The 1MB field generated here plus the padding results in a document size slightly bigger than 1MB (1048600 bytes instead of 1047576). This can lead to errors in randomly generated test cases that are playing on autosplitter thresholds corner cases (e.g. maxChunkSize` 56MB and surplus 18MB).
Changing the 1024 * 1024 with 1024 * 1023 seems to generate documents of exactly 1MB, allowing the unit tests to be always be correct.