A Projection Is an Average, Not a Line

An FDI projection is the model’s expected value for a player’s stat in one game: roughly, the average outcome across all the ways the game could plausibly go, given the information available two hours before kickoff. It is frozen at that time and never revised.

An average does not have to be a possible outcome. A passing-yards projection of 238.4 is not a prediction that the quarterback throws for exactly 238.4 yards; it is the centre of a wide range. For a small count the gap between the average and any single outcome is even starker — an expected value of 1.4 describes a mix of zeros, ones, twos and more, never 1.4 itself.

The FDI projection, in the model vs market guide

A sportsbook line is set so the book is comfortable offering prices on both sides. It usually sits near the middle of the outcomes — closer to the value the result finishes above about half the time than to the average. Lines on player stats are typically set at a half point, such as 234.5, so that a result cannot land exactly on them.

The line also comes with a price on each side, and the two prices need not be equal. A line of 1.5 offered at +120 on one side and −150 on the other is a different market from 1.5 at −110 on both. The line alone never tells the full story; its prices and the time it was observed are part of it.

Many player stats are lopsided. A player is far more likely to finish near their usual output than at three times it, but the occasional big game pulls the average up. When outcomes are skewed like that, the average sits above the middle of the outcomes even when both are estimated perfectly.

As an illustration, a stat whose outcomes are usually 3 or 4 but occasionally 9 might have an average of 4.3 while half of all outcomes finish at 4 or below. A projection of 4.3 beside a line of 3.5 or 4.5 is therefore not automatically a disagreement: part of the distance can come from comparing an average with a middle.

FDI shows a difference — projection minus the same player’s current line — because it is useful orientation: it says which side of the line the model’s average falls on and by how much. It is plain subtraction, shown only when both numbers exist for the same player and the line is current.

• It is not a probability: a +0.3 difference does not mean anything is 30% more likely.
• It is not a signal: model error, skew, the book’s margin and late information can all explain it.
• It is not comparable across stats: +3.9 yards and +0.3 of a small count are on different scales.
• It disappears when the line is not current, rather than being computed from an old market.

See projection, line and difference on the sample board

Sometimes the distance between a projection and a line is larger than skew or rounding can explain. That is worth understanding, but it is a prompt to ask questions, not an answer. The useful questions are mostly about information and timing rather than about the model.

• When was the line observed, and has news arrived since — a change in role, a late scratch, weather?
• Is the line priced evenly, or is one side noticeably more expensive, which says the market leans that way?
• Do the books agree, or is this one quote an outlier while most books sit elsewhere?
• What is the model’s typical miss for this stat, measured on games recorded before they were played?
• Was the projection frozen before information the market has already absorbed?

Most large gaps turn out to have an ordinary explanation on that list. The ones that do not are still only one model’s view against one market’s price — context for a reader’s own judgment, not a conclusion FDI draws for them.

Read the projection as what the model expects, the line as where the market has set its balance point, the prices as what that balance costs on each side, and the capture time as how current all of it is. When a model also estimates the chance of finishing above the exact number the line is set at, FDI shows that probability only in the player’s dossier and only at that exact match — as context, never as an instruction.

And read every projection with the model’s recorded history in mind. A projection is only as informative as the model has proved to be on games recorded before they were played.

How FDI records projections before kickoff