Fixed tests did not confirm transfer of load readers between encoding and delay in 12 SEEG patients and two fMRI samples

We tested transfer of load readers between encoding and delay. Public studies supplied human SEEG power and BOLD fMRI. Three patients and three MRI participants supplied development. A local code lock preceded new scores in 12 patients and two MRI samples. Those samples had 16 and 12 participants. Phase calibration used training data. Regressions controlled recorded trial history. Exact bounds allowed either order of repeated MRI conditions. Both transfer directions needed positive mean lower bounds in every sample. The shared test failed, p=0.999919. Mean encoding-to-delay lower bounds were +0.04248, -0.18797, -0.17632, in that order. Reverse bounds were -0.01226, -0.10900, -0.11736. Nominal MRI means were near zero. Negative lower bounds do not show reversed coding. All uncertainty intervals and negative results are supplied. In an exact model, identical ranks gave full load-response vectors with cosine -0.9999. The model allowed free offsets common to all sites. Thus rank transfer cannot identify the full response direction under those assumptions. This is a checked method and a negative benchmark. It does not show absent load information, neural code rotation or causality. Clinical sampling, separate cohorts, source timing assumptions, selected MRI regions, unknown repeat links and measurement filters limit scope. Code, compact electrical inputs, source terms and pins, results, figures and repeat steps are supplied. Exact MRI author rows remain external.

working memory · SEEG · fMRI · phase transfer · spatial ranks · trial history · partial identification · reproducibility