This is honestly a better idea than any intuition I can give you.

Also anecdotally, I have encountered situations where one batch of replicates was either much lower quality or somehow different than the rest and we ended up dropping those by first finding them using an empirical setup similar to the one described above.