A fixed test did not confirm that class tags improve forecasts of repeat disagreement in six SEEG patients and two fMRI studies

We tested whether a sound class tag improves forecasts of repeat disagreement beyond a separate response estimate. Public SEEG and fMRI studies supplied three estimates per sound. Development used two patients and four MRI adults. Fixed tests used six other patients, 16 B2021 adults and eight NH2015 adults. Five outer sound folds and three inner folds kept test sounds out of fitting. The target was log absolute disagreement between two estimates. The third estimate supplied response-size predictors. Mean gains were 0.0162%, 0.0331% and 0.0153%. Their 95% person t limits were [-0.0219, 0.0543], [0.0153, 0.0509], and [-0.0142, 0.0447] percent. The shared test of positive means failed, p=0.161972. Negative person scores and all specified controls remain. A later descriptive check against a constant gave mean gains below 0.1% for the full model in each study. An exact model gives positive class-tag value under one universal variance law, because a tag can improve knowledge of the true mean. It does not explain the human data. The contribution is a checked forecast assay and a failed shared benchmark. Clinical sampling, separate cohorts, source selection, shared schedules, only three estimates and different measurement filters limit inference. The package gives code, source versions and terms, all errors, uncertainty, figures, audits and repeat steps. No positive neural finding across both modalities or cause is established.

SEEG · fMRI · repeat disagreement · auditory cortex · prediction · uncertainty · negative result