OrchestraBench injects seeded failures into multi-agent workflows and finds recovery splits into three tiers: tool faults recover, ambiguous delegation partially recovers, and latent semantic failures never recover, while cascade radius grows with pipeline depth.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
OrchestraBench: Evaluating Multi-Agent Orchestration Failure Modes, Recovery, and Decomposition Quality
OrchestraBench injects seeded failures into multi-agent workflows and finds recovery splits into three tiers: tool faults recover, ambiguous delegation partially recovers, and latent semantic failures never recover, while cascade radius grows with pipeline depth.