Valid Per-Field Selective Risk Control for Document Extraction: Three Failure Modes, a Validity Ladder, and When Conditioning Pays Paper • 2608.14639 • Published 28 days ago • 2
Evaluating Tool-Using Language Agents: Judge Reliability, Propagation Cascades, and Runtime Mitigation in AgentProp-Bench Paper • 2604.16706 • Published Apr 17 • 1
Evaluating Tool-Using Language Agents: Judge Reliability, Propagation Cascades, and Runtime Mitigation in AgentProp-Bench Paper • 2604.16706 • Published Apr 17 • 1