{"title":"Model Collapse: Preventing Recursive Training Rot","region":"Global","category":"Data Integrity","description":"Ensuring agents don't ingest synthetic hallucinations as ground truth.","lastUpdated":"2026-02-23","steps":["Tag all agent-generated data with a 'Synthetic' metadata flag.","Implement a 'Human-Verified' anchor for training data subsets.","Audit vector store for high-density 'Echo' clusters.","Use 'Statistical Diversification' to filter out low-entropy text.","Rotate training sources to include high-authority external docs."],"url":"https://checklist.day/model-collapse-preventing-recursive-training-rot"}