Stop averaging the last three years for every position

When a fantasy site projects next season, they usually average a player's last few years. Same recipe for every position. We tested 16 years of data and that recipe is wrong.

What the math actually says

Here's the optimal weight to put on each prior season when projecting next year, learned by checking what would have minimized error across 16 years:

The picture

How much weight should the most recent season carry?

RBs lean on last year 23× harder than they lean on three years ago. QBs barely lean at all — the most recent season is only 3-4× heavier than the three-year-old one.

Why it works this way

Short careers and shifting depth charts make RB and WR production highly current-year. Last season is everything. QB skill is durable — mechanics, processing, accuracy — so older seasons still predict. DB film holds up best of all.

What to do

1. For RBs and WRs, ignore old seasons. A 2023 line for a 2026 projection is mostly noise.
2. For QBs, don't write off a quiet last year. A QB who got hurt or switched teams in 2025 isn't a 2025 player — his 2024 and 2023 should pull projections up.
3. Question any "3-year average" ranking. It's wrong everywhere; it's just less wrong for older positions.