Exact sample runs crossed random splits of the first 120 SEEG stimulation windows in eight of nine files
We tested full stimulation windows from nine public files in OpenNeuro ds006065. The files came from seven people. We retained three SEEG channels and the first 120 windows per file. Each window had 1,500 samples. We tested 1,071 neighboring window pairs. We found 64 exact shared runs and 710 close waveform matches. In one fixed random 80/20 split per file, 14.3% of test windows had an exact train match. With close matches included, the share was 94.6%. The mean gives each person the same weight. An ordered split with two windows removed had 0 detected matches across the split. The package includes source arrays, match tables, split files, figures, code, and pinned retrieval records. The result concerns full exported windows. It does not measure error in a BCI decoder. Close matches are weaker evidence than exact shared words.
SEEG · brain-computer interface · waveform overlap · data leakage · train-test split · intracranial EEG · stimulation windows · signal processing
References
4- usesTwenty reconstructed pulse blocks give higher late bipolar map agreement than five in seven people from ds006065 sci:sha256:d5d43d3b27b59b9165ef06329e145efd331b49cc4fa65856c51da9414336ca6c
Used its archived original TSS source code and dataset version receipts to define source window lengths, counts, and named phase channels. Downloaded new stimulation samples. Did not use its response maps.
- usesdoi:10.18112/openneuro.ds006065.v1.0.0
Source of the nine stimulation files and retained windows analyzed in this publication.
- supportsdoi:10.1038/s41467-025-59417-7
Provides the stimulation context and reports that consecutive pulses were at least one second apart, an assumption used to motivate the gap rule.
- usesdoi:10.5281/zenodo.14735080
Its archived source code and version receipts define the source window lengths, counts, and named phase channels used to plan this audit.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesUnder the source pulse layout, anchored mean subtraction after pulses raised pre-pulse phase locking in five SEEG control files for a centered 4–10 Hz filter
We reused its nine compact stimulation arrays, source receipts, pinned range retrieval code, source layout records, and source code archive. We fitted new earlier-window mean curves and computed new phase and locking sensitivity results.
- usesDetected sample sharing across SEEG train and test splits changed with the input time range in nine files from seven people
Provided the nine three-channel data arrays, full-window match graph, fixed splits, source receipts, and retrieval code. This study verifies the exact runs and tests their intersections with six input time ranges. It also adds an exhaustive short-response exact search and new split tools.