"Ensemble" is my jargon for a collection of alternative MSAs of the same sequences. One MSA in an ensemble is called a replicate.
With stratified ensembles and diversified ensembles, the same sequences appear in all alignments.
With stratified and diversified ensembles generated by align or super5, all replicates have equal accuracy on benchmark tests, noting that the replicate generated with default parameters is no better than any other on average. Therefore,
(1) there is no preferred replicate, including a replicate generated with default parameters, and
(2) differences between replicates must be due to alignment errors, because two different alignments cannot both be correct.
MSAs in a resampled ensemble are derived from the same original sequences by sampling columns. Here, the sequence labels are the same as the originals, but letters are shuffled due to sampling.
