Typically, accounting for player injuries can lead to an overcorrection. This is because (1) most players are role players; therefore, most injuries happen to players that have less of an impact on the game, and (2) the replacement value of the substitute player is often quite close to the value of the injured player.

It is simply the case that even among bench players, rostered players at the D1 or professional level are high level—and thus are able to fill in without as much of a drop-off as fans might assume. However, injuries to superstars are different, and I do adjust to injured players over a certain usage. I don't believe anyone else accounts for this, although I could be wrong.

I don't use any pre-season or introduce any non-game performance weighting. Everything in the model is a possession that took place during the current season. However, I have a methodology that identifies possessions that are most representative of a team's performance, which establishes the basis for the simulation.

In Men's Division 1 college basketball, going back 3 seasons, I was 61% Against the Spread (ATS). Straight up, the model was higher, but please note that Straight Up records in College Basketball are higher in the NBA due to the number of "Buy Games" that teams play each season that match superior opponents against inferior ones. 

Not every opponent is equal. The expected adjusted performance, both as a game score and per possession, tells you how a team is expected to perform analytically against a certain opponent. FFor instance, when facing weaker opponents, the expected score tends to be lower due to the relative ease of scoring against these defenses compared to an average one.