The analysis of data is one of the many items in the PSP which is incredibly painful if done by hand, but which lends itself to automation. I've only implemented a very little bit of automation here, using Makefiles to generate graph targets, etc.-- there is much more that could be done.

In particular, it seems that a serious practicioner of the PSP should have at their disposal a database of information into which he or she could "check in" programs; from this database, a set of analysis programs should be able to generate a full and complete report, graphs and all, at the drop of a hat. The evalpplog program helps a great deal, but even it is limited in its expandability. It might be an interesting exercise to convert logs and other data into roughly human-readable XML or SGML, and use that format to store project summary data, etc-- making some of the reports much easier to parse and extrapolate from.

As is often the case with tabular data, a visual graph of data is more impressive and more understandable; it is quite one thing to look at a scattering of numbers indicating estimation error; it is quite another to see the slope of the line as estimates grow closer and closer to target. This is interesting stuff. Unfortunately, the current state is not automated enough to be used regularly in practice. I can tell already that, unless I can generate several more tools, I'll primarily resort to the human-readable reports generated by evalpplog.