Investigating whether diagnostic AI false positive predictions contain pre-clinical disease signals. Mining MIMIC-IV sepsis data and CRC liquid biopsy discordances to find patients whose false positive flags preceded actual diagnosis by months. If validated, this turns model errors into novel early biomarkers.