1999–2026
Benchmarks
Every model in The Prediction Tracker's archive and our own rows, scored the same way: flat one-unit stakes against the closing spread at standard −110 juice. A win rate above 52.4% beats the vig — the field mostly doesn't, which is the point of benchmarking.
Source: The Prediction Tracker, seasons 1999–2026 · 57 models · 230,013 graded picks · plus 2 rows of ours · rebuilt October 21, 2026.
Career records — every graded season pooled into one row per model.
- Models tracked
- 57
- Graded picks
- 230,013
- Beat the vig
- 1 / 57
- Profitable all-time
- 1 / 57
1999–2026, plus our own rows
across the archive
win rate ≥ 52.38%
units at −110
| # | Trend | |||||||
|---|---|---|---|---|---|---|---|---|
| 1 | Betting Benchmarks — board playsours | 6 (2021–2026) | 666 | 52.10% | -0.53% | 50% | -3.5u | |
| 2 | ARGH Ratings | 25 (2000–2024) | 5,811 | 51.87% | -0.98% | 48% | -57.0u | |
| 3 | FF Winners | 13 (2014–2026) | 3,035 | 50.97% | -2.69% | 38% | -81.6u | |
| 4 | Cleanup Hitter | 13 (2014–2026) | 3,169 | 50.93% | -2.77% | 38% | -87.7u | |
| 5 | Beck Logistic Regression | 25 (2002–2026) | 5,971 | 50.93% | -2.77% | 40% | -165.5u | |
| 6 | Sagarin Prediction | 25 (2002–2026) | 6,221 | 50.88% | -2.87% | 36% | -178.7u | |
| 7 | Betting Benchmarks — model lineours | 6 (2021–2026) | 1,472 | 50.82% | -2.99% | 33% | -44.0u | |
| 8 | rp-excel | 18 (2008–2025) | 4,444 | 50.79% | -3.04% | 39% | -135.2u | |
| 9 | Kambour Ratings | 13 (2006–2026) | 2,808 | 50.71% | -3.19% | 31% | -89.5u | |
| 10 | PerformanZ Ratings | 28 (1999–2026) | 7,029 | 50.70% | -3.20% | 36% | -225.0u | |
| 11 | Sagarin | 28 (1999–2026) | 6,964 | 50.69% | -3.23% | 29% | -224.9u | |
| 12 | Massey Ratings | 24 (1999–2026) | 5,793 | 50.66% | -3.28% | 33% | -189.8u | |
| 13 | Talisman Red's Ratings | 13 (2014–2026) | 3,170 | 50.66% | -3.28% | 15% | -104.0u | |
| 14 | PiRatings | 13 (2014–2026) | 3,209 | 50.64% | -3.33% | 31% | -106.7u | |
| 15 | Game-Time Decision | 19 (2008–2026) | 4,187 | 50.63% | -3.34% | 32% | -139.7u | |
| 16 | Prediction Tracker: Median | 23 (2004–2026) | 5,609 | 50.62% | -3.37% | 35% | -189.1u | |
| 17 | Dokter Entropy | 20 (2006–2025) | 5,074 | 50.57% | -3.45% | 30% | -175.3u | |
| 18 | StatFox | 23 (2001–2023) | 5,476 | 50.55% | -3.50% | 26% | -191.6u | |
| 19 | Sonny Moore's Ratings | 28 (1999–2026) | 6,921 | 50.51% | -3.57% | 21% | -246.8u | |
| 20 | Beck Least Squares HFA | 28 (1999–2026) | 6,920 | 50.49% | -3.61% | 25% | -249.6u | |
| 21 | Prediction Tracker: Average | 27 (2000–2026) | 6,714 | 50.45% | -3.69% | 26% | -247.9u | |
| 22 | Benchmark: Line Open | 24 (2003–2026) | 4,775 | 50.43% | -3.73% | 25% | -177.9u | |
| 23 | Ken Ashby's Accuratings | 23 (2004–2026) | 5,271 | 50.35% | -3.88% | 26% | -204.3u | |
| 24 | Coffey Ratings | 25 (2002–2026) | 4,463 | 50.28% | -4.01% | 28% | -179.0u | |
| 25 | Computer Adjusted Line | 23 (2004–2026) | 3,827 | 50.27% | -4.02% | 17% | -153.9u | |
| 26 | Beck Points Per Game | 13 (2014–2026) | 3,275 | 50.20% | -4.17% | 15% | -136.5u | |
| 27 | Roger Johnson | 19 (2006–2026) | 3,807 | 50.14% | -4.27% | 37% | -162.5u | |
| 28 | Dunkel Index | 25 (2000–2026) | 6,111 | 50.12% | -4.31% | 28% | -263.5u | |
| 29 | Director of Information | 7 (2012–2026) | 1,648 | 50.12% | -4.31% | 14% | -71.1u | |
| 30 | Beck Turnover Adjusted Least Squares | 15 (2012–2026) | 3,803 | 50.09% | -4.37% | 27% | -166.2u | |
| 31 | Lou St. John | 14 (2013–2026) | 3,468 | 50.09% | -4.38% | 29% | -151.9u | |
| 32 | ESPN FPI | 12 (2015–2026) | 3,003 | 50.05% | -4.45% | 33% | -133.6u | |
| 33 | Pigskin Index | 26 (2000–2025) | 6,081 | 49.98% | -4.59% | 12% | -279.3u | |
| 34 | Pythagorean Ratings | 28 (1999–2026) | 6,889 | 49.92% | -4.70% | 11% | -323.6u | |
| 35 | Beck Least Squares Regression | 28 (1999–2026) | 6,968 | 49.91% | -4.71% | 29% | -328.2u | |
| 36 | PiRatings Mean | 13 (2014–2026) | 3,203 | 49.83% | -4.87% | 23% | -156.1u | |
| 37 | Beck Least Absolute Value Regression | 28 (1999–2026) | 6,925 | 49.82% | -4.89% | 21% | -338.6u | |
| 38 | TeamRankings | 11 (2016–2026) | 2,716 | 49.78% | -4.97% | 9% | -134.9u | |
| 39 | Beck Scoring Efficiency | 28 (1999–2026) | 6,855 | 49.74% | -5.03% | 11% | -345.0u | |
| 40 | Sagarin Golden Mean | 13 (2014–2026) | 3,269 | 49.71% | -5.10% | 23% | -166.7u | |
| 41 | Stephen Kerns | 21 (2006–2026) | 4,845 | 49.58% | -5.35% | 19% | -259.4u | |
| 42 | DCI Pro Football | 12 (2014–2025) | 3,167 | 49.54% | -5.42% | 17% | -171.6u | |
| 43 | DRatings | 14 (2013–2026) | 3,470 | 49.54% | -5.43% | 7% | -188.3u | |
| 44 | Beck Elo | 27 (2000–2026) | 6,685 | 49.44% | -5.62% | 11% | -375.5u | |
| 45 | Benchmark: Line Midweek | 14 (2013–2026) | 2,534 | 49.29% | -5.90% | 14% | -149.5u | |
| 46 | PiRatings Bias | 13 (2014–2026) | 3,205 | 49.17% | -6.12% | 23% | -196.3u | |
| 47 | DPD's Dogs | 20 (2007–2026) | 4,679 | 49.03% | -6.40% | 10% | -299.5u | |
| 48 | RWP Team Ranks | 14 (2011–2025) | 3,575 | 48.98% | -6.49% | 7% | -232.2u | |
| 49 | Versus Sports Simulator | 16 (2011–2026) | 4,044 | 48.89% | -6.67% | 6% | -269.7u | |
| 50 | Hanson Crew | 10 (2016–2025) | 2,415 | 48.86% | -6.72% | 10% | -162.3u | |
| 51 | Congrove Computer Rankings | 3 (2024–2026) | 622 | 48.55% | -7.31% | 0% | -45.5u | |
| 52 | OddsShark Computer Picks | 4 (2013–2023) | 824 | 48.42% | -7.56% | 0% | -62.3u | |
| 53 | Roundtable Ratings | 13 (2012–2024) | 2,881 | 48.32% | -7.76% | 15% | -223.5u | |
| 54 | Sagarin Recent | 6 (2016–2026) | 1,426 | 48.25% | -7.89% | 33% | -112.5u |
54 of 59 rows (min. 3 seasons — ) · win % pooled over every graded pick · trend is cumulative profit by season · shaded rows are ours
Reading these tables
Coverage is not equal. Some models are graded for all 28 seasons, others for one — so the all-time view opens on models with 3+ seasons, sorted by ROI, which compares like with like. Set min. seasons to "any" and sort by profit to see raw cumulative totals instead (where longevity dominates).
ROI is profit per unit staked. At −110 a coin-flip model returns −4.5%, so anything positive is genuine edge — Profitable yrs is the share of a model's own seasons that finished in the black, which separates steady edges from one lucky year.
Picks are graded against the closing spread, the hardest available benchmark. Models publishing earlier in the week may beat the number they actually bet while still losing to the close — see our methodology.
Our rows
The shaded rows are ours. They are graded by exactly the rule above — one flat unit against the closing spread, pushes excluded — and they are not counted in the summary tiles, since the field is what we measure against.
- Betting Benchmarks — model line (2021–2026 · 1,472 graded picks)
Our projected margin for every game, graded against the closing spread like every other model here. - Betting Benchmarks — board plays (2021–2026 · 666 graded picks)
Only the games the board actually played (P(cover) at the opening line >= 0.53), on the side it picked, graded against the close. The record page grades the same plays against the open they were bet at.
Seasons through 2025 are a walk-forward backtest, not published picks. Each season's numbers come from a model fit only on earlier seasons, and the play threshold was chosen on that window — read them as descriptive. From 2026 on, every row is the board as it was published each week, graded once the games are played; the record page grades the same plays against the opening line they were actually bet at, which is the number we stand behind.
Source and citation
Model predictions come from The Prediction Tracker, Todd Beck's long-running collection of computer-model NFL forecasts. The site has gathered every participating model's number for every game since 1999 and publishes one archive file per season (nfl99.csv through nfl26.csv). The 2026 file is the complete season through the Super Bowl. Closing spreads and final scores come from nflverse.
We re-grade those files ourselves rather than quoting the site's totals. The Prediction Tracker's own results page scores models against its midweek line; every table here uses the closing line, so our records will not match the ones on their season totals exactly. Model names and homepages follow the site's legend; the Line Open and Line Midweek rows are the site's own market columns, included as benchmarks.
Cite as: Beck, T. The Prediction Tracker — NFL, https://www.thepredictiontracker.com/prednfl.html. Predictions 1999–2026; graded by Betting Benchmarks against closing spreads from nflverse.