Diagnose

diagnostics() is the per-group dictionary for programmatic and audit access. spline_redundancy() reports knot spacing, basis correlation and effective rank, and discretization_impact() measures what binning the smooth terms into rating-table bins and grids does to the predictions. iteration_diagnostics() is the per-iteration IRLS table, a DataFrame available after fit(record_diagnostics=True); reml_diagnostics() and training_telemetry() are the solver’s own records as plain JSON-serialisable dictionaries with no tracking dependency, ready for MLflow, files, logs or a governance system.

diagnostics	Per-group diagnostic dict for programmatic / audit access.
spline_redundancy	Spline redundancy diagnostics: knot spacing, basis correlation, effective rank.
discretization_impact	Analyse the impact of discretizing this fit's smooth terms.
iteration_diagnostics	Return per-iteration IRLS diagnostics as a DataFrame.
reml_diagnostics	Return dependency-free REML telemetry for external tracking.
training_telemetry	Return dependency-free training telemetry for external tracking.