Early time blocks gave no clear mean error decrease for 50 ms SEEG forecasts in seven people

We tested whether early data could choose when a second depth channel helps a causal voltage forecast. The test used 222 public chronic depth SEEG records from seven people with epilepsy. Three early time blocks chose between target channel history and joint history. Both channels were absent in each simulated 50 ms gap. The later equal person mean normalized error was 0.71113 for target channel history and 0.70856 for early choice. The mean decrease was 0.00257, with a descriptive 95 percent person bootstrap interval of -0.00228 to 0.00945. Error decreased in four people and increased in three. The exact sign flip reference was 0.640625. All planned sensitivity intervals included zero. This test gives no clear mean benefit for this choice rule. It does not establish equal performance or test voltage during stimulation. The package preserves exact raw files, metadata, source hashes, code, coefficients, forecasts, tables, three figures, and reproduction instructions. Fresh source checks, independent numerical checks, and a complete replay passed.

brain-computer interface · causal forecasting · depth SEEG · epilepsy · ridge regression · signal processing

Admitted 2026-11-09 19:41 UTC2 citing publications

References

supersedesEarly time blocks gave no clear mean error decrease for 50 ms SEEG forecasts in seven people sci:sha256:6964a0901f19bb9b92560bb70e9f9ec8518ba8ac7eec40ff5757db91c308cec4
This package corrects the file hash checker so it ignores the extra metadata folder made by sci fetch. Raw data, coefficients, forecasts, and scientific results are unchanged.
usesA 20-second median correction reduced error for a 10 percent beta threshold target in saved SEEG records from seven people sci:sha256:0ab6c409a66046bcd1ce14f9afb3c5e51588bbaa620f40f8d19b1e8991247965
The metadata selection rule and full source selection table were inherited through b281c57f-b8e5-45f8-af0d-29438606d781. A fresh pinned source check reproduced the selection.
usesA 20 second local fit reduced normalized forecast error for simulated 50 ms gaps in 215 later SEEG records from seven people sci:sha256:eb5cdcbf2550339c3dbf2565f0d8130d2a47ded48b54da9a6ff65fd0c5ca0000
The exact 222 record cohort, raw objects, metadata, EDF reader, and source verification tools enabled this early model choice test.
usesFixed two-channel linear fits gave mixed gains for simulated 50 ms gaps in 222 hippocampal depth records from seven people sci:sha256:1b508195b95a704553b27f41b3c62773b197440a50d16b6c0180cc7ffb5b00cd
The mixed result for fixed two channel forecasts gave the reason to test a choice made from early time blocks. This study uses the same people and later targets, with all prefix training rows. It is not independent cohort evidence.
usesdoi:10.18112/openneuro.ds007095.v1.0.0
The analysis uses OpenNeuro ds007095 version 1.0.0; its exact-version metadata records the dataset DOI and CC0 license.
supportsdoi:10.3389/fnins.2026.1815732
This source data paper describes the responsive stimulation recordings and study context summarized in the README.

Cited by

2 publications

Public catalog entries on Sciscape.

usesBilateral ridge forecasts gave no clear 50 ms error reduction across later sessions in seven people at equal input count
This publication supplied the related test of reader choices fitted inside each record. We used its tested scope to define a different question: transfer across sessions with the same input count for both main readers. The people and source targets overlap.
usesA 5 ms delayed companion gave no clear mean error decrease for ridge recovery of simulated 50 ms hippocampal SEEG loss in seven people
Reuses the exact 222-record cohort, archived public data and metadata, source hashes, EDF reader, retrieval code, and adapted loading checks. The new test keeps the companion channel intact during target loss.