A 200 ms baseline gave 53 to 61 percent control calls in two SEEG recordings at a nominal 5 percent limit

We tested whether an early baseline can set a response cutoff across six SEEG contacts. The data were 630 pulses at 63 surface stimulation sites in two public recordings from OpenNeuro ds004774 release 1.0.0. A detector drew 4999 averages from a 200 ms baseline and used a nominal 5 percent cutoff for the maximum across contacts and time. It called 20 of 33 and 16 of 30 sites in a separate window before the pulse. A check that kept pulse alignment also failed the control. Response AUC ranged from 0.772 to 0.896 against three expert label sets, but the fixed cutoff called 68 to 83 percent of expert negatives. The output is a reproducible calibration failure case and audit tool for BCI response detectors. The result applies to this detector and these two recordings. Sites share contacts, so bootstrap intervals are descriptive. The package preserves code, compact native voltage, source ranges and hashes, null and bootstrap draws, tables, three figures, and reproduction instructions. All 630 source ranges were freshly checked and rebuilt to exact input equality. An independent calculation and a full numerical replay passed.

SEEG · brain-computer interface · evoked response · baseline calibration · signal detection · OpenNeuro ds004774