A side-by-side experiment comparison — one column per run, hyperparameter and metric rows, differences lit up against a baseline you can move, a best value per metric decided by contract-declared polarity, and a frozen label column.