MAVEN improves LLM reasoning on QA benchmarks by using a blackboard-style multi-agent system with explicit Skeptic-Researcher-Judge deliberation to produce verifiable trajectories rather than implicit states.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
MAVEN: Multi-Agent Verification-Elaboration Network with In-Step Epistemic Auditing
MAVEN improves LLM reasoning on QA benchmarks by using a blackboard-style multi-agent system with explicit Skeptic-Researcher-Judge deliberation to produce verifiable trajectories rather than implicit states.