In principal you can use whichever you like, but we suggest you use the human species for the two simulated test sets and dm3 for the fly set.
Yes, we'll take your pairwise alignments and we'll perform the analyses on them. But to be fair to other groups and to check the consistency of the pairwise alignments, we're also going to compute the transitive closure for the pairwise alignments you submit in order to test the induced alignments, too.
For example if you give us two alignments, one with species A to B (notation: A-B) and one B-C we will induce A-C from the first two and combine all the alignments into a single file in order to test. If you gave us all three pairs we would create induced alignments for the same three pairs by the same process but by holding out one of the three at a time. I.e. you submit A-B, B-C, A-C we would induce (A-C)* by using A-B and B-C, then similarly produce (B-C)* and (A-B)*, and then include those alignments in the tests along with your A-B, B-C, A-C.
The end result will be two analyses: one for the submitted pairwise alignments and one for the total induced alignment.