Review

Artifact review and validation workflows classify outputs into accepted, revised, or rejected categories through self-review, apply quality-diversity filtering to ideas, and trigger multiagent debates for low-confidence revisions to refine results.

The best way to filter ideas is applying Quality-Diversity filtering using a MAP-Elites-like approach, which activates when more than three accepted items exist, ensuring diverse and high-quality artifact selection before escalation.

Debate escalation should be used for low-confidence revisions identified during the self-review phase, triggering a full multiagent debate to critically evaluate, refine, or reject the artifact before final aggregate reporting.

Yes, this multi-agent review process applies to diverse artifacts including ideas, gaps, experiments, and results, running a self-review SOP to classify each artifact and produce a consolidated ReviewResult for downstream decision-making.

Aggregate reporting works by consolidating the outcomes of self-review, quality-diversity filtering, and debate escalations into a final ReviewResult, providing a unified view for downstream decision-making in multi-agent strategy contexts.