What this is
A free hobby project that projects weekly fantasy points for quarterbacks, running backs, wide receivers and tight ends.
How the projections work
It predicts how many fantasy points one player will score in one game.
Separate checks stop the results of the game being predicted from leaking into the inputs.
What it looks at
Six kinds of information go into every projection.
How to read a projection
Every row on the Rankings board looks like this one. This example is Caleb Williams, the top-ranked quarterback in Week 7.
- Rank 1, tier 1vs NE57%20.5
Tiers and the starter line
Tiers group players whose projections are close. A dotted line marks a drop in projection between two tiers, and the rank blocks take the color of their tier. The model cannot tell most neighbouring players apart, so a tier means "grouped by where the biggest drops fall", not "provably different". Tiers are drawn for the top 36 at each position and everyone below shares a final Rest tier. Tier lines show when the board is sorted by projection.
The solid blue starter line marks where starters end in your league: teams times starting slots at that position. It starts at 12 teams with 1 QB, 2 RB, 2 WR and 1 TE. Change the league on the Rankings page. Flex spots are not counted.
Model vs recent form
The Form page lists the gap between a player's projection and their average points over their last 3 games this season. Early in the season that is every game played so far, so the last 3 and last 8 averages match. The model weighs more than recent scoring, so a hot streak can sit far above the projection. A gap is not a prediction of a slump. Players with no games this season are left out.
How accurate it is
These numbers come from 2024 and 2025 games the model never saw. A projection is a best guess, not a promise. Each week's projections are saved before games and checked on the Results page.
Typical miss is how many points a projection is off in a normal game, on average, so a smaller number is better.
Rank correlation shows how well the model orders players from best to worst, where 1.00 is a perfect order and 0 is no better than chance.
What it can't know
The model does not know about trades, depth-chart news, or injuries that happened after the data was gathered. Check the status tag (Questionable, Out, IR) next to each player before you set a lineup.
- 01Rookies and players with no recent games rest on less information, so treat their projections with extra care.
- 02Each player gets one number. The band around it shows the typical miss for that position, not a true interval.
- 03Only quarterbacks, running backs, wide receivers and tight ends are covered.
- 04Lineup odds add each starter's points curve and treat players as independent, except a quarterback and his own receiver, who are counted as linked. An Out starter counts as 0 points.
Scoring
Half-PPR with 100/200-yard bonuses (project league rules, see ARCHITECTURE.md).
Data sources and licensing
Three public sources feed the projections.
- nflverse-dataPlayer statistics and weekly injury reports.CC BY 4.0
- nflverse nfldataSchedules, results and betting lines.Public data
- Sleeper public APIStatus tags such as Questionable, Out and IR. It also provides your league and roster if you choose to connect Sleeper. This site is free and non-commercial.Public API
The projections are our own work built from that data, and are not endorsed by nflverse. The nflverse player data is shared under the CC BY 4.0 license, which asks for credit to nflverse.
Disclaimer
Projections are estimates for entertainment and research. They are not betting or financial advice, and they carry no guarantee of accuracy. This site is not affiliated with the NFL or any team.