Using the Flat Earth Transformation (Müller, 1977) results in a coordinate singularity at the planet's center, which is transformed to infinite depth. This singularity makes it difficult to compute ray paths and travel times for phases that come close to the center. With the "flat" sampling mode, this problem is effectively removed. It is thus possible to compute ray paths and travel times even for inner core phase for epicentral distances close to 180°.
The take off angle optimization used by TTBOX for shooting rays finds very good solutions in almost any case and acceptable solutions in the few exceptions.
Comparison with the travel time difference between the Jeffreys Bullen travel times (Jeffreys & Bullen, 1940) shows that TTBOX is well suited to distinguish the JB velocity structure from the IASP91 velocity structure by travel time. This important for a use with inversion programs.
Travel times computed using TTBOX are - in the limit for ideal depth sampling - within ±0.05s or less of the printed IASP91 travel time tables. Deviations increase slightly with increasing epicentral distance. What is astonishing is that the completely independent TauP Toolbox of Crotwell et al. (1999) produces not only a very similar deviation from the printed tables, but also the same distance dependence. On the other hand, the program onset, which uses essentially the same computational methods as were used in the production of the reference data by Kennett (1991) gives travel time differences that are very similar to the printed reference, but do not correlate with those of TTBOX and TauP Toolbox. It is obvious that at least two of the data sets compared in the previous sections contain a systematic offset, but it is important to notice that carrying out more tests alone does not allow to recognize the biased data set, since truth is not subject to majority decision.
From all this, I conclude that Type III, IV and V errors are not the main cause for the observed 0.05s differences between TTBOX and the IASP91 reference, and that TTBOX is in fact a valid tool for travel time computation.
A drawback of computing in MatLab is that MatLab is relatively slow. However, this does not matter much in many cases (and at today's CPU speed) and is balanced by the integration with other MatLab routines and the built-in graphics routines.
(But I am not selling used cars. You are invited to draw your own conclusions.)