Clarus Metrics PRO

ACI Conformance Benchmarks Independent eval architect Inference drift • boundary intelligence • model safety Scope termination and stopping behavior instruction boundary discipline inference drift and collapse surfaces VL-JEPA failure modes (video reasoning) alignment via behavior, not narrative Artifacts 30+ public micro-benchmarks (20-row probes) lightweight probes for internal evaluation focus on failure detection over accuracy Not trying to sell models Not training, not fine-tuning, not a lab. Evaluating how systems fall apart — before deployment. Open for collaboration Research groups • safety teams • eval units • robotics • multimodal inference Contact email / HF messages team@loopwell.ai