Paper: Jev matches GPT-6 Astra on RAG claim checks within 0.6 pts at ~187× lower cost, but lets more unsupported claims through.

Benchmark/paper thread with media. rohanpaul_ai summarizes a study comparing Jev (label+probability) to GPT-6 Astra on 495 human-labelled RAG claims, plus a cascade. Jev was within 0.6 accuracy points at 187× lower cost but approved 23.2% unsupported claims vs Astra 14.1%. Advice: measure false accepts, not only headline accuracy.