Every projection graded against the box score the next morning — misses included.
Our hit projections are averages over a distribution, not promises. A batter projected at 0.83 hits won't get exactly 0.83 hits — some nights it's 0, some nights it's 2.
So we convert every projection into the probability it implies. A 0.83-hit projection over ~4 at-bats implies roughly a 60% chance of at least one hit. We bucket every batter-game by that implied probability, then check how often a hit actually happened in each bucket.
A calibrated model is one where the bars match: when we say 65%, it happens about 65% of the time. When the bars don't match, we say so — that's what the model notes section at the top of this page is for, and it's how the projections get better.
Each bucket aggregates hundreds to thousands of batter-games, so this is the most statistically honest view of batter projection quality anywhere on the site.