In ds004993, a no-repeat SEEG control does not resolve the early-run confound in repeated-sentence responses

Question: in ds004993 sub-W2, the 70-150 Hz (HFA) response of 13 temporal SEEG pairs is largest at the first presentation of each repeated TIMIT sentence. Repeat number and trial index correlate at r = 0.967, so repetition and an early-run effect are nearly the same contrast. Method: the same numpy pipeline as the prior analysis was applied to sub-W2 and the planned no-repeat control, sub-W3. A pre-registered early excess statistic compared mean normalized responses on trials 1-10 with trials 11-100, using 5000 trial-order permutations and a bootstrap confidence interval. Results: sub-W2 E = +0.51 SD (p = 0.0004, 95% CI 0.23 to 0.80). In sub-W3, with 25 responsive pairs and no repeated sentences, E = +0.25 SD (p = 0.125, 95% CI -0.00 to 0.53; 20/25 pairs positive). The sub-W3 linear trial slope was -0.005 SD per unit (p = 0.78). Neither pre-registered decision rule was met: the sub-W3 effect was not significant, and its confidence interval included the sub-W2 value. An exploratory sub-W2 test after a linear time term found a first-presentation coefficient of +0.36 SD (Freedman-Lane p = 0.023), but the test separates repetition and run position only weakly. Limits: two patients had different contacts and sentence sets; responsive pairs were not independent; power for a 10-trial early window was low. The control is inconclusive: a time-in-session effect can still explain part or all of the sub-W2 excess. Do not assign the excess to repetition suppression alone; model run position as well as repeat number in noise-ceiling estimates.

SEEG · intracranial EEG · high-frequency activity · repetition · time in session · TIMIT · OpenNeuro ds004993 · permutation test · preregistration