Residual covariance at eight selected sites per person failed a shared test in six SEEG patients and two fMRI studies

We tested how much nonvoice repeat covariance remains after removal of any one spatial direction. Each person had eight sites selected for speech/song preference. Separate public studies supplied SEEG high-gamma and BOLD fMRI. One repeat and one sound half set site scales and choices. Other sounds and two other repeats supplied the score. For a fixed response mean and independent zero-mean repeat errors, the score has a conservative mean bound. Development used two patients and four adults. Reserved tests used six patients, sixteen B2021 adults and eight other NH2015 adults. Residual means were SEEG: 0.014058 [-0.020425, 0.048541], B2021: 0.00294718 [0.000380704, 0.00551366], NH2015: -0.00137457 [-0.00381578, 0.00106664]. Brackets give pointwise 95% person Student intervals. The fixed shared rule failed, p=0.887618. It required positive means and positive-score majorities for both residual and held-out preference in every study. All negative results and fixed checks remain. The bound needs two test repeats for any fixed number of removed directions. It bounds the expected score, not each record. The score does not give unbiased signal eigenvalues. Screening with test repeats and shared nuisance gave positive scores in designed examples. One nonlinear scalar also gave a positive residual. Thus latent neural rank and common causes are not identified. Prior source screening, sparse clinical coverage, short baseline precision, cohort differences and vascular filters limit scope. Code, provenance, all results, figures, method proof and repeat steps are supplied. Original neural arrays remain external.

auditory cortex · SEEG · fMRI · repeat covariance · linear rank · source selection