Prediction metrics

Regression metrics

• R² — variance explained by the model. 0.9+ is strong for chemistry & materials data; 0.7–0.9 is typical and usable with uncertainty bands.
• RMSE / MAE — error in target units. Always read with the target's scale.

Cross-validation protocol

All metrics come from cross-validation, not a single train/test split — on small data a single split is noise. The reported interval and extrapolation flags come from the same fold structure, so they are consistent with the headline metrics.

Uncertainty metrics

• Conformal interval — a distribution-free range for each prediction; roughly 90% of true values should fall inside it.
• Extrapolation flag — set when the input sits outside the training envelope. A flagged prediction is a hypothesis.