We measure whether AI models retrieve facts or fabricate them — across 15 domains, from foundational to frontier difficulty.
PROOF OF METHOD · 18 OPEN-WEIGHT MODELS · ONE INSTRUMENT, ONE SET OF CONDITIONS · EVERY SCORE PUBLISHED
FOR REGULATED ORGANIZATIONS
We run this same measurement on the model you host and deliver signed, dated evidence for your governance file. No customer data required, nothing installed, fixed price.
Model Validation →
Reliability: 0–100, composite of recognition and accuracy scores · Recognition: pre-generation signal · Accuracy: generation signal
Put this instrument on your model.
One email gets you the complete sample report. Same-day reply, no call required.