Shared bipolar signals gave median directed scores below 0.00004 nats in independent AR(1) models from two SEEG records at 128 Hz

We tested whether shared bipolar subtraction can give directed prediction from independent contacts. An exact AR(1) example gave 0.0488588 nats in one direction and zero in reverse. Both predictors used the full past. We calibrated independent Gaussian AR(1) models from both complete resting SEEG records in OpenNeuro ds005670 version 1.0.0. At 128 Hz, 192 and 149 shared pairs gave median larger scores of 0.00003537 and 0.00003105 nats. Five pairs reached 0.01 nats. They used two contacts with strong 50 Hz power. An exploratory 50 Hz filter reduced all shared scores below 0.01 nats. Disjoint and equal-coefficient controls gave numerical zero. All 653 primary models passed a separate Riccati solver. A second raw-data extraction reproduced all arrays and result tables exactly. AR(1) did not reproduce higher-lag contact dynamics. Independence was imposed, and only two patients were tested. These scores do not measure neural connections or false-link rates. The archive contains code, moments, proofs, source hashes, retrieval instructions, results, and validation.

bipolar signals · Granger prediction · AR(1) models · SEEG · signal processing · reproducibility