At last, with the completion of program 6a, we have a tool which will, given some historical data, give us a prediction for size with a minimum of fuss and hassle. This is tremendously exciting, as the prediction it came up with given all my data and my initial estimate for program 6a was actually fairly close (283 LOC predicted; 299 actual). While my first-guess estimates are always wildly off, there's a chance they're wildly off in a predictable manner, and that's a good step on the way to estimation.

I continue to be pleased with Eiffel. While I haven't analyzed the numbers yet (waiting until after the PSP lessons proper), I can say that its standard library has surprised me less often than the C++ standard library, despite one workaround problem (the "raise-to-the-nth-power" function in Eiffel only allows integral powers, which confused me a bit on this program until I realized I could fake the algorithm by using a square root in combination with an integral power). The automatic documentation has been essential.

My numbers are stabilizing as I add more projects to the database; the ratios if total time to phase time have been remarkably stable, which now prompts me to ask more questions such as: am I spending the right amount of time in each phase to best use my resources? Should I be looking to eliminate defects earlier? It would be interesting to try and forcibly change the numbers and see what the results would be.