Reproduce “Dream2Discern: Learning from Hard Latent Variants for Runtime Failure Prediction in Multi-Agent Systems” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Boxuan Zhang, jianing Zhu, Xiaotian Han, Ruixiang Tang
Dream2Discern predicts whether an agent run will fail from partial recorded interaction histories. During training, it perturbs the auditor’s query and value representations to raise prediction loss, then trains on the perturbed representations with outcomes; at inference, it removes the perturbation. On StepShield, Qwen3-4B reaches 90.28 ± 0.46% wACC versus 79.17% for AgentForesight-7B, measuring prediction rather than intervention benefits.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.