Canonical workflows¶
Version 0.55 introduces a canonical workflow layer. It does not add a new estimator. Its purpose is to make the existing package navigable by scientific question rather than by function count.
Start with one of five routes:
| Scientific question | Canonical route | Primary entry point |
|---|---|---|
| What are the dominant modes of continuous gaze variation? | Continuous gaze exploration + FPCA | fit_fpca() / fit_mfpca() |
| How does an experimental predictor change a continuous response over time? | Experimental functional regression | fit_function_on_scalar_regression() |
| How does a repeated-trial functional response vary with predictors while respecting participant/trial hierarchy? | Repeated-trial functional mixed effects | fit_functional_mixed_effects_regression() |
| How does a repeated binary/count functional response change with predictors? | Generalized binary/count responses | fit_generalized_function_on_scalar_regression() |
| Is the scientific target recurrent or nonlinear temporal structure rather than a mean trajectory? | Nonlinear/recurrence analysis | recurrence_matrix() / rqa_metrics() |
Each route follows the same discipline:
- define the scientific response and sampling unit;
- make preprocessing choices explicit;
- fit one declared model/specification;
- inspect uncertainty and diagnostics appropriate to that model;
- perform sensitivity analysis only when scientifically predeclared;
- report the estimand, clustering/resampling unit, assumptions, failures and provenance.
The capability inventory contains the full advanced surface. The API stability policy explains canonical, advanced, diagnostic and experimental status.