A public, reproducible benchmark is not published yet.

A useful benchmark would state target count, true positives, false positives, false negatives, precision, recall, exploit-validation rate, and time-to-detection — plus enough methodology for someone else to reproduce the run.

No public dataset or score is available. When we publish one, it will use labeled lab targets (Nubesti Security Lab), not unnamed “enterprise customers,” and it will include the methodology next to the numbers.