...

In any case, we have a consensus from the users' side that piecewise ephemerides is the way to go. So it seems that a change to gtephcomp's ephemeris selection algorithm may be what we need.  

Even if we agree that in general and in most cases we'll want piecewise timing solutions, allow me to insist that the pain-to-gain of adding higher-order derivatives to both the D4 and to gtpphase etc is small. It's just some columns, and some higher-order terms in the Taylor series. We would then have the flexibility of using ST's with ephemerides found in the literature. The alternatives are a) to track down the author and convince him/her to re-generate 3-parameter timing solutions for the epochs that suit us, or b) extract the dates from the FT1 file and use TEMPO instead of ST's.