"Up to 80% of a dog's alerts are wrong" can't be evaluated without more information about base rate. If this is occurring in a context where the dogs are only

"Up to 80% of a dog's alerts are wrong" can't be evaluated without more information about base rate. If this is occurring in a context where the dogs are only being brought out when there's already prior cause to expect narcotics with >10% likelihood, that doesn't look good for the police. But if that number is more like 1%, even an 80% false positive rate corresponds to doing more than half of the remaining classification work (it's appropriate to look at this on a logistic scale). (Also, "up to" should be unpacked; a 95% confidence interval is a lot more informative than a rough upper bound.)