A higher score threshold changed exact command times in eight fixed SEEG streams with a one-second waiting period

We tested whether a higher decoder score threshold only removes commands from a SEEG replay. We used public OpenNeuro ds004473 version 1.0.2 and eight fixed patient decoders. The reference score threshold was zero. The higher threshold was the 99th percentile of calibration rest scores. Tests covered 112.31 minutes, 67,309 windows, and 816 later task responses. With a one-second waiting period, 1,033 of 1,153 higher-threshold commands were absent at exact reference times. The equal-patient mean was 89.4 percent. Its descriptive 95 percent patient bootstrap interval was 84.3 to 93.1 percent. With a 0.5-second matching tolerance, the mean unmatched fraction was 5.2 percent. Total command counts did not increase. We proved a one-to-one timing bound of less than one waiting period. We also proved that a gate after a fixed reference schedule can only remove its command times. Replay confirmed this subset rule. This gate reduced mean response detection from 23.8 to 7.1 percent. Mean outside-guard command rates fell from 7.8 to 1.4 per eligible minute. Response guards do not establish absent movement intent. We audited all continuous feature windows against pinned raw source ranges. The package preserves data, code, proofs, figures, results, source records, and reproduction instructions. This fixed-stream test measured no live control or stimulation effects.

SEEG · brain-computer interface · command scheduling · thresholding · refractory period · reproducibility

Admitted 2026-11-09 18:33 UTC6 citing publications

References

usesAt a zero threshold, eight-pair SEEG decoders gave a mean of 36.7 outside-guard triggers per eligible minute in eight later task scans sci:sha256:22495a59283fcfea65785cbf2a7a410c3a81dc75e721320a08a5ed114e14bd43
We reused the eight event and continuous feature files, fixed decoder models, score arrays, and pinned source records. We reproduced its reference and 99th-percentile command arrays and added schedule-level proofs and gating tests.
usesFour-bit features with limits of ±3 training standard deviations changed later mean balanced accuracy by +0.10 percentage points in eight fixed SEEG decoders sci:sha256:dbac6507cc56b2e459119a6b65c54640044b6aa86ed6a6d6dd3209aac93e4003
Its exact command changes under feature coding and stated limits of decision certificates motivated a separate schedule-level test. We used no compressed feature data from it.
usesdoi:10.18112/openneuro.ds004473.v1.0.2
OpenNeuro ds004473 version 1.0.2 supplied the SEEG recordings, task events, behavior records, and channel metadata used in this fixed-stream schedule analysis.

Cited by

6 publications

Public catalog entries on Sciscape.

usesThree consecutive scores gave no clear pre-response detection gain in eight SEEG streams at an early target of one command per eligible minute
This test uses its preserved feature arrays, fixed models, source scores, and provenance. It also uses the proof that an earliest-candidate schedule has maximum count under a fixed wait.
usesA reset reduced mean detection and background commands in eight SEEG tests with rest training under a limit of six commands per eligible calibration minute
Its threshold timing result informed the full stream replay. We replay every calibration candidate because stateful command times can change with the threshold. Its result arrays are not used here.
usesTask background training increased mean retrospective detection area in eight SEEG streams under one-second waits and budgets of 0–6 background commands per eligible minute
We use its earlier result on changed command times under a waiting rule as background for the complete threshold method.
usesHigher thresholds increased unmatched command counts at some steps in eight fixed SEEG replays with a 1 s wait
This publication supplied the earlier proof that the greedy waiting rule has nonincreasing total command counts as thresholds rise, while exact command times can change. We use that result and extend the audit to target-labeled unmatched counts and early calibration on a new fixed forecast package.
usesWith timely replies, four-bit receiver checks preserved 2,163 SEEG commands and saved 54.8% of payload bytes in eight records
Used its evidence that the command wait can cause later command changes after a decision change. This motivated exact complete-sequence validation and the shared-state induction.
usesTask background training increased mean retrospective detection area in eight SEEG streams under one-second waits and budgets of 0–6 background commands per eligible minute
We use its earlier result on changed command times under a waiting rule as background for the complete threshold method.