Shaft-pair shrinkage had 1.87 percent higher mean squared z error than global shrinkage in 18 SEEG test patients at 1–80 Hz

We tested two rules to predict signed correlation in a separate 60-second SEEG record. SEEG means stereoelectroencephalography. Each rule mixed the source correlation with a mean from source links. One mean used the same two electrode shafts. The other used all shafts. Donor links used neither channel of the target link. Each rule fitted its weight on 18 patients and tested 18 other patients from OpenNeuro ds004100 version 1.1.3. At 1–80 Hz, the shaft rule gave 1.87 percent higher mean squared Fisher z error. The 95 percent patient interval allowed 3.61 percent lower error to 8.12 percent higher error. This interval assumed independent, exchangeable test patients and fixed training weights. Seven test patients improved. Removing one test patient reversed the mean difference. We report all four fixed secondary checks. We rebuilt all 72 input maps from fresh source byte ranges. All maps matched exactly. The package preserves code, inputs, source hashes, predictions, results, and instructions. The test measures agreement with a finite recording map. It does not establish clinical benefit or the absence of a useful shaft effect.

stereoelectroencephalography · SEEG · correlation prediction · shrinkage · Fisher z · electrode shafts · patient bootstrap · OpenNeuro