I want to address your second issue. I work in the semiconductor industry and a wafer fab has a dozen or more major process steps that are used hundreds of times in the production process. There are hundreds of variables for each step, so there are literally hundreds of thousand variables to track. Machine learning enables us to optimize the process steps individually and collectively. In the old days, craftsmen used to do this on a micro level, e.g. if this pin needs to mate in that drilled hole, then take the pins with a larger measured diameter and match them with the holes with larger diameters to maximize performance and reliability.
MLB teams have much more data than ever before. Imagine taking Crawford's size, reach, flexibility, strength, previous injuries, parental health, ultrasound of tendons .... add in game situation, pitch count, type of pitches, pitcher fielding plays, ball stickiness, etc. I'm sure there are other datasets to add to this. Run the regression with an output of player health and, eventually, teams will be able to predict injuries with some accuracy. At a rudimentary level, they already do this with reliever days and pitches, and young starter innings year to year (Si's Tom Verducci used to highlight this one).
I'm sure the Sox have something already cooking, and I hope they team up with the world class medical practitioners and facilities in Boston.
