The most recent revision to the code changes the first-guess proceedure to be considerably more robust and reliable.
The exit wound position is now guessed to be closest to a high-charge tube that is surrounded by neighboring tubes with high charge. In the brute-force search stage, this position is varied by moving it to different points on a hexagonal grid projected onto the surface of the sphere.
The track direction, in the brute-force search stage, no longer uses a simple grid in theta and phi. Now, theta and phi coordinate pairs are loaded from the TFTM 2 bank, which contains a list of points pre-calculated to be equally-spaced about the sphere. (That is, the dot product of any two neihboring direction vectors should be the same, unlike the case of a simple grid in theta-phi). This has the effect of profiding much more homogeneous coverage of parameter space, while decreasing the computation time. (Thanks to B. Nickel at Guelph for this computation, which uses points interpolated on the faces of an isocohedron, then projected onto a sphere.)
Finally, the best-guess time used for each trial is now computed using a two step procedure: find best (mean) time offset for all tubes, then cut outliers and re-average. This helps robustness considerably, since now events with many tubes hit by reflected light will no longer pull the muon time away from the peak.