Fiddler's Centor runtime routes uncertain AI evaluations to a reasoning model

A Fiddler team member says the Centor Evaluation Runtime uses a fast model to score every span or trace. It attaches an uncertainty score and sends cases outside its competence to a reasoning model for re-scoring. The team member says the escalated cases can reveal where the fast model is weak or traffic is drifting; labeling them for future versions can reduce escalations over time.