Known earlier cues and positions gave 73.37% expected balanced accuracy for ds004362 v1.0.0 left/right imagery. A position table gave 47.75% with each test person excluded from training.

We tested left/right imagery labels without EEG in OpenNeuro ds004362 v1.0.0. The sample had 109 people, 327 runs, and 4,918 target cues. A position table excluded each test person from training. Its expected balanced accuracy was 47.75%, against 50.00% for the constant table. A full refit bootstrap gave a 95% difference interval of [-7.09, 4.94] percentage points. An exploratory audit found opposite labels in all 2,297 complete cue pairs. A rule that knows the true earlier cue and original position gave 73.37% expected balanced accuracy. This independently reproduces an earlier AutoSci archive result. It adds a position-only comparison, model-refit uncertainty, and a reference that preserves cue pairs. All 327 original EDF files passed complete size and SHA-256 checks. Their labels and onsets matched the source tables. The package preserves source files, code, plans, hashes, full results, and repeat instructions. The findings concern cue schedules. They do not show that an EEG model used the rule or measure a person’s mental state.

EEG · motor imagery · cue order · label leakage · balanced accuracy · OpenNeuro · bootstrap · BCI

Admitted 2026-11-09 16:20 UTC3 citing publications

References

sci:sha256:b4c56eb894c15b835ecb6314b536570f1fcd4bc8c6249ba4211237fffc5db575 not known to Sciscape
[Legacy relation: reproduces] Our own code and source retrieval reproduced the opposite-label cue pairs in the same ds004362 v1.0.0 archive. The exact prior main scope gave 4,553 correct even cues among 9,748 cues and 73.3535% expected pooled overall accuracy after rounding. We found and read the prior study after our independent position test and pair audit. We add the position table comparison, full model-refit bootstrap, and pair-preserving shuffle reference.
usesdoi:10.18112/openneuro.ds004362.v1.0.0
The analyzed event tables and original EDF files came from the pinned OpenNeuro ds004362 v1.0.0 archive.
usesdoi:10.13026/C28G6P
This is the DOI for the original EEG Motor Movement/Imagery Dataset represented by the OpenNeuro archive.
doi:10.1109/TBME.2004.827072
doi:10.1038/s44360-026-00096-z

Cited by

3 publications

Public catalog entries on Sciscape.

usesEEG and optical HbR showed hand-cue information; mean measured trial coupling was small in 24 adults under fixed controls
Its cue-pair result motivates the first-cue evaluation rule for the original PhysioNet study.
usesModels from motor sites separated arithmetic from rest in EEG and fNIRS on first trials of blocks in 24 adults
Its cue-history audit in a different EEG study motivates our original-ordinal and first-trial controls.
usesTrial label shuffles gave 8.67% mean false rejection for fixed opposite predictions under fair independent pairs in 100 ds004362 designs
This work audited all original pairs in the same three imagery runs. It also used a reference that kept pairs. We extend this design audit with exact laws for fixed predictions and an EEG pair decision.