We publish our accuracy. The industry hides theirs.
Dynasty General is built on calibrated prediction. We backtest our engine against named industry baselines (FantasyPros ECR, FantasyPros ADP, KeepTradeCut) on dynasty cumulative outcomes from 2022 to 2024. The numbers below are reproducible from the source rankings and historical actuals.
Two engine versions are reported, both as they stood on 2026-05-08. v0 uses position rubric scaffolding plus corpus-grounded age curves with no signal codes loaded. v1 adds 182 RB player-years of signals (role tier, role-at-new-team, traded flag, compounding-news count) coded by an LLM extractor with a pre-draft knowledge cutoff, RB only. v1 averages +0.022 absolute Spearman over v0 across the three years, with the gain concentrated on 3-year cumulative.
What this page is not: the accuracy of the product you use today. The live board changed on 2026-06-19; every value now runs through the position rubrics blended with the FantasyCalc market prior at a market-dominant weight, and the signals tested after this run (route participation, scheme and coaching history, offensive-line proxies) did not beat the market in backtest. The live engine has not been re-scored here. Each cell below is the newest run of that cell; superseded runs are dropped when the board is rebuilt.
Dynasty cumulative-PPR Spearman rank correlation
Higher is better. Each row is a (source, prediction year) snapshot. Cumulative window adapts to outcome data availability: 3-year for 2022, 2-year for 2023, 1-year for 2024 (will become full 3-year when 2025 and 2026 outcomes are ingested).
2022 prediction year
| Source | Spearman | MAR | n | Window |
|---|---|---|---|---|
| Dynasty General v1 (1QB, RB signals loaded) | 0.4415 | 61.40 | 87 | 3yr |
| Dynasty General v0 (1QB, no signals) | 0.3901 | 56.63 | 86 | 3yr |
| FantasyPros ADP | 0.4627 | 54.75 | 85 | 3yr |
| FantasyPros ECR (1QB) | 0.4631 | 51.40 | 91 | 3yr |
| KeepTradeCut (1QB) | 0.3538 | 52.83 | 78 | 3yr |
2023 prediction year
| Source | Spearman | MAR | n | Window |
|---|---|---|---|---|
| Dynasty General v1 (1QB, RB signals loaded) | 0.4744 | 33.09 | 67 | 2yr |
| Dynasty General v0 (1QB, no signals) | 0.4674 | 30.45 | 66 | 2yr |
| FantasyPros ADP | 0.5354 | 43.36 | 73 | 2yr |
| FantasyPros ECR (1QB) | 0.4327 | 37.17 | 72 | 2yr |
| KeepTradeCut (1QB) | 0.4485 | 30.79 | 62 | 2yr |
2024 prediction year
| Source | Spearman | MAR | n | Window |
|---|---|---|---|---|
| Dynasty General v1 (1QB, RB signals loaded) | 0.3464 | 56.89 | 70 | 1yr |
| Dynasty General v0 (1QB, no signals) | 0.3393 | 56.70 | 70 | 1yr |
| FantasyPros ADP | 0.0932 | 53.22 | 73 | 1yr |
| FantasyPros ECR (1QB) | 0.2001 | 48.95 | 73 | 1yr |
| KeepTradeCut (1QB) | 0.2358 | 52.00 | 64 | 1yr |
Headline (averaged across 2022-2024, top-100 predicted, 1QB)
- Dynasty General v1 (1QB, RB signals loaded): Spearman 0.421 (2022 0.442 · 2023 0.474 · 2024 0.346)
- Dynasty General v0 (1QB, no signals): Spearman 0.399 (2022 0.390 · 2023 0.467 · 2024 0.339)
- FantasyPros ECR (1QB): Spearman 0.365 (2022 0.463 · 2023 0.433 · 2024 0.200)
- FantasyPros ADP: Spearman 0.364 (2022 0.463 · 2023 0.535 · 2024 0.093)
- KeepTradeCut (1QB): Spearman 0.346 (2022 0.354 · 2023 0.449 · 2024 0.236)
Honest read: DG wins on AVERAGE; per-year picture is mixed. In 2024 (1-year cumulative) DG dominates by a wide margin because consensus rankings collapsed on injury-driven busts (CMC, Aiyuk, Cooper Kupp, Tyreek Hill). On the longer horizons (2022 3-year, 2023 2-year) FantasyPros wins by ~0.02-0.06. Average win is real; per-year ranking deserves scrutiny. Sample sizes 60-90 joined pairs per cell; confidence intervals roughly ±0.05 to ±0.10.
Methodology
We score each source's preseason ranking against the actual cumulative PPR points scored over the post-prediction window. The loss function follows VALIDATION_PLAN section 8.2:L_dynasty = 0.6 × KTC_6mo_value_MAE + 0.4 × cumulative_3yr_PPR_RMSE. v0 publishes the cumulative-PPR-rank correlation component; v1 will composite KTC value drift into the full weighted loss.
Source snapshots: FantasyPros rankings accessed under a paid trial in May 2026 from dynasty-overall.php?year=YYYY&week=0. KeepTradeCut snapshots from the public Wayback Machine archive. Sleeper season totals from the public stats endpoint.
We publish derived metrics (Spearman correlations, MAR, hit rates). We do not republish source rankings or player-by-player orderings. To reproduce a row: subscribe to the source, capture the preseason snapshot, score against actuals, compute the metric. Scripts are in the public repository under web/scripts/.
Industry prior art: FantasyCalc maintains a public performance- analysis page comparing their values to FantasyPros ECR by name. Comparative-accuracy publication is standard.
Caveats
- v1 signal coverage is RB-only (182 RB player-years, LLM-coded with a pre-draft knowledge cutoff, 2022-2024). QB / WR / TE predictions in this run use rubric + age curve only. The signals tested after this run (route participation, scheme and coaching history, offensive-line proxies) did not beat the market; see negative-results in the repository.
- Per-position breakdown shows DG loses to FP within-position on QB and RB across years; WR is roughly tied. The DG top-100 average win comes from cross-position bust avoidance (correctly downgrading aging starters that consensus overranked) more than from within-position ordering.
- The horizon-gating thesis (that workload-dependent signals like the bellcow boost should be horizon-gated) was considered after the partial-coverage v1 result and refuted by the full-coverage data. v1 wins biggest on 3-year cumulative. The boost is not horizon-gated.
- Sample sizes are small; confidence intervals are ±0.05 to ±0.10 on individual Spearman values. The average is more reliable than the year-by-year ordering.
- The 2024 prediction is currently scored on a 1-year window only. When 2025 and 2026 outcomes are ingested, this row will be re-scored on the full 3-year cumulative.
- Engine universe is restricted to KTC top-200; we cannot find a "diamond in the rough" outside that universe.
- Past backtest performance does not predict future seasons.
Backtest run 2026-05-08; page copy updated 2026-09-15. Dynasty General is not affiliated with, endorsed by, or sponsored by FantasyPros, Marzen Media LLC, KeepTradeCut, FantasyCalc, or Sleeper. Comparative claims reflect public product observation and our own backtest methodology as of 2026-05-08; see /scoreboard/methodology for sources and reproducibility notes. FantasyPros is a trademark of Marzen Media LLC, used here for comparative reference under nominative fair use.