A fixed acoustic distance rule failed a shared error rank test in 20 iEEG patients and 24 fMRI adults

We tested whether acoustic distance warns of neural encoding error beyond feature size and predicted response power. Separate public studies supplied signed iEEG local field potential and BOLD fMRI. Five patients and four adults supplied development. Fixed tests used 20 other patients, 16 B2021 adults and eight NH2015 adults. Models trained on 95 or 110 sounds and tested 48 or 55 other sounds. Mean adjusted rank scores were 0.07186, 0.13515 and 0.06606. Their 95% person intervals were [0.01992,0.12380], [0.09810,0.17220] and [-0.03816,0.17029]. The three-cohort rule failed, p=0.08879. All person-and-test-sound resample ranges included zero. Negative scores and all fixed controls remain. All 44 fixed unbounded response models had negative skill against their training means. A later exploratory input-clipping check gave positive skill in 41 people, but warning scores shrank or changed sign. A near-zero mean inflated one normalized variability descriptor in the MRI cough clip. An exact fixed-design example also gives growing distance with constant expected error. These checks limit interpretation; they are not fitted neural causes. Independent arithmetic checked all 44 primary people. The contribution is a reproducible failed warning benchmark and model diagnostic. It does not establish cross-modal warning generalization. Source selection, different cohorts and sounds, measurement noise, fixed fits and descriptor choices limit scope. Code, permitted electrical inputs, provenance, all outcomes, figures and repeat steps are supplied.

human neuroscience · iEEG · fMRI · acoustic encoding · model error · uncertainty · replication · negative result