Model verification compares endpoint behavior against official model fingerprints, while benchmark scores serve as a reference for capability comparison rather than a promise about any one provider.
This matters when you buy access through gateways, proxy layers, or region-specific resellers and need evidence that the endpoint behaves like the advertised model family.
The public product is designed to show verdicts, latency, drift, and stability signals without revealing protected prompts, probes, expected answers, thresholds, or internal rule ordering.