Current and projected
Where every team is now, and where the model expects them to finish. Games already played are treated as fact; only what is left gets simulated.
They answer different questions and disagree on purpose
Preseason is the model's talent estimate stretched over a neutral 162 games. It is the number the system is graded on, and it never sees a single result from this season.
Projected finish keeps every game already played and simulates only the remainder. By August that is roughly two thirds fact and one third forecast, which is why its interval is so much tighter than a preseason one.
The gap between them is the interesting column. A team far above its preseason line is either better than the model thought or has been lucky, and those two look identical in the standings. The run-differential and component columns on the projections page are where that gets separated.
Calibrated, not decorative
The 90% interval is fit so that it actually contains the answer about 90% of the time. Held out one season at a time, it covered 89.6% of preseason outcomes and 86% to 89% of in-season ones. A team sitting at 60 wins in early August still carries a range of roughly 12 games, because 45 games of baseball is a lot of variance even when the talent is known exactly.
Playoff odds come from the same simulation: 3,000 seasons run through the real remaining schedule, with division winners and wild cards resolved the way the actual format does it.