I don’t remember all of DP’s methodology or if he has changed it much since its inception, but I vaguely recall there being quite a few differences. Any time you have two different engines processing the same data, you are bound to have some large differences, especially in small (say, one year) samples.

Sky, above, that does not mean that Hawpe is screwing up 1/10 of his opps, or some such thing. A large number, plus or minus, is extremely likely to be a fluke. If you see someone with -37, it likely means that he is a bad fielder (maybe -18 if that -37 is in 150 games or so) and that the other -19 runs are “flukey.” “Flukey” could be hard balls to catch that “appeared” easy according to the data, 5 or 6 balls (or 10 or whatever) that 50% of all fielders would have gotten to, but Hawpe happened to have just missed, balls that an adjacent fielder would have or should have gotten to but didn’t (in UZR when any player catches a ball, no other players get docked, and when no one catches a ball every player whose position EVER caught a similar ball gets docked some amount). Etc.

When you see a number for a fielder, NEVER think in terms of that number and “how it could be.” Always think in terms of the proper regressed number and how THAT could be. That goes for all sample metrics – offensive, defensive, etc.