Beyond the Single Forgetting Curve: A Typologically-Extended Ebbinghaus Function and a Preregistered Test of Error-Type-Specific Memory Decay.

This preregistration specifies a Bayesian secondary analysis of the parent Confusion Fingerprint Index (CFI) three-arm parallel randomized controlled trial conducted in Kanpur Dehat, Uttar Pradesh, India (n = 90). The analysis tests whether the semantic type of cognitive error committed at encoding moderates the parameters of a typologically extended Wickelgren power-law forgetting function. Four error classes are distinguished by the CFI: Recall Failure, Partial Knowledge, Confabulation, and Interference. Binary item correctness is modelled across four retention intervals — immediate, 2 days, 7 days, and 14 days — using four nested Bayesian multilevel models (M0 null through M3 full typology) with crossed learner and item random effects and informative, source-anchored priors. The 7-day interval is included specifically to capture the documented one-week return of high-confidence errors and to help resolve power-law/exponential non-identifiability. Hypotheses are stated as directional parameter inequalities with symmetric falsification rules. Model comparison will use PSIS-LOO as the primary criterion, WAIC as a secondary criterion, and bridge-sampled Bayes factors for focal contrasts. Error coding will be completed blind to retention outcomes. The dataset is fixed and inherited from the parent trial; there is no new randomization, no optional stopping, and no human-participant results are reported here.