To solve the problem with the TDF, I reverted the code to go out in angle only to cos(theta) = .58. I also required that P_TDF be >= 0. For r > 10k, I set P_TDF = 0 (meaning the only contribution will be from the background noise).
Version 3.03 still has problems with accidentals throwing off the reconstruction.
The showers reconstructing high are:
Shower 10