Got it!
Ok, I think everything is fine. Dada2 uses the first x samples (sorted) to build up the n-reads values and train the error models.
Because errors are similar on each run, this is the default and should work well. You are good to go 
Colin