Shuffles across all time windows rejected 5.41 and 5.68 percent of log beta power controls with independent orders within ten-second blocks in two SEEG records

We tested time shuffles for log beta power correlation in two public resting SEEG records from OpenNeuro ds005670 version 1.0.0. Each record supplied a 120-second segment and 128 fixed pairs of bipolar channels on different shafts. The conditional model kept each ten-second block's power values and used independent orders within blocks. Each edge had 5,000 independent probes and 1,999 references for each rule. Shuffles across all windows rejected 5.410% and 5.682% of block probes at a 5% threshold. Matching block references, with correlation centered on its exact null mean, rejected 4.942% and 4.927%. Average beta differences were modest. Full shuffle edge rates reached 20.22% and 29.16%. A planned alpha check gave a larger average difference in the second record. Exact formulas show how block means and local variances set the conditional correlation moments. This is a finite control benchmark. It does not prove that block shuffles are valid for arbitrary SEEG or that the original signals are independent. All planned checks are reported. Fresh source hashes, independent decoding and power calculations, exact enumeration, and a primary repeat passed. The package preserves code, source versions, hashes, raw examples, power arrays, edge maps, full primary draws, all results, proof, figures, and reproduction steps. It reuses source code and the same two patients from the cited parent. Earlier features and results are not inputs.

SEEG · power correlation · block permutation · surrogate data · Monte Carlo test · conditional null model