Week 1 · 2026 Season
Week 1 Fantasy Football Projections
Week 1 projections and start/sit calls, every one with a P10–P50–P90 range. When the games are final we grade the call against what actually happened. Nothing gets pulled down after a bad Sunday.
Quick answer
Who should I start in Week 1?
Start the highest P50 at each lineup slot. The range decides the close ones. Take the wide P10–P90 band when you're chasing points and the narrow one when you're protecting a lead. We won't name a Week 1 starter before the model sees the slate. The full grade posts the day after the week's last game, starting Week 1.
Week 1 starters
Week 1 projections aren't published yet.
FAQ
When do Week 1 projections go live?
Week 1 projections post before the week's first kickoff, once the slate and the lines are in. A number can still move before kickoff when we rerun on a line change or a late injury designation. The grade is stricter than that: it lands on the first version we post before kickoff. A rerun updates the number you see, never the copy we answer for. Until the week posts, this page says so instead of running fake numbers. The draft board is live year-round, with a season-long P10–P90 on every ranked player.
How wide is a weekly projection range?
Wide. One game is a sample of one, so a weekly band runs far wider relative to its median than a season-long band does. The exact width is on this page per player once the week posts. The calls those bands ride on get graded week by week from Week 1, on the accuracy ledger — and so is the band itself: the ledger's Range hit column posts how often the actual finished inside the range we gave. In the season model's published backtest, 66.7% to 83.2% of outcomes landed inside a P10–P90 band built to hold 80%. Our ranges run narrow. We published that before anyone asked, and the weekly number will post the same way — narrow or not.
Do Week 1 projections account for the matchup?
Yes, through the game environment. The Vegas spread and total set the week's implied team totals, and a player's number scales with that environment — so a projection can move on a line change alone, with no news about the player. What v1 does not do yet: model plays, usage shares, and efficiency as separate steps. Those stay folded into one scoring rate, blended from this season's games, the last two seasons, and the position. Splitting them out is still ahead — the season model's v2 work was calibration and the WR change, not this.