Narrative Tracking
Nothing clustering yet — keep the feed growing.
Auto-grouped stories over the last 30 days using local embedding similarity (MiniLM, no LLM required). Clusters are labeled by their most characteristic keywords. When an LLM API key is configured, a v2 pass rewrites labels with semantic names (e.g. "THC beverage regulatory crackdown" instead of "thc · drinks · colorado").
🕸
No narratives yet
Run
python -m worker.refresh_narratives after you have at least a week of signals with final_score ≥ 3.