Track record
What we save, and when
Each week’s projections are saved before that week’s first kickoff, and never changed. Week 6 was saved Tue, Oct 13, before its first kickoff on Thu, Oct 15.
How it’s graded
After a week’s games, we score the saved numbers on real start/sit, FLEX and waiver calls, beside two simple alternatives: each player’s average over his last four games, and our preseason draft board.
Where each grade appears
Below, in 2026, week by week, once all of a week’s games are played: every graded week gets its own row. Week 6’s grade appears there once its games are played.
The draft boards
The Standard, Half PPR and PPR draft boards were saved the day they were served, each with a fingerprint to check it by, and we grade them here in January, win or lose. Custom-scoring boards aren’t saved. Check a board
Before it served
The test the weekly model passed first
A weekly model serves only after a test written down before it runs. The test replays the 2010–2019 seasons week by week, each season predicted only from the seasons before it, and compares the start/sit calls of what we serve (our model’s number blended with each player’s last-four-games average) with the average alone.
| Position | Model for week 4 | Model for weeks 5–6 |
|---|---|---|
| QB | +0.021 [+0.013, +0.030] | +0.028 [+0.020, +0.036] |
| RB | +0.018 [+0.015, +0.021] | +0.018 [+0.015, +0.022] |
| WR | +0.020 [+0.016, +0.023] | +0.020 [+0.017, +0.024] |
| TE | +0.017 [+0.010, +0.025] | +0.011 [+0.004, +0.019] |
| FLEX (RB, WR and TE together) | +0.019 [+0.017, +0.022] | +0.018 [+0.016, +0.021] |
The rule, fixed before the test ran: the whole interval above zero at QB, RB, WR, TE and FLEX, and none of the test’s 60 measures (three scoring formats, two sets of seasons, ten measures each) below zero. Each model passed. Kickers and defenses come from their own model, tested the same way against each team’s last-four-games average, the same in every format: K +0.038 [+0.027, +0.049], DEF +0.075 [+0.065, +0.086]. The defense test first read +0.045 [+0.037, +0.054]. That run read three teams’ defensive rows as zero, and it was run again on the corrected rows on October 2, by the same rule.
- Start/sit: of every pair of players at one position in a week, the share where the one projected higher scored more. +0.020 means 2 more right calls in every 100 pairs than the average made.
- FLEX: the same, with running backs, receivers and tight ends in one pool.
- The brackets: a 90% interval, the middle 90% of the edge over 2,000 resamplings of the 160 graded weeks.
Before start/sit used it
Counting a listed player’s chance to play
Start/sit counts a player the week’s official injury report lists Questionable or Doubtful at his projection times the share of players listed like him who played, 2009–2019 and 2024–2025. A test written down before it ran replayed the 2010–2019 seasons week by week on the projections saved each Tuesday, and compared start/sit calls with that step and without it, every listed player in the pairs, at 0 points when he didn’t play.
| Position | Change |
|---|---|
| QB | +0.0105 [+0.0063, +0.0149] |
| RB | +0.0285 [+0.0243, +0.0330] |
| WR | +0.0182 [+0.0142, +0.0224] |
| TE | +0.0160 [+0.0074, +0.0249] |
| FLEX (RB, WR and TE together) | +0.0225 [+0.0195, +0.0254] |
| The five, averaged | +0.0191 [+0.0163, +0.0221] |
The rule, fixed before the test ran: the five averaged with its whole interval above zero, and no position’s interval below zero. The result met it (graded 30 September 2026). In 2024–2025, which decided nothing, every position’s interval was above zero too: QB +0.0181, RB +0.0145, WR +0.0193, TE +0.0247, FLEX +0.0182.
Start/sit: of every pair of players at one position in a week, the share where the one ranked higher scored more. +0.0191 is about 2 more right calls in every 100 pairs.
Before the kicker and defense numbers used it
Setting the level of kicker and defense numbers
From week 6 a kicker’s or a defense’s number is our model’s number less a level: how far the model’s numbers sat above or below the results over the last season of games played, the same amount for every team that week. A test written down before it ran replayed the 2010–2019, 2024 and 2025 seasons week by week, each week reading only the results before it, and compared how far a season’s numbers sat from that season’s results with the level and without it.
| Position | Nearer by |
|---|---|
| Defense | +0.258 [+0.123, +0.396] |
| Kicker | +0.148 [+0.054, +0.243] |
The rule, fixed before the test ran: nearer in at least 9 of every 10 redraws of the twelve seasons, and over 2010–2019 no interval wholly below zero. Both positions met it (graded 2 October 2026).
Level: the average of a number minus the points scored, over a season’s games. Every team in a week takes the same level, so it changes no ranking and no start/sit call.
Scoring format