Bug Fixes & Robustness:
scp_geostatistical(): 1D coordinate vectors passed as s0 are now properly coerced to a 1-row coordinate matrix matching training dimensions.plot.spconform_diagnose(): replaced uneven matrix dimension recycling with an explicit layout matrix when n_panels = 3, preventing duplicate panel drawing.scp_geostatistical() and scp_areal() guarding against missing (NA) values in coordinates and responses.na.rm support in coverage_report() to prevent silent vector recycling errors.Performance & Computational Efficiency:
scp_geostatistical(): bandwidth is now evaluated once prior to the query loop rather than recomputed per target location, yielding up to a 50x speedup on large query grids.k_neighbors argument to scp_geostatistical() for localized $k$-NN calibration, defaulting to NULL for complete backward compatibility.Statistical Enhancements:
coverage_report() now evaluates and reports the strictly proper Winkler Interval Score (winkler_score) alongside empirical coverage and mean interval width.spconform objects:
predict.spconform() to extract point predictions or prediction intervals (fit, lwr, upr).residuals.spconform() to compute raw response residuals or absolute calibration scores.as.data.frame.spconform() to coerce conformal objects into tidy data frames.diagnose() now returns a classed spconform_diagnose object with dedicated print.spconform_diagnose() and plot.spconform_diagnose() methods.R/scp_areal.R for full ASCII compliance.spconform on CRAN.scp_geostatistical()).scp_areal()).spatial_kernel_weights(), areal_neighbor_weights()).diagnose()).print(), summary(), plot(), coverage_report()).