A 10 ms delay matched offline constant code flags for 250 ms SEEG windows in two R1286J recordings under an 11 sample rule

We tested a proposed power reader in two public SEEG recordings from R1286J in OpenNeuro ds005558 version 1.0.0. Each output used 250 samples at 1000 Hz. We marked runs of at least 11 equal native codes. The main rule withheld outputs with marked inputs on at least 103 of 114 channels. A rule that checked only the current sample released 3,021 and 9,806 such outputs in the two sessions. These were 3.25% and 13.54% of outputs in the next marked word intervals. They occurred in 43 of 68 and 53 of 53 intervals. No such outputs occurred in 121 selected control intervals. A streaming rule delayed each decision by 10 samples. All 1,002,588 tested raw decisions matched the offline rule, including each channel flag. A proof shows why this delay is sufficient and can be necessary. The rule assumes samples arrive in order and treats marked codes as invalid inputs. The package has source code, pinned source checks, compact data, all fixed settings, and three figures. There is one participant. This study does not measure neural recovery, memory change, or the original classifier output.

SEEG · brain stimulation · signal processing · input validity · constant code runs · streaming detection · artifact flags