List your company

LLM-Agent Oversight on HotpotQA

0.436Scalar RegretScalar routing
0.41420.42510.4360.4469Jun 19, 2026

Evaluation results

MethodLinks
Scalar routingIntervention Type=Scalar2026.060.436———
Action-conditioned (prefix) routingIntervention Type=Action-conditioned (prefix)2026.06—0.4170.0190.02