With only dynamic errors, we have:
a = 93.3% of played errors have been detected.
b = 93.3% of played errors have been detected and well classified.
c = 3.3% of good notes have been recognised as errors.
Results are quite good too. Problems come when we have combinated errors: general dynamic errors / accentuation problems. An accentuated note in a too loud played part may be detected as an error while it was right.