Two energies bounded power changes from linear filter state in 35 SEEG records with complete input

Stored linear filter state can change later SEEG power. We tested this effect in OpenNeuro ds006519 version 1.0.0. The release had 35 records from 21 patients. The primary set had 1,659 channel occurrences off the stimulation shaft. We used fixed causal filters and complete later input. All power windows used samples after the cut. For a known state, two output energies gave an upper bound on the current power change. A rule that used only the stored state energy omitted a cross term. It gave 75,013 false passes across 1,448,307 primary post readings in three bands. These failures occurred in all 35 records. The full bound gave no false passes in this test. At the fixed primary endpoint, 250 ms after the marked post start, 96.7 percent of gamma readings met the five percent bound. Each patient had equal weight. The descriptive bootstrap interval was 95.2 to 98.0 percent. Pre controls had similar effect times. An added exploratory initializer reduced some early changes but did not restore full state. The package preserves a streaming tool, code, derived data, five figures, source hashes, fixed object versions, local plans, and reproduction steps. The method uses established mathematics. The results concern filter processing. They do not measure neural recovery or decoder gain.

filter state · signal processing · power bound · SEEG · brain-computer interface · causal filter