Bounds from known filter states gave longer beta than gamma reset delays in 35 SEEG records

We tested a rule that limits the voltage change from a causal filter reset. We used all 35 released SEEG voltage records from 21 people in OpenNeuro ds006519 version 1.0.0. The rule used the known filter state and a past baseline. A proof bounds the RMS change in every later block of approximately 25 ms to 10% of baseline band RMS. The proof assumes stable linear filters, identical later inputs, and an exact known state. At marked poststim window starts, the mean person-level delay was 229 ms for beta (13 to 30 Hz) and 82 ms for gamma (70 to 110 Hz). The paired difference was 146 ms, with a 95% bootstrap interval of 130 to 158 ms. Beta had the longer delay in all 21 people. At 100 ms, 5.8% of beta channels and 77.1% of gamma channels met the bound under equal-person weighting. The source does not give exact stimulation train ends. The rule limits filter reset effects. It does not establish neural signal quality or decoder accuracy. The package preserves the proof, code, compact states, voltage crops, results, figures, source hashes, and a pinned retrieval recipe. Exact compact replay and all numerical checks passed.

filter state · observability Gramian · SEEG · signal processing · filter reset · voltage bound · brain-computer interface