How Accurate Are Our Fantasy Projections?
Every week this site keeps the projections it published before kickoff, then scores them against what the players actually scored — next to the simplest honest forecast, each player's season-to-date average.
Log last updated Thu, Oct 8, 8:25 PM ET · tracking since Week 4
The 2026 season so far
Over Week 4 and 200 player-games, our projections missed by 5.00 half-PPR points per player on average, against 5.69 for each player's season average. In close start/sit calls the higher projection came out ahead 58.3% of the time (season average: 55.3%).
Swipe for rank correlation, top-N and close calls
| Position | Average miss | Rank correlation | Top-N hit rate | Close calls right | ||||
|---|---|---|---|---|---|---|---|---|
| Ours | Season avg | Ours | Season avg | Ours | Season avg | Ours | Season avg | |
| All positions Week 4 · 200 player-games | 5.00 | 5.69 | 0.440 | 0.359 | 54.2% | 54.2% | 58.3% | 55.3% |
| QB 30 player-games | 4.48 | 5.76 | 0.413 | 0.366 | 58.3% | 66.7% | 63.7% | 63.4% |
| RB 49 player-games | 5.97 | 6.22 | 0.494 | 0.517 | 70.8% | 70.8% | 57.4% | 60.4% |
| WR 79 player-games | 5.15 | 5.98 | 0.317 | 0.205 | 45.8% | 37.5% | 58.2% | 49.8% |
| TE 42 player-games | 3.95 | 4.48 | 0.536 | 0.348 | 41.7% | 41.7% | 56.9% | 48.7% |
Week by week
Swipe for rank correlation, top-N and close calls
| Week | Average miss | Rank correlation | Top-N hit rate | Close calls right | ||||
|---|---|---|---|---|---|---|---|---|
| Ours | Season avg | Ours | Season avg | Ours | Season avg | Ours | Season avg | |
| Week 4 200 player-games | 5.00 | 5.69 | 0.440 | 0.359 | 54.2% | 54.2% | 58.3% | 55.3% |
- Week 5 is not scored yet: 0 of 15 games final (last game Mon Oct 12, 8:15p ET).
How much a gap meant
Every pair of similar starters at the same position in the same week, grouped by how far apart the projections were as printed. The calibrated column is the 2024–25 rate the weekly rankings quote; the season-average column groups the same players by the gap between their season averages.
Swipe for the season-average comparison
| Gap | Reads as | Higher one won | Pairs | 2024–25 calibration | Season avg won | Pairs |
|---|---|---|---|---|---|---|
| Under 1 pt | Coin flip | 53.2% | 269 | 51.7% | 49.4% | 165 |
| 1–2 pts | Slight lean | 52.8% | 286 | 56.9% | 37.6% | 157 |
| 2–3 pts | Lean | 54.3% | 254 | 62.3% | 47.6% | 145 |
| 3–5 pts | Start | 59.6% | 322 | 67.8% | 55.5% | 292 |
| 5+ pts | Clear start | 66.4% | 432 | 78.3% | 61.3% | 804 |
The 2024–25 backtest
Before a single 2026 projection went up, the formula was fit on 2025 Weeks 4–11 and then tested week by week on games it had not seen, using only what was known before each kickoff: the rest of 2025, and all of 2024 with nothing re-tuned. These are those numbers, half-PPR, the same metrics as the 2026 tables above. The tracker exists to show whether 2026 looks like them.
The honest read: a real but modest edge. The average miss was 5.12 points against 5.34 for season averages late in 2025, and 5.11 against 5.19 across 2024. Close calls went right 63.9% and 62.1% of the time. Weekly fantasy scoring is mostly noise, so a small, steady margin over a strong baseline is what a real improvement looks like.
Where it lost. Quarterbacks early in the season: over 2024 Weeks 4–7, plain season averages missed by less (5.75 against 6.08) and got more close calls right (68.5% against 63.6%). Top-N at running back and receiver in 2024 was roughly a tie.
Swipe for rank correlation, top-N and close calls
| Test | Average miss | Rank correlation | Top-N hit rate | Close calls right | ||||
|---|---|---|---|---|---|---|---|---|
| Ours | Season avg | Ours | Season avg | Ours | Season avg | Ours | Season avg | |
| All positions 2025 Weeks 12–18 · 1,365 player-games | 5.12 | 5.34 | 0.417 | 0.369 | 56.4% | 52.4% | 63.9% | 60.6% |
| QB 2025 Weeks 12–18 · 214 player-games | 6.75 | 7.08 | 0.276 | 0.240 | 50.0% | 50.0% | 61.1% | 58.6% |
| RB 2025 Weeks 12–18 · 344 player-games | 5.47 | 5.71 | 0.442 | 0.409 | 66.1% | 67.3% | 65.6% | 64.6% |
| WR 2025 Weeks 12–18 · 530 player-games | 4.85 | 5.01 | 0.483 | 0.418 | 56.0% | 48.2% | 63.6% | 58.2% |
| TE 2025 Weeks 12–18 · 277 player-games | 3.96 | 4.18 | 0.466 | 0.411 | 53.6% | 44.0% | 61.0% | 56.6% |
| All positions 2024 Weeks 4–18 · 2,863 player-games | 5.11 | 5.19 | 0.475 | 0.428 | 58.6% | 58.0% | 62.1% | 61.6% |
| QB 2024 Weeks 4–18 · 441 player-games | 6.49 | 6.62 | 0.382 | 0.330 | 57.8% | 57.8% | 61.8% | 61.5% |
| RB 2024 Weeks 4–18 · 735 player-games | 5.31 | 5.36 | 0.559 | 0.548 | 69.7% | 70.8% | 65.5% | 65.7% |
| WR 2024 Weeks 4–18 · 1,144 player-games | 5.02 | 5.09 | 0.448 | 0.417 | 50.3% | 52.2% | 58.1% | 57.9% |
| TE 2024 Weeks 4–18 · 543 player-games | 3.93 | 4.02 | 0.511 | 0.419 | 56.7% | 51.1% | 64.7% | 59.4% |
| All positions 2024 Weeks 4–7 · 755 player-games | 5.07 | 5.28 | 0.477 | 0.392 | 61.2% | 57.3% | 61.5% | 59.4% |
| QB 2024 Weeks 4–7 · 117 player-games | 6.08 | 5.75 | 0.450 | 0.449 | 62.5% | 64.6% | 63.6% | 68.5% |
| RB 2024 Weeks 4–7 · 197 player-games | 5.52 | 5.71 | 0.525 | 0.476 | 68.8% | 66.7% | 65.3% | 63.0% |
| WR 2024 Weeks 4–7 · 299 player-games | 5.07 | 5.44 | 0.391 | 0.303 | 53.1% | 52.1% | 57.0% | 55.3% |
| TE 2024 Weeks 4–7 · 142 player-games | 3.62 | 3.95 | 0.541 | 0.339 | 60.4% | 45.8% | 62.5% | 52.3% |
Walk-forward, half-PPR. The model's settings were fit on 2025 Weeks 4–11 only; two quarterback settings were chosen after looking at 2024, so 2024 is not fully blind for those two. Source: this site's round B/C backtest.
Swipe for each season
| Gap | Reads as | Higher one won | 2025 Wks 12–18 | 2024 | Pairs |
|---|---|---|---|---|---|
| Under 1 pt | Coin flip | 52% | 52.8% | 51.2% | 7,798 |
| 1–2 pts | Slight lean | 57% | 57.2% | 56.7% | 7,128 |
| 2–3 pts | Lean | 62% | 64.3% | 61.4% | 5,862 |
| 3–5 pts | Start | 68% | 68.4% | 67.5% | 7,823 |
| 5+ pts | Clear start | 78% | 79.5% | 77.6% | 5,710 |
How the tracker keeps score
What is recorded. Every scheduled rebuild of this site copies the projections it is showing into a log, game by game. A game copied after it has kicked off is frozen and never changes again, so the projection that counts is the last one readers saw before kickoff — not a later revision. The whole log is public at /data/projection-log-2026.json: every player, his projected stat line and his half-PPR points.
What is scored. A week is scored once all of its games are final. Each player the site projected who then played is one player-game; his projection is compared with his half-PPR points. Players ruled Out were never projected, and a player who was projected but didn't play (a late scratch) is left out for both forecasts — the same rule as the backtest, so the two sets of numbers mean the same thing.
The comparison. Each player's season-to-date points per game before that week. It is the number most people would otherwise use, and a strong one: weekly fantasy scoring is noisy enough that an average is hard to beat.
- Average miss — mean absolute error in half-PPR points. Lower is better.
- Rank correlation — how well the order of the projections matched the order of the results (Spearman, 1 is perfect), computed per week and position and averaged.
- Top-N hit rate — of the top 12 QBs and TEs or top 24 RBs and WRs by projection, the share who finished in the top 12 or 24 that week.
- Close calls right — every pair of similar starters at the same position in the same week (the top 18 QBs and TEs or top 36 RBs and WRs by how much work they had been getting): how often the higher projection outscored the other. This is the start/sit question.
What the projections are. The same five-factor model as the weekly rankings: volume, efficiency, scoring chances, the Vegas line, and a market prior from Fantasy Football Calculator preseason ADP worth 2 games of evidence.
What these numbers can't tell you
One week is a small sample. A week is about 250 player-games and a few hundred close calls; a single week can swing either way by more than the whole backtest edge. The season-to-date table is the one to read.
Half-PPR only. The log keeps each projected stat line, so PPR and Standard can be checked from it, but the tables score half-PPR, the format the model was tuned on.
Stats can still change. Results come from nflverse's weekly stats, so a stat correction after a week is scored moves its numbers with it.
Kickers and defenses aren't tracked here. Their projections have their own backtests on the kicker and defense boards.
A fantasy tool, not betting advice. FantasyTools analysis is opinion for entertainment purposes only — not betting, financial, or professional advice.
Related tools
- Weekly rankings — this week's projections, with the math on every row.
- Rest-of-season rankings — the same model over the rest of the season, with its own backtest.
- Handcuff rankings — how much each backup's projection jumps if the starter sits.
- NFL stat corrections — the lines that changed after they were first published.
Get the week’s accuracy report in your inbox.
Two short emails a week in season. Tuesday: the waiver adds worth a claim, the defenses and kickers worth streaming, and who’s gaining targets. Saturday: injury report changes, line moves, wind and the close start/sit calls. No spam, unsubscribe any time.
We’ll email you once to confirm. See our privacy policy.
Next step
Seen the record? Here is this week’s board.
Every QB, RB, WR, TE, kicker and defense ranked by projected points in your league's scoring, with a floor-to-ceiling range and the math behind each number.
Open the weekly rankings2026 player stats, usage and the schedule via nflverse.com (CC-BY-4.0). Projections use preseason ADP courtesy of Fantasy Football Calculator (Sep 6–11 half-PPR drafts) as the market prior. The log was last updated Oct 8, 2026; the full log is at /data/projection-log-2026.json.