Fixed tests did not confirm past word surprise interactions in six iEEG patients and 18 fMRI adults

We tested whether recent observed word surprise lowers the conditional signal association with current surprise. Separate public studies supplied high-frequency iEEG and BOLD fMRI during two stories. Development used three patients and four MRI adults. Fixed tests used six other patients and 18 other adults. The past-interaction score means were +0.012529 and +0.070263. Their 95% person Student intervals were [-0.000162,0.025220] and [0.021510,0.119017]. The shared two-claim Holm p was 0.09375. A past-versus-future interaction test also failed, p=0.296875. All fixed controls and negative scores remain. Only 1.81-2.01% of the MRI past-product predictor squared norm remained after nuisance projection, versus 81.02% in iEEG. A later input-only scan found that 0.1 s sampling still left only 1.92-3.90% under the fixed hemodynamic model and timing controls in both stories. This is a design result, not positive neural generalization. A constructed linear surprise signal with zero history interaction gave nonzero scores under a response timing error in each source design. This is an interpretation limit, not a fitted human cause. Separate arithmetic checked all 24 reserved people. The package preserves permitted compact signals, source pins and terms, code locks, all results, figures, and repeat steps. The tests do not confirm shared neural adaptation or establish its absence. Different cohorts, stories, surprise models, clinical sampling, source filters, spatial averaging, motion, and hemodynamic assumptions limit scope.

iEEG · fMRI · word surprise · context · response gain · model error · natural speech