Changes

Substantive changes to this guide are recorded here. Corrections of typography and style are not listed; the full history is in the repository.

12 September 2026

Correction to the worked example. Earlier versions had the central example the wrong way round. They reported surprise ratings as lower at event boundaries than within events, on the grounds that boundary detection reduces prediction error. Event Segmentation Theory holds the opposite: a boundary is the point at which the current event model stops fitting the input, prediction error rises, and that rise is what triggers the update. Surprise should therefore be higher at boundaries. The example now runs in that direction throughout. The filter decisions in the worked example are unaffected, because the competing accounts make the same directional prediction either way.

New chapter. “Naming Your Predictions”, on why numbered hypotheses lose the theoretical content of a prediction, and what a working name has to do.

Cross-references. References between chapters are now Quarto cross-references rather than literal chapter numbers.