probably minor – amount of foul territory affects park factors, a place like Oakland gets a lot more foul pop-outs than RBiA, for example. Of course, that could be a general problem wtih FIP, as foul outs are somewhat fielding independent – obviously they are fielding dependent, but the alternative is a continued at bat, not a man on base.

I think Adam’s system is probably pretty good. Obviously using ERA to monitor pitching has an error associated with it due to fielding. And FIP has estimation errors associated with it as well and lack of park factors. Averaging a park factor corrected ERA (ERA+) and FIP probably gives you a decent estimate.

Ks and walks can vary dramatically from park to park.

we would find ourselves in contention.

are HRs, Walks, and BBs and they have a constant to make them scale like ERA. In a blowout you would assume the pitcher being blownout would give up lots of HRs and BBs so it is about as susceptible as ERA is to blowouts.

you’re supposed to assume baseball statistics are parametric and smoothly distributed. Stop rocking the boat.

In all seriousness, I think you touch on a key point here. Because runs scored aren’t independent of one another, the marginal probability of scoring (or giving up) an additional run decreases as the number of runs goes up. There are several reasons for this – runs gets scored in bunches (just guessing, but I’d bet 3 hits in an inning probably yields something like 1.3 runs, 4 hits probably 2.2 runs, 5 hits 3 runs, etc). Once down (or up) by a lot, teams use worse pitchers so as not to waste their good arms, so there is another bias against high score teams.

I think a similar bias will exist in Pythagorean records as well. I don’t know how significant these biases are, but you’re probably right that they exist.