OpenNeuro ds004993 W2 event onsets align at scale 1.000; the 1.024 optimum follows from mislabeling 102.4 Hz data as 100 Hz

For one participant (W2) in OpenNeuro ds004993, we tested whether TIMIT task events.tsv onsets require a 1.024 multiplier, as reported by one prior publication, or are correct as recorded, as reported by another. The EDF header specifies 512 samples per second, and every event sample equals onset × 512. The EDF annotation events also match at a constant absolute-clock offset, with no scale difference. We computed a common-average-referenced 70–150 Hz high-gamma envelope from 106 good SEEG contacts and scanned alignment scale a and offset b against 101 sentence onsets. With exact decimation by 8 (64 Hz), the best alignment was a = 1.000, b = +0.08 s, and mean Pearson r = 0.242 for the ten highest-scoring contacts; at a = 1.024 the best r was 0.071. Windowed lags remained 0.05–0.15 s over about 300 s. Inter-onset gaps were 1.075 ± 0.009 s over 100 gaps, and their slope on sentence duration was +0.012 (SE about 0.003), unlike the approximately −0.023 expected under the proposed scale error. Repeating the scan with decimation by 5 while labeling the true 102.4 Hz output as 100 Hz produced a = 1.024 as the optimum (r = 0.243), and the windowed lag increased to 6.6 s. Thus, for W2, events.tsv onsets should not be multiplied by 1.024 when data are read at the correct 512 Hz rate. The prior response and decoding values are probably valid because the onset and time-axis errors cancel. This analysis covers one participant; W3 was not run. The decimation-5 windowed lag search is noisy, and the global scan is the main evidence. The analyzed files came from a live OpenNeuro S3 snapshot read on 2026-10-11; its exact immutable dataset release is unknown.

intracranial EEG · SEEG · event timing · high-gamma activity · decimation · speech listening · OpenNeuro ds004993