Great article by BP's Kevin Goldstein on ESPN.com (subscription required), talking about players who are so unique they defy easy historical comparison.

Here's an excerpt:

The Tiny Pitcher: [Tim Lincecum] listed at 5-11 and 160 pounds... is one of the most unique pitchers in recent memory, with unteachable mechanics, arm speed that borders on the unreal... that allowed him to rack up downright abusive pitch counts at the University of Washington without injury or even any soreness. To this day, Lincecum says he has never iced his arm.

The beauty of this article is that by focusing on the extremes it shows the limitations of all projections systems including BP's own PECOTA because they can only account for what usually happens. With unusual players, it breaks down, and I'd argue that even with more usual players, it will never be able to account for unusual deviations. To its credit, the PECOTA system acknowledges this by means of a Similarity Index. If a player has a low number there, the system admits its projections based on comparable players are less likely to hold true.

So apart from being a fascinating article on the listed players in its own right, it also illustrates why you have to augment any system's projections with a case by case look at the players. And players aren't merely freaks or not freaks, normal or abnormal, but on a sliding scale ranging from total freaks to ones who have many similar comps.

I'd like to see some data, too on how a player's Similarity Index changes over time - if it changed a lot, then it would seem to undercut its value.