A $20 entry into a 150,000-entrant tournament
The engine builds an opposing field of 10,000 lineups from projected ownership, simulates the slate 5,000 times with players correlated inside their own games, and ranks every lineup by expected return against a payout curve built from that contest's settings (the same approximation described under Methodology below, not the contest's exact prize table), choosing on one half of the draws and reporting the other. This run scored 10,000 lineups in 23.5 seconds.
| # | Lineup | Salary | Avg score | ROI (simulated) | Cash % | Ownership | Top 1% |
|---|---|---|---|---|---|---|---|
| 1 | QBJared GoffRBChase BrownRBDe'Von AchaneWRAmon-Ra St. BrownWRGarrett WilsonWRMichael WilsonTESam LaPortaTEDalton SchultzDSTChargers | $50,000 | 130.5 | 635% | 31.6% | 131% | 3.66% |
| 2 | QBTua TagovailoaRBBijan RobinsonRBOmarion HamptonWRDrake LondonWRZay FlowersWRGarrett WilsonWRWan'Dale RobinsonTEHarold Fannin Jr.DSTRaiders | $49,900 | 130.1 | 218% | 29.6% | 164% | 2.78% |
| 3 | QBDaniel JonesRBBijan RobinsonRBChase BrownWRChris OlaveWRAlec PierceWRMichael WilsonWRWan'Dale RobinsonTEDallas GoedertDSTRavens | $49,800 | 135.8 | 564% | 40.9% | 173% | 3.60% |
| 4 | QBCam WardRBChase BrownRBKenny GainwellWRAmon-Ra St. BrownWRChris OlaveWRGarrett WilsonWRWan'Dale RobinsonTETucker KraftDSTEagles | $49,900 | 129.1 | 74% | 30.2% | 155% | 3.28% |
| 5 | QBC.J. StroudRBDe'Von AchaneRBD'Andre SwiftWRJa'Marr ChaseWRNico CollinsWRMichael WilsonWRWan'Dale RobinsonTEDalton SchultzDSTEagles | $49,800 | 125.7 | 457% | 23.0% | 133% | 1.86% |
| 6 | QBJared GoffRBBijan RobinsonRBChase BrownWRNico CollinsWRGarrett WilsonWRWan'Dale RobinsonWRDevaughn VeleTESam LaPortaDSTChargers | $49,800 | 127.1 | 42% | 25.3% | 143% | 2.10% |
| 7 | QBJustin HerbertRBChase BrownRBDerrick HenryWRZay FlowersWRWan'Dale RobinsonWRQuentin JohnstonTETrey McBrideTESam LaPortaDSTRaiders | $49,900 | 131.5 | 72% | 32.9% | 198% | 2.26% |
| 8 | QBJustin HerbertRBBijan RobinsonRBKenny GainwellWRChris OlaveWRGarrett WilsonWRDK MetcalfWRMichael WilsonTESam LaPortaDSTJaguars | $50,000 | 132.2 | 313% | 35.5% | 135% | 4.14% |
Generated 2026-08-19. The ROI column is not sorted, and that is deliberate. Lineups are ordered by the engine's own ranking, made on one half of the draws; the ROI shown is what each earned on the other half. Sorting the board by that column would just promote whichever lineup got luckiest in the half being reported, which is the exact bias the split exists to remove. When a tool's headline number always sorts perfectly, it is worth asking whether it was chosen and scored on the same draws, because that alone would inflate it.
What the engine knows about one lineup
NE @ SEA, with the juice removed
A −110/−110 market implies 52.4% + 52.4% = 104.8%. That extra 4.8% is the house's cut, not anybody's opinion. Strip it out and you can see what each book actually thinks, and what you are paying for the privilege.
| Book | Home | Away | Raw implied | Hold | Fair home % |
|---|---|---|---|---|---|
| DraftKings | -110 | -110 | 104.76% | 4.76% | 50.00% |
| FanDuel | -112 | -108 | 104.75% | 4.75% | 50.43% |
On this game FanDuel has the slightly lower combined overround, though the better price still depends on which side you take, and the books disagree by 0.43 points of fair probability once the vig is stripped. Neither is visible on the raw price.
How the simulator actually works
Projections come first, and they are ours. Betting lines and DFS salaries are inputs; other sites' projections are not. For the NFL, each team's passing yards are projected as one number from its implied total, the spread and its recent passing volume, then split between the quarterback and his receivers, so the pieces add up (since 17 October 2026). The NFL projection you build with blends that number with what the DraftKings salary implies. NBA projections (beta) are expected minutes times a per-minute rate, with each player's chance of playing estimated from his recent availability.
Injury news is read through the day. For the NFL we read the league's official practice and game-status reports, daily depth charts and weekly rosters every hour, and every 20 minutes on game days; for college football, the availability reports Power 4 schools publish. A player listed out or doubtful is removed, a questionable one is discounted, and on NFL Classic slates a player who will not dress (a practice-squad player, for example) is left out. If news lands after you build, the builder says so and offers to re-run the sims or late swap the lineups on screen (Classic slates).
The field is people, not optimizers. A tournament is not won against the best lineups, it is won against the lineups people actually enter. We build the opposing field from projected ownership with a per-lineup appetite for chalk, quarterback stacking and bring-backs, so it contains both the chalk-chasers and the contrarians a real contest has.
Players are correlated inside their game. Scores are drawn jointly through each game's correlation structure, not independently. A quarterback's ceiling and his receiver's ceiling arrive together, because in football they do.
Every player's range is shaped to his projection. Outcomes are shaped to that player's own projection and position: the NFL ranges come from historical results for similarly projected players at the same position, scaled to his projection, rather than a generic curve scaled off his median.
Prizes follow a payout curve built from your contest's settings. The curve uses the entry count, fee, rake, share of entries paid and first-place share, so it approximates the real prize table rather than copying it. Rank is read from where a lineup finished in the simulated field, then mapped to the block of places on that curve it stands for. Duplication is charged once, when copies genuinely spill past that block.
ROI is measured out of sample. The engine ranks lineups on one half of the draws and reports their return from the other half. Ranking thousands of lineups by a noisy estimate and then quoting the winner's own number always flatters it. That gap is luck, and we throw it away rather than sell it to you.
Which tools this describes. The lineup builder (/dfs/build/) and the top plays on the optimizer (/dfs/optimizer/) both rank this way: lineups are chosen on one half of the draws and their return is reported from the other. Top-1% finish rate is not the ranking key, and the builder shows it as a rank rather than a percentage, because the raw rates ran high against real contests. Method in place since 18 August 2026, checked against the production engine (build 9ee914a) on 15 September 2026. The worked example above was frozen on 2026-08-19 with the same method.
Depth is not a luxury. How much of the top 150 repeats between two runs depends on the slate and on whether the candidate lineups are regenerated. In our 30 September study on 550 NFL contests, with the candidate lineups held fixed, the median overlap at 5,000 draws was 43%. When the whole build was re-seeded, the overlap was 0 to 2%. We publish that because it is the honest limit of what a single lineup's ROI can tell you.
How it has held up. The ownership test, the field scorecard and the track record of our projections, props and college lines, misses included, are on How Our Model Works.
See it on this week's slate
The numbers above are frozen so you can audit them. The live tool runs the same engine against the current week, with your contest's size, entry fee and payout settings.
View pricing