Where should the batch processor sit relative to memory_limiter and any sampling or filtering processors?

Why this is the answer

The recommended order is memory_limiter first (so it can refuse data early), then processors that drop or sample, then batch last so that the batches sent to exporters contain only surviving data of the right size. Batching first wastes work on data that is later dropped, placing it before the limiter defeats early backpressure, and processors run strictly in the listed order.