Allowed repeat exchanges reversed within-load speed slopes in 28 fMRI adults; the SEEG mean interval crossed zero

We tested whether a neural score trained on memory load tracks response time within the same task and load. Separate public studies supplied SEEG from 12 patients and fMRI from 16 and 12 adults. Readers used other runs or an earlier session. A local plan fixed the speed endpoint and trial-link audit before new scores. The nominal mean slopes were -0.002605, -0.004014 and +0.004565 load-score units per residual log response time SD. All 95% person Student intervals crossed zero. The fixed positive cross-modal test failed. The reused fMRI tables did not verify which of two neural repeats came first within a condition. With full condition controls, exchanging every pair exactly reverses the speed slope. The same exchange preserves all-neural-row load AUC. In the measured records, one exchange reversed the nominal sign in 20 adults; two were needed in eight. No neural value or load label changed. These are sensitivity assignments, not detected source errors. Load AUC means were 0.657 in SEEG and 0.887 and 0.762 in fMRI. Load reading therefore did not validate this behavioral use. The new contribution is a checked sign audit, exact minimum exchange counts and a complete negative benchmark. It does not establish absent neural coupling or positive cross-modal generalization. Code, licensed compact inputs, source versions and terms, all results, figures and repeat steps are supplied. Clinical sampling, unequal tasks, source selection, phase timing and measurement filters limit scope.

human neuroscience · SEEG · fMRI · working memory · response time · trial links · partial identification · negative result · neural decoding · reproducibility