Multi-agent systems reveal hidden failures shaping task outcomes
MASIF: Implicit Failures for Process Observability in Multi-Agent Systems
Artificial Intelligence
Summary
Multi-agent systems are groups of computer programs working together to complete tasks. Sometimes, these systems make small mistakes that don’t stop the task from finishing but still affect how the work is done. The authors call these hidden mistakes implicit failures and studied many examples to see how often they happen and how they impact the process. They created ways to understand and measure these mistakes, showing they provide useful information that final success or failure alone doesn't reveal. They also developed a technique that uses these mistakes to improve decision-making in multi-agent tasks.
What this means in practice
- •For robotics engineers: Identify and analyze hidden errors within teams of robots to improve task coordination and fault tolerance beyond simple success metrics.
- •For software development teams: Use implicit failure data to select higher-quality execution traces when training or evaluating multi-agent software, enhancing process insights.
Authors
Yanwei Huang, Arpit Narechania, Atanu Sinha
Abstract
Multi-agent systems (MAS) can complete a task despite local execution errors that are corrected, bypassed, or carried forward without individually determining its outcome. We call these implicit failures (IFs) and study them as signals for process observability beyond final task outcome. Across 1,828 traces from nine MAS-benchmark configurations, IFs occur in 54.3% of successful and 75.2% of failed traces. We develop taxonomies of error types and subsequent absorption mechanisms, supported by human and cross-model annotation audits. Counterfactual replay further probes how repairing an IF changes the final outcome. To examine the practical value of IFs, we evaluate IF-based best-of-N trace selection against existing process reward models, and introduce a novel RL algorithm, IF-GRPO, that uses IF costs to refine preferences among successful trajectories in GRPO. Experiments show that IF-GRPO exhibits some gains in accuracy and process metrics. Importantly, it exposes trade-offs between task performance, IF count and cost, and token-normalized IF density. Overall, our results show that IFs provide complementary process observability into how MAS reach their outcomes, including error recovery, propagation, and IF-based process differences that final task success alone does not expose.