One quirky metric I will keep my eye on: divide the history into N equal-length time periods--what is the largest value of N for which the portfolio beat the benchmark in N-1 of the sub periods? Or all of the N sub periods. Ideally this exercise would require a history of daily balances which I don't really have a simple way to maintain, but I'll have monthly.

Oh wow! :-)

In another (not stock-related) area, I have a spreadsheet called nineheads. The idea being that instead of doing elaborate statistical calculations (I've never been much good at stats) I simply divide a data series into 9 pieces and calculate the up/down value for each piece. If I get 8 or 9 "up" I assume the series is reliably "up", and 0 or 1 (i.e. 8 or 9 "down") means reliably "down". The 6 levels between those extremes are various levels of "not sure".
I haven't really used it in anger yet, or offered it for criticism (don't think it would be well received) but I like the simplicity of it. I derived the "9" value (with the help of some internet stats tutorials and trial-and-error spreadsheets). Seems like you're doing a similar but more generic thing with a variable N, so your "score" is the size of N, and my "score" is the number out of 9.

That's made my day, no, my year!!! :D

SA