Corrected conformal calibration by scan date returned all selected SEEG pairs at 90 percent in seven ds003498 patients

We tested candidate sets for high frequency oscillation count lists in public SEEG data. OpenNeuro ds003498 version 1.1.1 supplied 135 records from seven patients. Each record had 300 seconds. The target list kept all ties at the top 20 percent count boundary. At the 90 percent level, calibration by scan date returned all selected pairs. The corrected quantile needs nine calibration dates before it can select fewer than all pairs. Only zero to four calibration dates remained after fitting and target dates were removed. Calibration by record retained 73.78 percent of pairs under equal patient weights. Its descriptive 95 percent patient bootstrap interval was 49.98 to 92.03 percent. The package preserves source marks, complete primary sets, code, a rank proof, and nine fixed checks. Independent parsing checked every count row. A separate run matched all 15 numeric output files exactly. The exhaustive coverage result follows from a rank identity. It does not validate future coverage. A future guarantee needs exchangeable calibration and target units conditional on fitting data. The target is a detector-count list. The study does not measure tissue labels or clinical accuracy.

conformal prediction · stereoelectroencephalography · high frequency oscillations · whole-list prediction · calibration · uncertainty quantification