Scoring predictions for a large set of inputs. Each input is processed independently; there are no dependencies between predictions. 2. **Hyperparameter search** — Evaluating different hyperparameter configurations. Each trial trains an independent model; no information flows between trials (in grid