Sort the Dataset Examples: Keep, Fix, or Drop
Sort candidate examples by whether they are ready to train, need rewriting, or should be excluded.
Sort each candidate example by what should happen before training. Keep as-is Fix before using Drop from dataset A reviewed escalation summary that matches today’s style guide and current policy. A helpful response with the right tone but an outdated benefits amount from last quarter. A summary that invents authority by saying “I have already approved the refund for you.” A borderline fraud case that was reviewed yesterday and shows the exact escalation threshold. A strong customer reply whose stable empathy is useful, but whose cited clause has since been rewritten. A chat where the human agent breaks policy but…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in