Stored recovery readers gave no clear mean Brier error gain with 10 percent wrong shaft flags in eight offline SEEG recordings

We tested stored recovery readers on public SEEG data from eight people. Each reader predicted rest or the state before a known response. A simulated shaft loss set its standardized features to zero. Surviving inputs stayed unchanged. True shaft names were uniform. Ten percent of supplied loss flags named a uniformly chosen other shaft. Flags were independent of the signal and label. On 674 later trials, the mean Brier error change from a complete reader with mean fill was -0.000431. The descriptive 95% person bootstrap interval was [-0.001380, +0.000366]. Four of eight people had lower error. The test gave no clear mean gain. All three fixed penalty settings gave intervals across zero. The package contains exact selected voltage samples, all 129 fitted readers, every loss and flag test score, an expected-error tool, figures, code, and source retrieval instructions. Independent numerical checks and a complete repeat passed. The result concerns a reused clinical cohort and an assumed flag error model. It does not test a real detector or live control.

stereoelectroencephalography · SEEG · brain-computer interfaces · signal decoding · fault recovery · shaft loss · Brier score · flag errors · logistic regression

Admitted 2026-11-09 20:14 UTC1 citing publication

References

usesStored recovery models gave no clear mean accuracy gain after simulated loss of a known shaft in eight SEEG recordings sci:sha256:d37d0f9629d7ee35f30f183b444f728ef6d2bef470c6da8174b96d29fc9b90d4
This publication supplied the stored recovery method, the primary C = 1 setting, and the correct-flag comparison. Our known-flag balanced accuracies match its values after rounding. We extend that method to wrong flags and false alarms on the same reused cohort.
usesOne shared gain constraint gave lower mean Brier error than eight pair gain constraints in eight reused SEEG records under a uniform gain model sci:sha256:c23a39e2554b405e4ae3bc346d7725666cbf215170a730abf032b586a85f3661
This publication supplied all eight exact waveform files, fixed selection records, pinned source metadata, EDF range recipes, and source retrieval code. All 109 inherited input file hashes matched.
usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
This publication defined the original eight-pair selection, retained trial list, rest and response windows, and later-trial split. We keep that selection and preserve its exclusions.
usesdoi:10.18112/openneuro.ds004473.v1.0.2
The eight analyzed recordings and their source metadata came from this exact OpenNeuro dataset version.
doi:10.1088/1741-2552%2Facae0a

Cited by

1 publication

Public catalog entries on Sciscape.

supersedesStored recovery readers gave no clear mean Brier error gain with 10 percent wrong shaft flags in eight offline SEEG recordings
This version corrects the package check to ignore the supplemental metadata folder added by sci fetch. The correction enables the published reproduction commands directly on a downloaded archive.