Under a fixed mean test, load contrasts differed by phase in later sessions of 12 SEEG patients and two fMRI samples

We tested whether high-minus-low load response vectors share structure and differ between encoding and delay. Separate human studies supplied SEEG and fMRI. Cross-block products used scale estimates from an unused third block. First tests used 11 SEEG patients and 16 and 11 fMRI adults. All shared claims failed. Phase-change means were 0.03422, 1.15524 and 0.36559. A separate fixed follow-up processed all 42 unused electrical sessions in the 12 reserved patients. 39 sessions in 12 eligible patients entered its primary; sessions were averaged within people. The new electrical phase-change mean was 0.12172, with 95% person t interval [0.03002, 0.21342]. The later-session phase-change test passed the fixed numerical rule in all three study groups. Its four-claim Holm p was 0.0277974. Shared-structure tests failed. Correct-only and history checks were uncertain. The follow-up reuses people and known MRI scores; it is not independent cross-modal confirmation. A common-code simulation gave positive ordinary squared distance without true change. A gain change alone gave positive corrected change. The nominal shared-structure t test also had an 8% false-positive rate at a 5% level in one specified noise model. Source selection, adaptive trial history, measurement filters and distinct cohorts limit inference. The package adds 3.40 MB of compact phase data from 6.11 GB of checked raw inputs, code, plans, all scores, figures, audits and repeat steps. It does not establish a common neural code or causality.

working memory · SEEG · fMRI · cross-validation · load contrast · phase change · public data · replication