Marginal scores gave 39.1 percent site accuracy versus 39.2 for precision blocks at 20–100 ms after one known simulated pair loss in four SEEG sessions

We tested site identification after one known simulated bipolar output loss in public SEEG data from OpenNeuro ds004457 version 1.0.2. Four sessions retained a signal. Surviving voltages stayed unchanged. The test used four calibration pulses per site and 536 separate test pulses. The fixed covariance fraction was 0.5. At 20 to 100 ms, equal-patient mean accuracy was 28.1 percent with mean replacement and 39.1 percent with marginal scores. The observed precision block gave 39.2 percent. The primary marginal gain over mean replacement was 10.97 percentage points. All four patients improved; the exact two-sided sign-flip value was 0.125. The marginal difference from the precision-block control was -0.09 points. Its descriptive 95 percent patient bootstrap interval was -0.73 to +0.55 points. This test did not show a gain from covariance reinversion. It did not establish equal performance. Common conditional replacement gave 29.1 percent accuracy. Established block algebra identifies a site penalty from that replacement. We extend a published contact-loss test with distinct output-loss masks and mechanism controls. All 110 source hashes, fresh raw input reconstructions, 5,810 numerical checks, and a full replay passed. The package preserves code, inputs, models, decisions, three figures, source records, and reproduction instructions. This is an exploratory test on reused pulses. The site-wise time split is retrospective. The results do not establish population or live interface performance.

stereoelectroencephalography · brain-computer interfaces · signal decoding · missing outputs · Mahalanobis distance · marginal scoring · covariance · robustness