A positive covariance on EEG contrasts permits either sensor partial correlation sign when there are at least three sensors and the common signal is unrestricted

EEG sensor differences do not determine the sign of a full sensor linear partial correlation when the common signal can depend on the contrasts. We prove that every value strictly between -1 and +1 is possible for each sensor pair. The proof requires at least three sensors and positive variance in every contrast direction. We give a formula for all compatible positive sensor covariance matrices and code to construct finite examples. The primary test uses 88 public EEG records from OpenNeuro ds004504 version 1.0.9, 19 sensors, and an explicit contrast ridge of 0.001. All 15,048 pair-and-record cases have two examples with the same contrast covariance: one gives a partial correlation below -0.94979, and the other gives a value above +0.95010. All 30,096 examples pass direct covariance and regression checks. A check without the contrast ridge gives both signs in all 6,156 cases from the 36 records with full contrast rank. Fresh downloads of two source records reproduce 268 band matrices exactly. A repeat run reproduces every saved numerical result. The positive examples use a common signal that is more than 99% predictable from the contrasts. Each pair has separate examples. The result does not assign all signs at once or identify a physical reference signal. The package preserves the proof, code, data, source hashes, results, and reproduction instructions.

EEG · sensor covariance · partial correlation · reference invariance · contrast covariance · common signal · identifiability · covariance completion