Reproducibility with versioning

I created a custom dataset with tfds build with a single train split. I want to share this dataset with my colleagues and ensure reproducibility such that if we all use a 80:10:10 split on the dataset, the split sets yield the same data on all our workstations. The versioning section in the documentation states that as long as the major version remains unchanged between runs, "the data as read by the client is the same". Does that mean that my colleagues and I have the same contents in the splits if everyone uses version 1.0.0?

Start with the versioning section linked in the issue and the behavior of a custom dataset created with tfds build. Determine whether version 1.0.0 guarantees identical 80:10:10 split contents across workstations, then clarify the documentation so the reproducibility conditions and expected result are explicit.