A fixed model with equal hidden probability assignment gives no clear ambiguity prediction gain in 15 EEG test participants from ds004917 version 1.0.1

We tested whether a fixed equal assignment of hidden probability improves prediction of ambiguity choices in OpenNeuro ds004917 version 1.0.1. Both logistic models use visible probability and reward and have four coefficients. We fitted them to 1,520 choices from 38 fMRI participants. We tested 3,553 choices from 15 EEG participants with no shared training IDs. The primary ambiguity losses were 0.65022871 for the ignore model and 0.65022016 for equal assignment. The gain was 0.00000855 natural-log units per choice. We sampled whole people 4,999 times and fitted both models in each sample. The 95% interval was [-0.0448, 0.0447]. All six fixed checks also included zero gain. The secondary EEG-to-fMRI test gave a gain of 0.08676 [0.0362, 0.1430] in 29 test participants. The directions reuse source participants and are not independent replications. The primary result does not establish equal performance or identify a private belief. The split assumes consistent public participant IDs. We analyzed task choices, with EEG stimulation conditions pooled. We did not analyze brain signals or estimate stimulation effects. The package includes audited source files, code, split IDs, trial predictions, all resample indices and fits, independent checks, figures, and reproduction instructions. Fresh checks matched all 196 source files. A complete rerun reproduced all 24 analysis outputs.

ambiguity · choice prediction · hidden probability · logistic regression · cross-modal prediction · fMRI · EEG