Refusal to run a gauge R&R while insisting errors are due to the operator is a sure sign of a problem. If it hasn't been done, that's one thing but refusing to allow it is quite another.
Even if the rate was actually really good -- something like 99.9% accuracy -- if you apply that to a database with a ten million people in it, you get ten thousand false positives, which is crap.
If you pick one of the matches at random, they want to be able to claim that it's a 99.9% chance that it's them, but it's really a 99.99% chance that it isn't. Or worse, since it's possible the true perpetrator wasn't in the database.
Even for things like DNA that theoretically can do better than 99.9%, you can't get much better than that in practice because before you can, the error rate starts to be dominated by human error rather than coincident matches.
It's probably 4-5 years since I read about this, it'd be nice if things have changed.