[Chris is me]

First, we figure out what an alliance needs to win the match. We try to be as specific as you need, but no more. Being too specific constrains you too much (committing to "two scorers and a defender" in Lunacy was the downfall of many teams, for example), but being too general doesn't form the alliance you want. In Breakaway, we wanted an alliance that could shoot 6 in auto, had an agile striker, high traction feeder with excellent ball control, and a defender that could at the very least clear balls in auto, with a hanger or two thrown in there if possible.

Then you figure out what stats will best demonstrate who is the best at the roles you've dictated for your alliance. In Lunacy, +/- was a good indicator for almost any alliance. Breakaway, it got more complicated. Auto kicks, front scores, mid feeds, pushing, hanging, etc.

Then you assemble your list or lists based on your data and who's best at what you want. You figure out what you want in a first pick and what you want in a second pick, then you make them. This year there were two general approaches to the two picks, mainly the "front last" second pick and the "back last" second pick (even if they started in the middle to allow the primary scorer access to 3 balls in auto). Alliances like 254 / 233 / 3357 picked their frontbot last, while alliances like 2056 / 1625 / 3138 picked their backbot last. This depended on what was available in the field, what choices people had, etc. I know both times for my team we ended up picking the backbot last, but for different reasons. 1124 was inexplicably still around for the 12th pick of the draft (granted, this was before they changed their code and became the frontbot they played as at CT and champs), and they had an auto that could theoretically clear 2 balls out of the back. At CT, the much deeper field left 173, 3125, and 102 available for that 12th pick; 173 had a lot more consistency on Saturday morning than 3125, so the choice was made to pick them.