convergence-debugger

What problem does it solve?

AI deliberations can sometimes fail to converge, leading to wasted time and resources. This skill provides an expert guide to diagnose and fix issues within the AI Counsel's convergence detection system, ensuring efficient and accurate consensus building.

Core Features & Use Cases

• Transcript Analysis: Learn to interpret deliberation transcripts for convergence status, similarity scores, and voting outcomes.
• Configuration Validation: Identify common misconfigurations in convergence thresholds, early stopping, and round constraints.
• Backend Diagnostics: Troubleshoot issues with semantic similarity backends (Jaccard, TF-IDF, SentenceTransformer) and ensure optimal selection.
• Voting vs. Semantic Conflict Resolution: Understand how voting outcomes override semantic similarity and debug parsing issues.
• Use Case: Your AI council is running for the maximum number of rounds even when models seem to agree. This skill helps you pinpoint whether it's a misconfigured early stopping threshold, a semantic similarity backend issue, or a problem with how models are casting their votes.

Quick Start

Guide me through debugging why my AI deliberation is not converging, even though the models' responses look very similar across rounds.