How these rankings are made
A projection is one number standing in for a range of things that could happen on Sunday. Ours is built from what players have actually done, what their situation is now, and what the betting market expects their team to score. No part of it is a judgement call made after seeing the number.
This page is the whole method in plain terms. Every section opens for the detail underneath, and nothing here is a summary of something we won’t show you.
- Built by
- OUR MODEL, ALONE
- Human edits
- NONE
- Input families
- 5
- Revised after publish
- NEVER RETROACTIVELY
Five families of input. Everything the model sees falls into one of them, and each is something you could look up yourself.
Production history What the player actually did, weighted toward recent form.
The largest family by some distance. We use per-game rates rather than season totals, so a player who missed time is not punished twice, and we look at both the best three-game stretch and the final three games of last season — one says what the ceiling looked like, the other says how the year ended. Two-season averages come in where a single year is too small a sample to trust, which is most of the time at defence and kicker.
Role and situation Whether the job itself changed.
A projection built only on history assumes the player walks into the same situation, and often they do not. We flag a change of team and a change of head coach, because both reset the assumptions underneath a usage rate: a new coordinator can halve a target share without the player getting any worse. Where the depth chart around a player has changed materially, that lands here too.
Age and contract stage Where a player sits on their own curve.
Age is carried to one decimal because the difference between 29.1 and 29.9 matters at running back and barely registers at quarterback — the curve is position-specific, and we fit it that way rather than applying one penalty across the board. Contract stage is a usage signal, not a value judgement: players on rookie deals tend to be given the ball while a team finds out what it has.
Market prices What sportsbooks expect the team to score.
This is a price, not an opinion. A sportsbook line has real money on both sides of it, and it moves as information arrives — which makes it the fastest-updating input we have and the main reason a weekly projection changes between Tuesday and Sunday. We read three: the implied team total, the spread, and the game total. A back on a team implied for 27 points is in a different week from the same back implied for 17, and no amount of production history sees that coming.
Schedule strength The quality of what they have to play.
Measured from opponent efficiency rather than from opponent record, which is noisy and reflects the wrong things. The season-long board uses the aggregate over a full schedule; the weekly board uses the single opponent in front of them. It matters most at defence, where the opposing offence is essentially the whole projection, and least at kicker.
Same order every time, run start to finish on every publish. Nothing is adjusted by hand afterwards.
Project per-game production Every player gets a projected points-per-game figure.
The model produces one per-game number per player, in a base scoring format, from the five input families. Per-game rather than season totals is a deliberate choice: it keeps a sixteen-game player and a twelve-game player comparable, and it is the unit an actual lineup decision is made in.
Convert to your scoring format The same projection, re-scored for PPR, Half-PPR or Standard.
Reception volume is the only thing that differs between the three, so the conversion is applied to the projection rather than the model being run three times. Switching format on the board re-scores every figure — the ranking you see is always the ranking for the format you selected, not a PPR board with different labels.
Rank within position Players are ordered against others at their own position.
There is no cross-position ranking on the board, because a tight end and a quarterback are not competing for the same roster slot. Defences are ranked as teams rather than players, which is why that board has a different shape.
Measure against replacement VOR says how much better than a replacement-level starter.
Replacement level is the projection of the player at the position's last startable slot in a standard league. Subtracting it turns six separate position boards into figures that can be compared across positions — which is the number that actually matters on draft day, when the question is not “is this the best receiver left” but “is this receiver worth more than that back”.
Publish to a fixed depth Each board stops where our conviction does.
Below a certain rank the projected gap between adjacent players is smaller than the week-to-week noise around either of them, and ordering them is a false precision. We set that depth per position — currently 18 at QB, 46 at RB, 62 at WR, 17 at TE, 18 at K, 16 at DST — and the board simply ends there.
The weekly board publishes Tuesday evening, ahead of Wednesday-morning waiver processing, then recomputes as the week’s information arrives. The Tuesday number is the one we are measured on — later refreshes never rewrite it.
- TUE, 6PM ET
- The week's board publishes, ahead of Wednesday-morning waiver processing on the major platforms. This is the figure our accuracy record is measured against.
- WED – SAT
- Recomputed as practice participation, role news and market prices move. The board shows how far each projection has travelled since Tuesday.
- SUN, AM
- The last recompute before the early kickoffs, for final start and sit decisions.
- PROJ
- Projected fantasy points per game, in the scoring format you have selected. One decimal.
- VOR
- Value over replacement: the projection minus the projection of a replacement-level starter at that position. Comparable across positions.
- Δ
- On the weekly board only: how far the projection has moved since Tuesday's publish. Green up, red down, unsigned zero when it has not moved.
- BOARD DEPTH
- How many players we publish at that position. The board ends there deliberately.
The useful half of a method is knowing where it stops. None of the following is a limitation we intend to remove later — they are the shape of the product.
- No accuracy guarantee.
- Weekly variance is larger than the difference between a good projection and a mediocre one. We publish our full historical record and we do not promise any particular week.
- No injury prediction.
- We do not forecast who gets hurt, and a projection assumes the player is available. Late news is your job on Sunday morning; the board will not know before you do.
- Our own model only.
- Every number on this site is our model's output. We do not blend, average, or reconcile it with anyone else's rankings, and we do not publish other people's projections alongside ours.
- No ordering of the factors.
- Player pages list what the model looked at, not how much each thing mattered. Ranking the inputs against each other would imply a precision the model does not have.
- Running backs show no factor table.
- The RB projection is built without a market blend and its inputs do not reduce to a handful of printable figures. The ranking and projection are the complete output there — we would rather show nothing than show something misleading.
METHOD LAST CHANGED · 12 AUG 2026 · CHANGES ARE DATED AND KEPT