'Good evidence' isn't the same as definitive evidence. If that were the criterion, no evidence would be admissible, since there's an unquantifiable degree of doubt about all testimony. Lie detector evidence has the advantage that we know what that degree of doubt is.
Taking a low figure, if a lie detector test is accurate 85% of the time, that probability can be inserted in the 'probability string' represented by the probabilities of all the separate pieces of evidence put before the court, and a more accurate overall probability of guilt or innocence obtained. No individual piece of evidence has to be definitive.
'Genetic fingerprinting' mirrors the approach. The very high probabilities of accurate identification produced by the technique, are calculated from a string of much lower probabilities, represented by 'matches' at up to 6 to 9 individual sites on the genome. An individual 'match' might have a 25% chance of being purely coincidental, but, taken together, matches on a large number of sites give identification probabilities of more than a million to one. Such a test is regarded as very good evidence in court.
The problem, of course, is juries. Choosing 12 people at random off the street, to perform a complex task for which they have neither knowledge nor experience, has its drawbacks.