multi-round-deliberation

Multi-agent discussions reach consensus by applying an iterative Assess, Check Convergence, Modify, and Re-assess workflow. Agents critique and revise artifacts until all participants pass the convergence criteria or hit a maximum round cap.

Iterative writing cycles prevent superficial feedback by injecting a shared value framework into agent prompts and running full re-assessments after modifications. This ensures agents resolve issues rather than merely noting them before convergence checks.

Complex decision-making workflows handle iterative refinement through round-based assessment and modification steps. Agents evaluate the artifact, apply changes, and re-assess the full output until consensus is achieved or the maximum round limit is reached.

Iteration and convergence checks apply directly to story design and paper writing workflows. The protocol integrates with these phases to guarantee quality by enforcing round-based assessments and modifications until all agent concerns are fully resolved.

If agents do not reach consensus during a review workflow, the iteration stops automatically when the maximum round cap is reached. This stop condition prevents infinite loops while ensuring multiple rounds of critique and modification are attempted.

Full artifact re-assessment after modifications prevents regression in multi-agent workflows. By re-evaluating the entire output rather than just the changed sections, the system ensures that new modifications do not introduce unresolved issues into the converged result.