[thinks] Oh, I see. You're right, if Dominions uses Jazzepi's discontinue-on-fail algorithm P(3 sites | 2 sites) = site freq, and maybe you shouldn't skip them. However, I don't believe that it does discontinue on fail because (w/ site freq 45) I don't see anything close to 55% of provinces having zero sites. The distribution of sites I actually see looks more like a binomial, which makes skipping 2-site provinces reasonable after all.
Binomial probability calculator
P(1+ sites) = 91%
P(2+ sites) = 61%
P(3+ sites) = 24%
P(at least 3 sites | at least 2 sites) = 24/61 = 40%
Assuming that all sites are equally likely in a given terrain, that gives a 91/8 = 11% chance of finding a site in a "fresh" province with a site-searching spell, and a 40/8 = 5% chance of finding a site in a province that already has 2 sites.
-Max
Edit: no, that's wrong. There's more than an 91/8 = 11% chance of finding a site with a spell, since there could be more than one site there. But I'm too lazy to do the math at the moment, and I've made my point.