DFS Projections Vs Simulations: Why Medians Lose NFL GPPs

Every NFL DFS lineup I have ever regretted came from trusting one number too much. A projection said a receiver was worth 14.2 points, the salary made the math work, and the lineup looked clean right up until Sunday played out a game script the projection never told me about. The number was not wrong, just incomplete, and the difference between those two things is the difference between cashing a double-up and winning a tournament. This piece is the case for why: what a median projection actually is, what it throws away, and why the part it throws away is the part that wins NFL GPPs. By the end I will put two receivers side by side with the exact same projection, where one belongs in your tournament lineups and the other is filler, and the projection alone cannot tell you which is which.

The Quick Answer

A DFS projection is one number, the middle of a player's range of possible outcomes. A simulation plays out the whole range, thousands of times, with the game scripts and teammate correlations attached. NFL tournaments pay their biggest prizes to extreme outcomes, so the tool that models extremes beats the tool that averages them away. The full mechanics, a worked two-receiver example, and the one format where projections are still the right tool are below.

What A Median Projection Actually Is

A projection is a compression. Behind a printed 14.2 sits an entire distribution of things that can happen to a football player on a Sunday: the two-catch dud in a run-heavy blowout, the ordinary 5-for-60 day, the 9-for-140 with two scores when his team falls behind and throws 45 times. A median projection takes that whole spread and reports the 50th percentile, the outcome he beats about half the time. Our team builds these inputs all season, and the process behind them is real work; our guide to how NFL DFS projections get made walks through it. You can browse the results yourself in the NFL DFS projections, ownership and stacks in the DataHub.

So the number is faithful to the middle of the range. The problem is what the compression deletes: the shape of the distribution around that middle. Two players with identical medians can have wildly different chances of a 30-point game. The single number carries no information about ceiling, no information about floor, and no information about which teammates spike alongside a player when his big day arrives. For half the formats in DFS that loss barely matters. For the other half it is everything, and the payout structure decides which half you are in.

Why NFL GPPs Pay The Tails

In a guaranteed prize pool with six figures worth of entries, the money is concentrated violently at the top, and the lineups that take it down are the ones where a stack hit its 95th percentile together. Competing for first place takes a score in the extreme right tail of your lineup's own distribution, the kind of outcome a median projection is specifically built to average away. Our breakdown of what Milly Maker winners actually look like makes the same point from the results side.

Variance cuts the other way too, and the spread deserves a plain look. Our NBA DFS Strategy show broke down one night in April 2026 where the best lineup in the pool by simulated ROI finished 42,189th in the contest. The process was right and the result was ugly, because one contest is one draw from a distribution. A median projection hides that spread from you; a simulation forces you to look at it. Once you accept that a GPP is a fight over tail outcomes, the question stops being "which players have the best projections?" and becomes "which combination of players gives me the most chances at a slate-breaking score?" Those are different questions with different answers, and the second one needs three inputs the first one never uses.

The tools that answer the second question are about to cost more. Stokastic NFL annual pricing rises on September 1: Core moves from $549.95 to $649.95, Max from $749.95 to $899.95 and MVP from $1,299.95 to $1,499.95. An annual plan started before then keeps today's rate for the full season, playoffs included. The full Early Bird breakdown has the tier-by-tier math.

Lock In Early Bird Pricing

The Three Things A Median Cannot Hold

Ceiling: The Shape Of The Range

Volume drives NFL scoring, but the kind of volume matters. A slot receiver who catches six short targets every week and a deep threat who alternates two-catch games with 150-yard explosions can carry the same median while living in different universes. The deep threat's profile, high air yards on fewer targets, produces exactly the boom-or-bust distribution that tournaments reward and cash games punish. The median sees the two receivers as interchangeable. Sampling the full range thousands of times, a sim sees that one of them hits 25-plus points in a meaningful slice of outcomes and the other almost never does.

Correlation: Points Arrive Together

An NFL passing touchdown is scored by two players at once. When a quarterback hits his ceiling, his top receiver usually hits his ceiling in the same afternoon, which is why the QB-plus-pass-catcher stack is the foundation of tournament lineup construction. Median projections are computed player by player, so a spreadsheet of medians treats your quarterback and his receiver as independent line items and quietly understates what the pair does together on their best day. Simulations play out whole games, so the correlation is native: in the sims where the QB throws for 350, the receivers eat too, and the stack's simulated ceiling reflects the sum of outcomes that actually co-occur.

Game Script: The Spread Is A Projection Too

Every median projection silently assumes a blended game script. Real games are not blended. A team that falls behind by two scores abandons the run and force-feeds its receivers; a team protecting a lead hands off into the fourth quarter and its passing game goes quiet. The Vegas spread tells you which scripts are likely, and a simulation samples across them: some sims play out the shootout, some the blowout, some the grind. Take a real number from our first look at the DraftKings Week 1 board, the August 1 snapshot of a slate DraftKings can still reprice before kickoff: Jahmyr Gibbs is the most expensive player on it at $8,000. In sims where Detroit leads wire to wire he racks up fourth-quarter carries; in the sims where they trail, his volume shifts toward the passing game and part of that $8,000 goes to waste. One number cannot carry both futures. A distribution can.

A Worked Example: Two Receivers, One Projection

Here is the promised head-to-head. The numbers below are illustrative round numbers to show the mechanic, not tool output for real players; the shapes are the archetypes described above.

A projections-only process cannot separate these players; every tool that consumes only the median treats the two rows as identical. A tournament player should almost always prefer Receiver B, and the reasons live entirely in the deleted information: nearly five times the tournament-winning outcomes, arriving in the same sims where his quarterback erupts, which means a stack built around him multiplies its ceiling instead of adding two medians. The honest cost is the 30% dud rate. Simulations price that risk, and they tell you the tournament payout structure is paying you to accept it.

Salary pressure makes the distinction sharper, and here is where the Week 1 board comes back. Quarterback pricing on that first slate is compressed: Josh Allen tops the position at $7,000 and only $1,000 separates him from the eighth-priced QB. Cheap access to elite quarterbacks makes full stacks unusually affordable, which raises the ceiling of every lineup built to exploit correlation. Reading pricing shape and distribution shape together is the whole craft, and it is a read a single column of medians cannot make.

Projections Vs Simulations Vs Optimizers

One distinction worth keeping clean, because the terms get blurred together. An optimizer is a consumer of projections: it takes the median column and solves for the highest-projected lineup under the salary cap. Built that way, it inherits every blind spot described above and adds a new one, since thousands of entrants feed similar medians into similar solvers and converge on the same builds. That crowding problem, and why simulated fields beat deterministic solvers, is its own topic; our NFL DFS Sims vs optimizers breakdown covers it in full. The short version for this piece: projections are an input, an optimizer is one way to consume that input, and a simulation is a different consumer that keeps the whole distribution alive instead of collapsing it first. The Stokastic Sims also fold in projected ownership, so the output ranks lineups by how often they beat the field, and the field is the thing a GPP is actually played against. Ownership is where the ranking gets interesting in practice: hand the sim two builds with near-identical projected totals, one carrying a heavily owned chalk receiver and one carrying his lightly owned teammate, and the low-owned build ranks higher, because in the sims where both receivers hit, it passes thousands of duplicated entries instead of splitting the prize with them.

Where Median Projections Still Win

None of this makes projections obsolete, and this article should say so plainly. In cash games, double-ups and 50/50s where beating roughly half the field pays the same as finishing first, the median is precisely the number you want. Cash play is a hunt for floors, and a high-floor lineup built straight off projections is the correct tool for that job. The compressed quarterback board from earlier reads differently through this lens: when the Week 1 slate's top-priced QB costs only $1,000 more than the eighth-priced starter, paying up for the highest floor at the position is nearly free, and ranking those floors is exactly the job the median column does best. The simulated-tournament machinery, percentage-to-first framing, leverage and heavy stacking are GPP concepts and should stay in GPPs.

Projections are also the raw material simulations are built from. Every sim run starts from a projected distribution per player, which is why we publish both and why the free Sims access exists: the fastest way to understand what a distribution adds is to run one on a real slate next to the projection column and watch the rankings disagree. When they disagree, the sim is usually seeing correlation or ceiling the median cannot. Our walkthrough on how to use the NFL DFS Sims shows the full workflow once the season data goes live.

The Season Starts Priced In

Back to the first regret. The receiver who burned me was a distribution I never looked at, in a game script I never priced. The thesis of this whole piece fits in that one sentence: NFL GPPs are decided by ceilings, correlations and scripts, and the median is the one number that contains none of them. The players who internalize that before Week 1 build different lineups than the field all season.

The timing matters this year for a boring financial reason. Stokastic NFL annual pricing goes up on September 1, and an annual plan locked before the increase keeps the Early Bird rate through the season and playoffs. Running month-to-month at the new $149.95 Core rate for the five months of season you cannot skip comes to $749.75, about $200 more than the $549.95 Early Bird annual, and the gap grows from there on Max and MVP. If simulations are going to be part of your process this season, the cheapest version of that decision exists for a few more days.

See NFL Early Bird Pricing

More on this:

• How To Use NBA DFS Projections To Build Better Lineups
• The 4-Step Kalshi Prop-Edge Workflow For DFS Players
• How To Make Your Own NFL DFS Projections
• NBA DFS Projections: How To Build Better Lineups
• MLB DFS Ownership: How To Use Projected Ownership To Optimize Your Lineups

FAQ

What Is The Difference Between DFS Projections And Simulations?

A projection reports one number per player, the middle of his range of outcomes. Simulations play the slate out thousands of times and keep the whole range, including each player's ceiling games, the teammates who spike with him, and the game scripts that produce those outcomes.

Are Median Projections Useless For NFL DFS?

No. They are the right tool for cash games, where beating half the field rewards floor over ceiling, and they are the raw input every simulation is built from. They are the wrong tool to use alone in large-field tournaments, where payouts concentrate on extreme outcomes.

Do The Stokastic Sims Replace Projections?

They consume them. The Sims start from projected ranges for every player, then simulate full games so correlation, game script and projected ownership shape the output. The result ranks lineups by how often they beat a realistic field rather than by their summed medians.

Should I Use Projections Or Simulations For Cash Games?

Projections. A cash lineup wants the highest floor you can buy, and the median is a floor-and-middle number. Save the simulated-tournament tools, leverage and heavy stacking for GPPs, where differentiation gets paid.

Why Does Correlation Matter More In NFL Than In Some Other Sports?

Because NFL scoring events are shared. A passing touchdown scores for the quarterback and the receiver at once, so their big games arrive together. Lineups built to capture that co-movement have higher ceilings than the sum of their individual projections suggests.

When Does Stokastic NFL Early Bird Pricing End?

September 1, 2026, when NFL annual and monthly pricing increases. Annual plans started before then keep the current rate for the full year.