An insurer scores a large file of existing policy records overnight. Results are needed the next morning, and no user waits for an immediate response. Which inference approach is the best conceptual fit?

1. Batch inference.
2. Human labeling of every input record as the only valid form of model inference.
3. Model pre-training instead of inference.
4. A continuously running real-time endpoint solely to keep each response below an interactive latency target.