The theory yields quantitative predictions of transfer gains between old and new systems

What it is

Cognitive Complexity Theory does more than assert that shared rules help; it can quantify the help. Predicted learning time for a second system equals "unshared rule count × per-rule constant," and comparing that with learning from scratch ("total rules × constant") yields a quantitative prediction of the transfer gain.

Why it happens

The gain comes from the difference set of the two rule inventories. For candidate designs A and B, each is diffed against the old system; the smaller difference set predicts shorter learning time. Because the magnitude comes from the size of the difference set rather than a vague sharing ratio, two non-identical designs can be ranked.

Studying it

Validation is prediction-versus-observation: modelers compute the difference set from production-system models, state a learning time prediction, then have novices learn the new system and regress observed time on the prediction. When the prediction beats a total-rule-count baseline, the difference set is a valid predictor.

Where it stops holding

Quantitative prediction assumes a stable per-rule constant and a consistent rule granularity chosen by the modeler. The difference set captures declarative condition-action knowledge only; it does not predict fluency, error rates, or affect. Transfer gains are relative quantities suited to ranking alternatives, not absolute schedule promises.

Applying it

• Score redesign candidates by rule-difference count and prefer the smallest difference set for releases within the same period.
• Turn predictions into acceptance thresholds, such as "experienced users' first-use time on new tasks under 1.3x the old task," and verify before launch.
• Keep each version's rule model so the next redesign reuses comparisons at the same granularity.

Related

• Same group: B4.14.1 Learning time can be estimated from the number of rules the user must newly acquire · B4.14.2 Mastered rules transfer; the more two interfaces share, the cheaper the second is to learn · B4.14.4 Rule counts depend on the representation the modeler chooses, so different modelers may produce different counts
• Nearby: B4.12 Keystroke-Level Model and operator constants · B5.02 Effectiveness, efficiency, and satisfaction
• Search terms: transfer prediction · learning time estimation · interface consistency