Lag relations repeated beyond site scores in 19 SEEG patients and 50 resting fMRI adults; a ten-person film test failed

We tested whether pairwise lag relations repeat after removal of one additive score per recording site. Separate public human studies supplied bipolar SEEG and BOLD fMRI. A difference of squared lag correlations gave a polarity-invariant pair field. We projected out the site scores and tested the residual field in two records per person. A local plan fixed the first test. In 19 SEEG patients, mean residual repeatability was 0.780, with a 95% person-bootstrap interval of 0.733 to 0.826; all 19 scores were positive. The fixed test across two films in ten fMRI adults failed: mean 0.011, interval -0.025 to 0.045, five positive scores, conjunction p=0.623. A separate follow-up was fixed before scores in 50 MICA-MICs resting fMRI adults. Two 150-second epochs, separated by at least 117 seconds of unused frames, gave mean 0.323, interval 0.292 to 0.355; all 50 scores were positive. The extension conjunction had Holm p=0.000003815 across two test opportunities. This supports limited within-record generalization across the two modalities. It does not confirm cross-film transfer or common anatomical edges. Full-strength copying of the first residual field reduced mean error by 53.7% in SEEG but increased it by 46.1% in resting fMRI. A delayed-common-input Gaussian model also gave repeatable cycles without links between observed nodes. Stable cycles therefore do not establish feedback or nonlinear neural dynamics. Clinical sampling, distinct time scales, source processing, hemodynamics and residual noise limit interpretation. The package preserves the failed test, fixed plans, source versions and hashes, code, all person scores, uncertainty draws, figures and an offline repeat.

SEEG · fMRI · lag correlation · Hodge decomposition · repeatability · common input · prediction error · human neuroscience · negative result