convergence-debugger

Debug convergence detection issues by checking transcripts, config, backends, and voting status conflicts.

• Examine transcript for convergence info and voting outcomes
• Verify config settings for thresholds and min_rounds_before_check
• Ensure correct backend (SentenceTransformer preferred)
• Test semantic similarity manually with sample texts
• Resolve voting vs semantic status conflicts (voting overrides semantics)

• Convergence status depends on semantic similarity or voting
• Voting overrides semantic status when present
• Semantic similarity thresholds (0.85 default, 0.40 divergence threshold)
• Min_rounds_before_check must be ≤ (total_rounds - 1)
• Jaccard backend has low similarity scores (only word overlap)
• SentenceTransformer backend provides best semantic similarity
• Early stopping fails if no votes or thresholds are met

1. 1.Run `ls -lt transcripts/*.md | head -5` to view recent transcripts
2. 2.Check `config.yaml` for convergence and early stopping settings
3. 3.Test backends with `python -c` commands to verify availability
4. 4.Use `grep -A 5 'VOTE:'` to check for voting in transcripts
5. 5.Run `test_similarity.py` to validate similarity scores
6. 6.Adjust thresholds or reduce `min_rounds_before_check` if convergence not detected