Single-image fits of a simple temporal model did not confirm a shared repetition prediction gain in 8 ECoG patients and 6 fMRI adults

We tested whether single-image responses predict repeated-image responses in two separate human studies. We fitted a bounded exponential temporal filter with static compression to six single-image response areas at each local unit. Six repetition conditions were held out. The source data contained 98 ECoG contacts from eight patients and 108 fMRI region and image fits from six adults in eight experiment records. We retained all 72 possible links for the two repeated fMRI participants. The temporal model reduced repetition error relative to a total-exposure model by a mean of 15.0% in ECoG, with a 95% person bootstrap interval of 4.3% to 25.4%. The fMRI mean was 5.3% to 9.8%, with interval union -1.9% to 16.5%. The fixed cross-modal test did not pass (p = 0.125). Shorter ECoG windows and pooled fits weakened the gain. At a planned tolerance equal to 1% of squared duration responses, median supported time constant ranges were 116-fold in ECoG and 61-fold in fMRI. At 1% of residual error, they were only 2.1-fold and 2.4-fold. These grid ranges are not statistical confidence intervals. The data do not confirm a common neural time constant or a shared prediction gain. The package supplies a new prediction benchmark, code, both original source files, provenance, all fits, negative results, controls, figures, and offline repeat steps. Cohorts, clinical status, spatial sampling, source processing, and the assumed area readout limit the tested scope.

ECoG · fMRI · visual cortex · temporal integration · repetition · prediction · parameter constraints · participant dependence · reproducibility