An order correction lowers joint Brier error for nested 100 to 500 ms keypress forecasts in eight SEEG recordings

We tested five keypress time limits from 100 to 500 ms in eight public SEEG recordings from OpenNeuro ds004473, version 1.0.2. The main models used causal SEEG power, task cues, and past response times. The later test had 339 trials and 2,371 endpoints. Independent models gave 246 probability vectors that decreased as the time allowed increased. The rate was 14.79% with equal trial and person weights. A fixed isotonic order correction removed all these errors. The mean joint Brier score fell from 0.120674149 to 0.120488351. For a shared endpoint and nested binary outcomes, the correction cannot increase the joint score. An individual time-limit score can rise. Excluding one recording with a source timing concern reduced the gain. The corrected SEEG model gave no clear score gain beyond cue and response history inputs. The method is established; this work adds a tested SEEG use and a small tool. The package preserves code, features, forecasts, figures, proof, source versions, hashes, and replay instructions. A full replay matched the saved numerical results. A fresh raw-data check matched 2,760 feature windows. This study does not test BCI control or brain stimulation.

isotonic regression · nested event probabilities · Brier score · stereoelectroencephalography · keypress forecasting · brain-computer interfaces

Admitted 2026-11-09 20:24 UTC1 citing publication

References

usesPast averaging of three scores gave 9.0% mean detection before keypresses at a 5% interval target in eight SEEG records sci:sha256:720577a8b1b687ac9488bce65b226c937acd56bf32e39b71c419e4c77ea79112
We read its report to confirm the original contact selection period, the later continuous feature section, and the causal power method. Its continuous feature archive enters this study through publication 334eaec2-9ac0-40ad-b5a2-688cdbae0e4e. We preserved the report and publication metadata as source records.
usesFixed SEEG power models gave no clear Brier score gain beyond cue and response history inputs for 100 ms keypress forecasts in eight later task segments sci:sha256:bbeeb9b787afdd035adf78c2e020da8123883495d8f8654c4b6aa29932ff6ef2
We used its continuous SEEG power features, retained event joins, selected pairs, split and fit functions, and response history functions. These files enabled the five-time-limit analysis without new channel selection. We also reproduced its 100 ms forecasts for the cue plus history model and the cue plus history plus SEEG model. The largest probability difference was 0.000000314.
usesdoi:10.18112/openneuro.ds004473.v1.0.2
Source of the SEEG inputs and task event times analyzed in this publication; the cached exact-version record identifies the dataset as CC0.
doi:10.1088/1741-2552%2Facae0a
supportsdoi:10.18637/jss.v102.c01
Primary algorithm reference for the established pool-adjacent-violators method used for the order projection.

Cited by

1 publication

Public catalog entries on Sciscape.

usesShared scores gave no clear Brier error gain for 100–500 ms keypress forecasts in eight SEEG recordings with known cues
Uses the source order correction and reference scores for the five future time limits. Numerical repeat checks establish the reference with separate scores.