The value of sampleSize in the _id and metadata fields in the persistent sample doc schema is the size of the sample that was requested when analyze was run, not necessarily the actual size of the sample (which can be inferred from the length of the docs array). The actual size may be smaller when the collection size is smaller than the requested sample size, or with chunk sampling.

This has become a recurring source of confusion, so we should more accurately name the field requestedSampleSize.