PlayDelta reads the gap between price and model.
Market quotes and model estimates answer different questions. The board keeps their sources, timing, and limitations visible.
Methodology reviewed September 28, 2026 · 21+What PlayDelta is
An independent analytics interface that helps users understand where a sports market and a model read may disagree.
What PlayDelta is not
PlayDelta is not a sportsbook, exchange, broker, financial adviser, or betting-instruction service.
A probability model independent of market prices.
- 1. Build team historyUse completed regular-season games from MLB's official schedule service. Market prices are never training inputs.
- 2. Estimate strengthCombine opponent-adjusted win strength, smoothed run differential, recent form, and designated home advantage in a regularized logistic model.
- 3. Test later seasonsFit on 2021–2023, choose model settings on 2024, and evaluate the frozen choice on 2025. Results enter the model with a conservative delay to reduce timing leakage.
- 4. Match the exact gameRequire an unambiguous matchup and official pregame start time. Ambiguous doubleheaders, unsupported games, and stale data produce no comparison.
- 5. Compare the same outcomeSubtract that team's current ask from its modeled win probability. Estimates expire at first pitch. A positive gross gap stays a research observation, not a trade signal.
Model probability minus ask price.
For example, an estimated 55% win probability against a 52¢ ask produces a +3 percentage-point gross gap for a $1 winner contract. Fees, slippage, settlement differences, and model error can erase that gap. The board does not label it a net trading edge.
Team results do not explain every game.
This version does not adjust for the announced starting pitcher, confirmed lineup, injuries, park, weather, or in-game score. Its probabilities are research estimates. Historical forecast accuracy does not establish profitability.
2025 test results
2,424 eligible games, with model settings chosen before calculating this test. Lower Brier score and log loss are better.
| Forecast | Brier | Log loss |
|---|---|---|
| Experimental MLB model | 0.2452 | 0.6834 |
| Fixed Elo comparator | 0.2443 | 0.6814 |
| Training home-win rate | 0.2483 | 0.6897 |
| 50/50 forecast | 0.2500 | 0.6931 |
The fitted model improved on a constant home-win forecast, but did not beat the simpler Elo comparator. It also underestimated the aggregate home-win rate in this test. These results support continued research, not a claim of trading advantage.
Historical schedules were retrieved after the games, rather than preserved in real time. Result delays reduce timing leakage but cannot prove exact historical data availability.
Coverage and evidence
The public board compares supported MLB and NFL pregame winner contracts with independent experimental estimates. The NFL model updates team strength from verified final results using Elo, a rating system where wins and losses change each team’s rating. It does not model starting quarterbacks, injuries, weather, or live game state. Missing models stay unavailable; probabilities are never manufactured from the exchange quote.
NFL contract fair value includes the half payout for a tie: win probability plus half the tie probability. Board gaps are before fees. The forward scorecard compares forecasts saved before kickoff with actual winners and final scores. It tracks recorded, graded, pending and excluded games, plus probability error and calibration. No profitable edge has been validated.
Model evaluation uses log loss, Brier score, and calibration against verified game outcomes. Pending and excluded games do not count toward these metrics. The scorecard evaluates forecast quality; it does not calculate trading returns.
Sources: MLB schedule and results · Kalshi fee information
Referral disclosure
PlayDelta may receive a referral reward if you sign up through a Kalshi referral link. PlayDelta is not currently sponsored or endorsed by Kalshi. Eligibility, market access, and account rules are controlled by Kalshi, not PlayDelta.