Projections: An Informed Starting Point, Not the Answer

What a Projection Is Actually Claiming

A projected point total is not a forecast of what a player will score on Sunday. It is the middle of a range of things he might plausibly do, compressed into one number so it can be sorted. Underneath that number sits a chain of smaller estimates: how many plays his offense will run, what share of them he will be on the field for, what he tends to do with the touches he gets, and whether he suits up at all. Each link in the chain carries its own uncertainty, and the figure printed on the player card is what survives after they are multiplied together and the uncertainty is thrown away.

That compression is simultaneously why projections are useful and why they mislead. Useful, because a lineup decision requires you to put two players in an order, and only a single number does that cleanly. Misleading, because two players carrying the identical figure can describe completely different bets, and nothing on the card distinguishes them. The projection answers "which of these is bigger." Managers routinely read it as though it answered "what will happen," which is a question it was never built to take.

How Accurate Are They, Honestly?

Better than the alternatives, and considerably worse than the confident tone implies. When Fantasy Football Analytics measured projections against expert rankings and crowd-sourced ADP, projections accounted for roughly 53% of the variance in points players actually scored. Expert rankings managed 27%. Crowd rankings drawn from average draft position managed 23%. On that evidence, projections were close to twice as informative as either ranking method — which is worth remembering the next time somebody in your league dismisses the numbers as made up.

The second half of that finding deserves equal weight. Something like half the variation in real scoring was explained by none of the three methods. Injuries nobody saw coming, a coordinator deciding to run the ball forty times, a fumble on the goal line, and ordinary randomness all live in that unexplained half. A tool that gets you most of the way to an ordering is genuinely valuable. It is still not a decision, and treating it as one is how managers end up defending a bad start by pointing at a number.

One Number, Two Very Different Bets

Picture two flex candidates both projected for 11.5. The first is a high-volume back who will get his fifteen carries and three targets whatever happens; his realistic outcomes run from about 7 to about 16, and he lands near the middle most weeks. The second is a deep receiver whose figure is carried almost entirely by the chance of one long touchdown. He finishes with 3 points most weeks and 28 occasionally. Their projections match. Their distributions barely overlap.

Which one you want depends entirely on the week you are having, and that is information the projection does not carry. If you are a heavy favorite, the narrow player protects a win you are already likely to get. If you are down twenty points with one player left, the narrow player guarantees you lose by twelve instead of twenty. Sorting by projection alone silently picks the same answer in both situations. Recognizing which shape you are holding is most of what separates a good start/sit call from a defensible one, which is why it gets its own decision framework.

Why Two Sites Disagree, and Why That Helps You

Different providers build projections differently — some weight recent games heavily, some lean on multi-year baselines, some hand-adjust for news and some do not — so the same player routinely carries different numbers in different places. The instinct is to hunt for the one source that is right. The evidence says not to bother: Fantasy Football Analytics found individual providers swing wildly from year to year, with sources that led a position one season finishing near the bottom a few seasons later, while a simple average of many sources beat individual sources in 69% of head-to-head comparisons.

That turns disagreement from an irritation into a signal. When several independent sources cluster tightly on a player, the consensus is well founded and the number deserves the weight you are about to put on it. When they spread out — one site has him as a low-end starter and another as a bench body — the spread is telling you his role or his health is genuinely contested, and that is precisely the player whose week you should not decide from a single figure. Read the agreement, not just the average.

Three Things a Projection Structurally Cannot Know

• •Anything that has not happened yet, or that happened after the model last ran. Every projection carries a timestamp, and depth charts move faster than the systems reading them. A Friday number can be describing a Wednesday world.
• •Your league's exact rules. Most published projections are computed for a generic format. If your league awards six points for a passing touchdown, or a bonus at the tight end position, or fractional points per reception, the published figure is answering a slightly different question than the one your lineup asks.
• •What the coaching staff actually intends this week. Models infer game plans from history. They cannot see a staff that has decided to lean on the run, protect a returning starter with a snap count, or roll out a package they have never used.

The Movement Matters More Than the Level

Here is the part almost nobody uses. The most valuable thing about a projected total is frequently not what it says today, but that it is different from what it said yesterday. The level is somebody's opinion. A change is a fact about the world: a model revised a player downward because a practice report landed, a depth chart shifted, or a teammate was ruled out and the touches moved. The number often moves before the story does, because the inputs feeding it update on their own schedule rather than on the news cycle's.

This reframes what checking projections is for. Reading them once, on Sunday morning, gets you the level and nothing else — you see where the number ended up and have no idea whether it has been sitting there all week or dropped four points on Thursday. Comparing today against last time is what converts a static opinion into an alert, and it costs nothing beyond remembering the previous figure. It is also the reason a projection drop is worth investigating even when you cannot immediately find a headline explaining it.

The Four Checks Worth Running Beside the Number

1. 1Is the role real?Snap share, target share and touch counts tell you whether the projection rests on established opportunity or on an efficiency spike that may not repeat. See usage data for what to look at and over how many weeks.
2. 2Is he going to play, and in what shape?A projection is normally computed for a healthy player in a normal role. A designation, a limited practice, or a snap-count plan changes the answer without necessarily changing the number. Our guide to reading injury reports covers how much each designation is worth.
3. 3Has it moved since you last looked?Direction and size, not just the current figure. A number that fell overnight is reporting something. A number that has been flat for a week is reporting nothing at all, which is its own useful answer.
4. 4Does it hold in your league's scoring?Recompute anything close in your own format before it decides a lineup. The gap between two players is often smaller than the gap between two scoring systems.

A Working Rule

Treat the projected total as the opening argument rather than the verdict. It earns that status honestly — it beats rankings, it beats your gut, and it beats the guy in your league who watched one highlight. What it does not do is know things that have not been published, price your specific rules, or tell you how wide the range under it is. Start there, spend ninety seconds on the four checks, and overrule the number when one of them says something the number could not have seen. That is the entire discipline, and it is why we monitor projection changes rather than simply reprinting the current figures.