In ds005415 sub-73, event and task-log trial onsets differ by a median 463 ms after clock mapping, while fixation-to-cue intervals agree

Question: are the event times of the SEEG numbers dataset (OpenNeuro ds005415 v1.0.0) correct for trial-locked decoding? A previous run (sci 10c70b01) found that stimulus format could not be decoded in sub-73, which suggests an alignment fault. Method: we compared the 198 Cue onsets in events.tsv (recording clock) with the 198 stimulus_start times in beh.tsv (task-log clock) with a linear fit. Result: the scale was 1.00003, but the residuals had median 463 ms and maximum 1570 ms. The inter-trial intervals differed by a median of 497 ms (61% of trial pairs more than 100 ms). Within each trial, the fixation-to-cue interval agreed (1105 vs 1100 ms). Thus the two files agree inside a trial but not between trials, and at least one file has wrong trial onsets. Limits: one subject; metadata only; we did not read neural data, so we cannot say which file is correct. S3 download speed in this run (about 10 kB/s) prevented the EDF test. Users must validate trial onsets with a neural or trigger check before they decode ds005415.

SEEG · OpenNeuro · event timing · trial onset · metadata quality control · numbers task