Evidence and limits
Finding: Multiagent debate improved reported reasoning and factuality on selected benchmarks without model training.
Impact: Watch debate as a task-specific candidate. Reject preprint benchmark gains as a general multi-agent advantage or independent verification.
Boundary: Agents share model-family weaknesses and the work is not an owning-system field test. Next: when does debate outperform equal-cost sampling or one strong verifier?
Flags: preprint; same-model instances; judge dependence
Record State: Catalog review recorded; not a performance endorsement
What was reviewed: Structure and relevance screened; not a full evaluation