Fitted reference weights failed a shared prediction test in 20 iEEG patients and 16 fMRI adults at 48 calibration sounds

We tested whether a short calibration record can set useful weights for other people's sound response maps. Separate public studies supplied signed iEEG LFP means and BOLD fMRI. Five patients and four MRI adults supplied development. A local code lock preceded new tests in 20 other patients and 16 other adults. Each person supplied 48 calibration sounds in eight fixed splits. The score tested response differences within source sound classes. Mean prediction gains over equal reference weights were -0.00015% and -0.10033%. Their 95% person Student limits were [-0.04831,+0.04802]% and [-0.22827,+0.02762]%. The single shared gain-and-skill rule failed, p=0.942314. Mean skill over category-constant prediction was positive in both studies. Calibration map error fell, but this fit did not give a confirmed prediction gain. Separate MRI target estimates gave -0.50854% mean gain. Eight adults from another MRI study gave -0.54650%, with the same MRI sounds. All negative people and checks remain. An exact noiseless model also lowered calibration map error while raising test response error. This model does not explain the human scores. The contribution is a fixed reference-weight method, a negative human benchmark and reusable residual summaries. Small reference groups, clinical coverage, source screens, different libraries and measurement filters limit scope. No positive cross-modal personalization effect or common cause was confirmed. Code, source pins and terms, figures and repeat steps are supplied.

human neuroscience · intracranial EEG · fMRI · auditory cortex · reference weighting · kernel prediction · negative result · reproducibility