It seems to me that there are two ways one can check on the efficacy
of the typical autosomal test that is administered by FTDNA or
23andme (yes, I realize they have somewhat different tests).
1. simulate genetic data (essentially, create genetic data for a
simulated family over, say 5 generations or what have you) and run
the test on the fake data to see if you can retrieve the known
2. retrieve genealogical data >from a known and well documented tree
and run the tests on folks >from every branch of the tree and see if
you retrieve the known relationships.
The advantage of (1) is that it is much cheaper and you can change
the data set in any number of ways (can make the data more "noisy",
can make fake ancestors that are partially related to each other or
totally unrelated, etc.). The disadvantage is that you could probably
never simulate reality with 100% accuracy.
The advantage of (2) is that it is more similar to what the average
user of the product wants to be doing. The disadvantage is that
probably the company involved will have to subsidize some testing (I
think its worth it though).