Dream2Discern: Learning from Hard Latent Variants for Runtime Failure Prediction in Multi-Agent Systems

A paper titled "Dream2Discern: Learning from Hard Latent Variants for Runtime Failure Prediction in Multi-Agent Systems" proposes training runtime auditors to predict terminal outcomes of LLM-based multi-agent systems from observed execution prefixes. The work addresses the limited corpus of trajectories with verified outcomes, noting that expanding that corpus requires costly execution and verification.

Reliable deployment of LLM-based multi-agent systems calls for runtime auditors that predict terminal outcomes from observed execution prefixes. These auditors typically learn from a limited corpus of trajectories with verified outcomes. Expanding this corpus requires costly execution and verificati