Sort examples before they poison the split
Separate training, validation, holdout, and exclusion examples for a fine-tuning dataset.
Sort each candidate fine-tuning row into the right split decision. Train Validation Frozen holdout Exclude sort-buckets score-chips Current-policy example that demonstrates the target answer format and has reviewer approval Representative ordinary case used for checkpoint comparison during iteration Rare high-risk escalation case reserved for final launch decision Duplicate of a training row with only the customer name changed Example from the old refund policy that product retired last quarter Clean hard-negative example that clarifies a confused label boundary Small stable slice used to compare checkpoint A against checkpoint B Board-visible safety case nobody may tune against repeatedly
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in