Tests with fixed selected pairs and delays rejected 99.2 percent of uniform circular audio rotation controls at a five percent threshold in five SEEG patients

We tested contact pair and delay selection with public speech data from five SEEG patients. The data came from OpenNeuro ds004703, version 1.1.0. Each evaluation control independently rotated audio within each passage. The model used uniform circular shifts. At a five percent threshold, tests that kept the selected choices fixed rejected 99.215 percent of controls. Tests that repeated selection in reference controls rejected 4.997 percent. These are equal patient mean rates. A second seed gave rates of 99.164 and 4.842 percent. Separate early-control selection gave 5.143 percent in the primary run. Each patient and seed used 9,999 reference controls and 20,000 evaluation controls. The general selection problem and maximum-score correction are established. This study adds a measured SEEG benchmark and an audit package. Independent direct calculations checked all 16,620,120 passage correlations and all saved result values. A clean offline repeat produced identical result hashes. Fresh raw data rebuilt one retained passage per patient with exact feature matches. The package preserves code, features, source metadata, retrieval recipes, all control draws, choices, scores, probabilities, figures, and instructions. The result applies to the defined conditional rotation model. It does not establish a biological null model, a neural speech source, or a clinical effect.

stereoelectroencephalography · speech processing · selection bias · circular rotation controls · calibration · maximum statistic · Monte Carlo testing