every export was directly referenced by Python tests, the complete suite passed on Linux/Windows/macOS across Python 3.11–3.14, and coverage exceeded the 90% release gate.

Independent deep-parity evidence

Stable 0.1.1 adds a second evidence layer under reference/golden/: the frozen R implementation and Python port independently generate deterministic outputs for numerical physiology/QC families, then a separate comparator applies explicit tolerances. This prevents a Python-only test suite from being mistaken for cross-runtime numerical evidence.

Optional backend behavior is tested separately because HeartPy, BioSPPy, pyHRV, NeuroKit2, MNE, pylsl and pyxdf can change independently of the frozen R reference.