Reference ranks matched sounds above chance at 24 selected sites in six SEEG patients and two fMRI studies

We tested two ordinal readers of repeated natural sounds in separate public human studies. Reference ranks used 64 other sounds at each site. Order votes used only the preference between two sounds at each site. Both rules have exact invariance to strictly increasing functions of processed responses, with fixed site labels. Development used two SEEG patients and four fMRI adults. We added the vote claim after development. A local code lock came before reserved tests. Tests used six other patients, 16 B2021 adults and eight NH2015 adults. At 24 sites, reference-rank mean accuracy was 59.25%, 80.12% and 81.83%. Their 95% person Student limits were [56.79,61.71]%, [77.90,82.33]% and [76.77,86.89]%. Order-vote means were 57.11%, 80.61% and 79.38%. Both shared tests passed Holm correction over three claims, p=0.046875. A fixed five-point retention test against standardized values failed, p=0.598190. Original and added claims, all people and controls remain. Short-baseline subtraction reduced electrical scores. An exact example showed loss of within-interval order under finite reference ranks. A zero-signal selection model gave above-chance scores. Thus these results support readers for selected processed measurements. They do not isolate a native neural code or a shared cause. Source screening, sparse clinical coverage, different ages, tasks and measurement filters limit scope. Code, source pins and terms, all results, figures and repeat steps are supplied. Original neural arrays stay external.

human neuroscience · SEEG · fMRI · ordinal decoding · reference calibration · monotone invariance · source selection