How the sims work

A projection is one number per player: the points they are expected to score. It is an average, and an average hides the two things a tournament player cares about, which are how far the outcome can swing and how often the good version happens.

So instead of scoring the slate once, we play it out thousands of times. Each run walks a game plate appearance by plate appearance, using a model of that specific batter against that specific pitcher, and records what every hitter ends up with. Because the run is a whole game rather than a list of independent players, hitters on the same team rise and fall together the way they do in a real game.

Thousands of runs give a distribution instead of a point estimate. Everything on the sims boards is read off that distribution.

A percentile is just a position in that distribution. When we say a ceiling of 135.3, we mean 1 simulated slate in 10 finished at or above 135.3. The median is the middle: half the runs landed above it, half below.

The range bar draws the 10th to 90th percentile, so it covers the middle 80% of outcomes and cuts off the tails on both ends. Every row on a board is drawn on one shared axis, which is what makes the bars comparable to each other rather than to themselves.

A wide bar is not a worse lineup than a narrow one. It is a lineup whose outcome depends more on how the night breaks. Which of those you want depends on the contest you are entering, and the sim does not have an opinion about that.

The board takes the lineups you have already saved for a slate and ranks them against each other by simulated ceiling. It grades what you built. It does not build anything.

The reason it exists is narrow and worth stating plainly. The optimizer maximizes a single projected total, and there are usually hundreds of lineups sitting within a point or two of the best one. On that measure they are effectively tied, and the optimizer has nothing left to separate them. The distribution is a second axis the projection does not carry, so it can break a tie the optimizer cannot see. That is the whole claim. The sim does not overrule the optimizer, and a lineup ranking lower here is not a bad lineup.

That is also why the projected total sits in the same row. When the two disagree you get a chip reading “Projection has it 3rd,” which is there to show you the disagreement rather than to resolve it.

If a row warns that some players had no sim draws, those players contributed nothing to the total. The number shown is short by whatever they would have scored, so treat it as incomplete rather than low.

A top stack is whichever team's hitters score the most as a group in a single run. Do that thousands of times and each team ends up with a probability: the share of simulated slates where they came out on top. Those probabilities sum to 100% across the slate, because exactly one team leads each run.

Where projected ownership is available, it sits in the next column. We show the two side by side and stop there. We do not publish a leverage ratio, because dividing one by the other buries the assumption that both numbers are equally trustworthy, and they are not.

Within-team correlation is how tightly a team's hitters move together in the sim. Higher means the stack tends to boom or bust as a unit.

This is an experimental alpha and the list below is the honest version of where it stops.

• Main slate only. Sims run for one main classic slate per day. Other slate types are not covered yet.
• No showdown. Showdown scores the captain at 1.5x, and the sim produces raw per-player scores. Summing them would hand you a confidently wrong number instead of no number, so we return no number.
• Missing players. A player with no draws in that slate's set contributes zero. The board says so on the row rather than hiding it in the total.
• Correlation is modeled conservatively. The within-team relationship the sim produces is the right shape but is deliberately held weaker than what we measure in real games. It is an active area of work.
• It knows nothing about your contest. Field size, payout curve, and duplication are not inputs. Nothing here is a recommendation about what to enter.

The simulation reruns several times through the day. Each board shows the timestamp of the run behind the numbers you are looking at, along with whether that run was working from confirmed lineups or projected ones.

Numbers change between runs, and that is the system working. A morning run cannot see lineups that are not posted yet, so a later run is built on more confirmed information. How much a number typically moves between an early run and a late one is something we are still measuring, so we are not going to put a figure on it here.