Two different jobs get called sampling. One asks whether the batch may be released. The other asks why one corner behaves badly. They are not taken the same way.
For a release question you want a composite, a single mixed sample built from equal amounts taken all over the load, because you want the batch average. For a diagnosis you want the suspect zone kept separate and labelled, because the whole point is the difference. Mix a bad zone into the composite and it disappears into an average that looks acceptable while the crop fails.