Charts
Diagnostic charts
The figures that answer “is this model or process sane?”. As with every family: LUC renders, it does not estimate.
Time-series diagnostics
The lag correlations are computed by the caller and the significance band is drawn at the half-width the caller supplied — no sample size, no band.
Autocorrelation
Autocorrelation and partial autocorrelation over one axis; both series draw a stem from zero.
Partial autocorrelation
The partial autocorrelation alone, for reading against the panel on the left.
Survival and risk
The step is mandatory: survival is constant between events, so interpolating between two estimates would invent a risk the sample does not have.
Survival curves
Censoring marks are supplied points, never inferred from the data.
Ranking and model quality
Two machine-learning diagnostics: one measures ranking quality, the other measures each feature's direction and uncertainty.
Precision–recall
Recall on x, precision on y; the dashed line is the caller's no-skill prevalence.
Feature importance
Ranked horizontal bars; `signed` colours by coefficient direction, and the intervals are the caller's too.
Scientific correctness
These charts will not do the statistics on your behalf.
Lag correlations, survival estimates, precision and recall, feature importances and their intervals are all drawn exactly as supplied. Binning, fitting, resampling and correction happen where the data lives, which is what stops the picture and the report from disagreeing.