Pair calibration lowered PAC forecast error in 49 SEEG patients and 16 MEG participants; the shared skill test failed

We tested a sparse forecast of reciprocal phase-amplitude coupling (PAC) differences in separate human SEEG and MEG cohorts from one public source study. Four reference sites supplied a site-score estimate. Up to 256 separate pairs set one scale factor. Other pairs supplied test loss. Eight patients and three MEG people supplied development. A local code lock preceded scores in 49 eligible reserved patients and 16 reserved MEG people. Calibration lowered loss in all 65 people versus the unscaled estimate. Mean gains were 30.35 and 53.92 points of zero-forecast loss. Their 95% person Student limits were [26.27,34.43] and [45.57,62.27]. The shared gain rule passed two-claim Holm correction, p=6.47e-10. The stronger rule required positive mean skill against a zero-difference forecast in both modalities. It failed, p=0.07319. MEG mean skill was 4.01 points, with Student limits [-1.57,9.59]; its bootstrap range was positive. The fixed rule was retained. A six-centimetre check left only three eligible SEEG patients. An added local-coupling control left the MEG added gain uncertain. An exact noise reference permits a 50-point gain with zero signal skill. All 78 input sets matched fresh source reads. Code, source pins, data terms, all results, figures and repeat steps are supplied. This is a method and benchmark result for processed estimates in one record per person. It does not confirm a common neural circuit, transmission direction, causality, new-study replication, or useful sparse prediction in both modalities.

phase-amplitude coupling · SEEG · MEG · matrix completion · calibration · cross-modal test · negative result