Separate references from two SEEG patients gave opposite temporal score signs in all 16 test fMRI adults for three sound groups

We tested whether a late electrical window carries more BOLD-aligned relative sound-group contrast energy than an equal early window. Separate public human studies supplied SEEG high-gamma amplitude and fMRI BOLD. Two patients and four MRI adults set references. A local code lock preceded tests in six other patients and 16 other adults. Three source groups formed the bridge; no item identity join was established. Early and late windows each had 40 samples. A cross-repeat kernel removed the isotropic contrast component and used common calibration. Mean late-minus-early scores were -0.01765 and -0.07423. Their pointwise 95% person Student limits were [-0.04292,0.00762] and [-0.09455,-0.05391]. The fixed shared positive rule failed, p=1. All 16 MRI differences were negative. Eight other MRI adults also gave negative differences under the same reference. Resampling reference and target people gave ranges across zero in both studies. All negative people and all controls remain. An unfitted equal-map model gave a mean late advantage of 0.691 with separate cosine normalization and about zero with the primary projection. Separate references from P1 and P2 gave opposite temporal score signs in all 16 test MRI adults. The reference omissions were planned; the person view was added after the primary result. The contribution is a temporal contrast assay, a measured reference-choice limit and a failed cross-study benchmark. It does not establish positive cross-modal neural generalization, absent late coding or causality. Sparse clinical sampling, two reference patients, source screens, group pooling, unequal error and different cohorts limit scope. Code, provenance, derived statistics, figures and repeat instructions are supplied. Original neural arrays and audio remain external.

human neuroscience · SEEG · fMRI · auditory cortex · temporal response · crossvalidated kernel · negative result · measurement error · reproducibility