Large Language Model (LLM)‑driven multi‑agent systems face a distinctive reliability challenge: actions taken by upstream agents can affect the correctness of downstream agents. Prior work has examined error propagation and tried to estimate uncertainty inherited from upstream agents, yet these approaches do not explicitly model how realized sender‑receiver interactions contribute to downstream error risk under different sender correctness states.
In this work we study agent‑level error‑risk estimation from execution traces. We propose Interaction‑Effect Propagation (IE‑Prop), an interaction‑conditioned model that learns separate effects for correct and incorrect upstream agents. For each sender‑receiver interaction the model combines these conditional effects using the sender’s estimated error risk, and recursively propagates the resulting risk estimates through the execution graph.
We evaluate the method on three multi‑agent architectures and three benchmarks. Across all architectures IE‑Prop raises the final‑node AUROC by roughly 0.05 points over the established baseline. Moreover, tracking risk across multiple steps adds an additional ~0.03 AUROC gain in sequential systems.
These results demonstrate that realized inter‑agent interactions provide predictive information beyond receiver‑local evidence, and that the benefit of multi‑step risk tracking depends on the system’s information‑flow structure.
Review: IE‑Prop’s conditional modeling of upstream correctness offers a finer‑grained view of risk propagation, making it a valuable tool for improving the reliability of LLM‑based multi‑agent deployments.