We do have one occasionally significant change to the data (besides players changing teams, like Julio Lugo). Something in the last week got out of whack on the pitcher's games pitched and games started, as a result of which a bunch of pitchers were listed as "Swing" pitchers even though we were projecting them to be full-time starters. I'm not sure how it got that way, but it has been corrected.
While I was fixing that, I also took the opportunity to drive some changes in how we handle starters and relievers. The ERA (and other stats) from PECOTA come with estimates for games started and relieved. Generally speaking, a pitcher working in relief will be able to strike out a few more hitters, give up a few less hits, and produce a better ERA and WHIP than he can while starting. Some rough estimates of that have been present in the mondo spreadsheet that puts together the depth charts, but we haven't carried those changes into the PFM output – until now.
The effects of the changes are probably best illustrated by looking at one pitcher, and there's no better example really than Joba Chamberlain.
There are internal safeguards in the main PECOTA program that inhibit really drastic changes; primarily, these are error checks, but in the case of a pitcher changing from starting to relieving (or vice versa) it prevents you from eliminating all the starts (or all the relief appearances). This manual fix works, but the longer goal will be to get the main program modified to handle this without breaking down. The "normal" line reflects Chamberlain's forecast up until the point he lost the fifth starter competition to Phil Hughes; the relief mode calculation is what you'd find in the PFM for the last week, until today (and still shows that way in the unadjusted section of the Pecota spreadsheet); the last line is what you'll find in today's PFM. The differences will be largest for those making the largest role changes – Chamberlain, Hughes, C.J. Wilson, Neftali Feliz, off the top of my head.
PS – Jorge Cantu, back to mere mortal status.