At a 5% site threshold, 9 of 28 iEEG and 18 of 19 fMRI reference people had a cortical Hurst map flag

We tested whether a 5% site decision also gives a 5% decision for a complete sampled cortical map. Public Hurst derivatives supplied separate electrical and fMRI reference cohorts. A fixed person split kept fitting, calibration and testing separate. The site rule flagged 9 of 28 reserved iEEG maps and 18 of 19 reserved fMRI maps. Exact 95% person intervals were 15.9–52.4% and 74.0–99.9%. The shared excess test gave p=0.00000565. Complete-map calibration reduced flags to 3 of 28 and 3 of 19, but wide intervals did not verify 5% error for either fitted rule. A later uniform four-site calculation gave mean flag chances of 17.4% and 21.3%. The established tolerance bound needs 59 independent calibration maps for a finite calibration-maximum threshold with 95% confidence of at most 5% map error. The reserved sets had 30 and 19 valid maps. Artificial shifts showed the detection cost of stricter decisions. An electrical median control had too few calibration maps for a finite threshold. These failures remain. The contribution is a new reference-map benchmark and a small calibration tool, not a new statistical theorem. Electrical sites were judged normal in people with epilepsy; fMRI people had healthy control labels. Cohorts, sampling, averaging and signal processing differ. No shared neural mechanism, disease diagnosis or causal effect is established. The package preserves public inputs, code, plans, source hashes, all results, four figures and repeat steps. Fresh source reconstruction and independent numerical checks passed.

intracranial EEG · fMRI · Hurst exponent · normative mapping · multiple testing · conformal prediction · tolerance limits · human neuroscience · reference calibration