What we can prove, and what we can’t.
Most dynasty value sites ask you to trust a number with no accounting behind it. This page is the accounting — including the parts that don’t flatter the model. Everything below is generated from the same files the site builds from; nothing here is hand-picked.
Values updated: Aug 25, 2026
Our model does not beat “just use last year’s points.”
We hid the 2022 season, rebuilt every input from 2021 and 2020 only, ran the model, and scored it against what actually happened over the following 4 seasons (440 players). Players who vanished from the league score zero rather than being dropped — 91 of them did, and excluding them would quietly grade us only on the players who stayed healthy.
The honest headline: ranking players by last season’s points per game predicts future production better than our dynasty value does. Here is the rank correlation with realized points, measured within each position — comparing across positions would flatter us, because quarterbacks score the most raw points and we deliberately discount them for 1QB scarcity.
| Predictor | QB | RB | WR | TE | Mean |
|---|---|---|---|---|---|
| Last season's PPG | 0.615 | 0.684 | 0.714 | 0.627 | 0.660 |
| Multi-season PPG blend | 0.457 | 0.648 | 0.647 | 0.511 | 0.566 |
| Dynasty Cheat Code value | 0.510 | 0.619 | 0.658 | 0.617 | 0.601 |
We are publishing that because it is true. Two things make it less damning than it first looks, and one of them is genuinely in our favour.
The buy/sell call adds something last year’s points do not.
Comparing all our buys to all our sells would prove nothing — buys land on better players to begin with. So we bucketed every player by their prior season’s points per game first, and compared calls only against players who had produced at the same level. Median realized points over the following 4 seasons:
| Prior production | Bullish | Neutral | Bearish |
|---|---|---|---|
| 0.3-4.0 ppg | — | 23 (n=77) | 28 (n=1) |
| 4.1-6.1 ppg | — | 45 (n=78) | — |
| 6.1-9.0 ppg | — | 119 (n=78) | — |
| 9.1-13.3 ppg | 676 (n=6) | 255 (n=71) | 266 (n=1) |
| 13.4-26.4 ppg | 1068 (n=19) | 595 (n=51) | 801 (n=9) |
Among players who were already producing at a high level, the ones we called bullish went on to score meaningfully more than the ones we called bearish. That is the model’s actual claim, and it survives the control. The samples behind those calls are small — read them as a direction, not a guarantee.
- Realized fantasy points are a proxy for dynasty value, not the thing itself. Dynasty value is trade value, and it prices years beyond any window we can test.
- A dynasty model is supposed to lose a one-year contest. It pays for youth on purpose. The gap between our model and last-season points narrows as the horizon lengthens — roughly halving from one year to four — which is the shape you want, but it has not closed.
- The opportunity signal could not be included. It needs the roster as it stood that offseason, and only the current one is available, so this test grades the engine without its one forward-looking input.
- One holdout season is one sample. We will keep publishing this as more seasons land, including when it moves the wrong way.
How the values are built is documented on the methodology page. If you think a specific call is wrong, tell us — that feedback is how the call-calibration guards got written in the first place.