The 18 SEEG records in ds005398 version 1.1.3 could not support a surgery outcome AUC

We tested whether the public SEEG subset of OpenNeuro ds005398 version 1.1.3 could support HFO removal evaluation against surgery outcomes. Only three of 18 patients had an available outcome. All three had Engel class I. The other 15 had no recorded resection and had unavailable outcomes and channel resection labels. Correct outcome AUC was undefined. A deliberate replacement of unavailable outcomes with zero gave surgery status an AUC of 1.00. Replacing both unavailable outcomes and channel resection labels with zero also gave STE, MNI, and channel removal fractions an AUC of 1.00. These were invalid controls, not prediction results. All 387,097 source event rows had resection flag zero. After duplicate removal, 25,170 full-record marks disagreed with known channel resection labels. The files did not resolve anatomical truth. Five fixed settings gave the same AUC pattern. The package preserves all used source payloads, a pinned retrieval recipe, code, results, and an outcome feasibility tool. Fresh public downloads verified 94 payloads. Separate decimal counts, rank AUC calculations, and clean rebuilds agreed. The claim concerns this release only.

stereoelectroencephalography · high-frequency oscillations · epilepsy surgery · surgery outcome · resection labels · AUC · OpenNeuro · reproducibility