Maximum Adjacency Search: should unit tests be relaxed to allow multiple correct orderings in the result?

When @daankolthof added more tests to MAS, it occurred to me that different but equally correct implementations of the algorithm might return different orderings when tie-breaking edges with equal weights. I haven't confirmed that this is a real problem, it's just a suspicion.

Assuming I'm right, the principle here is that unit tests should be written to test the implementation against the _specification_, not against itself.