Reducing manual verification?
First of all, with 1 real bug found out of 104 detected anomalies, there is no way you can use this at runtime. Not even if it gets a lot better.
Secondly, in order to decide whether it can reduce manual verification, we'd need to know the rate of false negatives. If it catches some bugs but misses others, then you can't stop the manual verification you're already doing. Which means the tool would add manual verification, with the amount added depending on the rate of false positives (which looks horrible, although it may improve). This may not be a bad thing, if it does catch the odd bug which would otherwise have slipped.