You can't be one-in-a-trillion confident about any particular DNA result, even if that's what the nominal probability from the DNA analysis itself seems to say, because it is objectively observable that there are plenty of other sources of errors of all sorts. You can only get down to that "noise floor" of probabilities. This can still be a very useful result, but it's important to not let it be any more important that it deserves.
Even if you have a magic machine that you can point at someone and it goes "boop" if they are guilty, you still must consider the probability that someone faked the "boop", or has swapped in a different shell that looks the same but makes the same "boop" when someone remotely triggers it or that the speaker was broken so even though the magic machine tried to "boop" nobody could hear it. All of these are quite realistic and of a much higher probability than the magic machine existing in the first place.
(This is also arguably the root problem with the still-popular "The computer said it so it must be true". Even if you assume the computer really is 100% accurate, which itself is a transparently false proposition when examined in daylight, there's still plenty of other reasons why one should not give the computer too much confidence.)
