The remote crossing difference interval included zero for mean and median references after synthetic pulses in 35 SEEG records

We tested reference effects on a known one-contact transient in public SEEG data. We used the first interictal record from each of 35 eligible patients in OpenNeuro ds004100 version 1.1.3. We tested 83,760 injections per reference on 4,188 contacts. The primary Gaussian pulse had a 10 ms time standard deviation. Its peak was 100 times the median stored calibration scale. The scale was 1.4826 times the median absolute deviation in the first 5 seconds. The detector threshold was 6 reference-specific scales. A remote target had a different inferred shaft name. A new crossing was present after injection and absent in the unaltered control. Rates had equal patient weights. The mean-reference rate was 1.593 percent. The median-reference rate was 1.292 percent. The paired difference was 0.301 percentage points. Its 95 percent patient bootstrap interval was -1.005 to 1.264 percentage points. The interval assumes independent, exchangeable patients and this fixed design. It included zero and did not establish a cohort mean reduction. Median reference had lower rates in 31 records. Three tied. HUP117 had a higher median-reference rate. Fixed checks showed dependence on pulse amplitude and threshold. A voltage check added after outcome inspection showed that a bounded median shift could exceed the mean shift. The package preserves native input samples, fixed source versions, exact median update code, crossing identities, and bootstrap draws. It also includes repeat instructions. Direct calculations and a complete offline rerun passed. This is a detector stress test. It does not measure clinical false positives or neural signal accuracy.

SEEG · intracranial EEG · median reference · mean reference · transient detection · threshold crossings · synthetic pulse · signal processing · bootstrap