The linear regression analysis program was a much better problem than the LOC counters-- not just in its simplicity, but also in its objective interpretation and a testable outcome. If you're building a baseline of data, particularly to compare your progress against others, this is a good example of the sort of program to use (1A was also good). From the descriptions of the rest of the problems, I have high hopes for them.

Now that we have a linear regression tool, I'm very tempted to use it for other comparisons: to see if there is a correlation between Eiffel development time and C++ development time, or program size, etc. I'll hold off on that until I have a full set of data and more tools, but it's an intriguing idea-- realizing that development time will be extremely biased toward Eiffel, as the Eiffel programs are developed after the C++ code.

However, I have issues with the instruction given in [Humphrey95]. The script for this PSP, while detailed, omits some important instructions (how to use the t-distribution? What exactly does "new reused" code mean?) which are findable in the text but still needlessly cryptic. I would rather have seen the LOC counting problems omitted as programming problems, and the use of "standard" LOC counters encouraged, keeping the problems objective and reducing the frustration (and data corruption) caused by programs 2A and 3A.