In ds007436, surprise and scaled log cue exposure give equal response-time fits with log(8 + trial) trends in each run

We tested all 348 task runs from 58 participants in OpenNeuro ds007436 version 1.0.0. All supplied surprise series fit a rule that counts earlier cue presentations. None fit the paper's equation that includes the current presentation, under the stated nine-cue prior. We reconstructed integer exposure counts and proved an exact regression equivalence. Surprise and scaled log cue exposure give equal fitted response times when each run has an intercept and a log(8 + trial) trend. The largest fitted-value difference was 1.15e-13 seconds. A planned cubic time model gave an auditory slope of 8.14 ms per surprise SD and a tactile slope of 0.80 ms/SD. The auditory p value was 0.0313 before correction and 0.0625 after Holm correction for two modalities. The public records lack unchosen action values and cue labels. This is a limited reanalysis and a test of model design, not an exact repeat of the source study. The package preserves source tables, hashes, code, results, a proof, and reproduction instructions.

multisensory learning · surprise · cue exposure · response time · regression identifiability · OpenNeuro