PlayDelta is an independent research tool, not a sportsbook or exchange. It does not accept wagers, broker trades, provide betting or financial advice, guarantee outcomes, or determine your eligibility for third-party markets.
Third-party market availability may vary by jurisdiction; verify eligibility directly with the exchange. Data may be delayed, incomplete, or unavailable.
MLB + NFL · Experimental models
A record we can measure.
Forecasts saved before the game. Results added after it finishes. A transparent look at how the probabilities hold up.
One forecast per game, saved before kickoff. Historical backtests are kept separate.
Recorded—Pregame forecasts saved
Graded—Verified final outcomes
Pending—Awaiting a final result
Excluded—Voided or invalid games
Probability quality
How forecasts compare with results
Awaiting results
Home-win probability error (Brier score)—
Measures how far the saved home-win chance was from the result. Lower is better; a constant 50% forecast scores 0.2500. A tie counts as no home win.
Home-win log loss—
Penalizes confident wrong forecasts more heavily. Lower is better; a constant 50% forecast scores 0.6931. A tie counts as no home win.
Metrics use verified final outcomes only. Pending and excluded games do not count toward accuracy.
Home-team probability calibration
Final games only
Across many games, teams given a 60% home-win chance should win about 60% of the time. Small samples can swing widely.
NFL predicted home-team probabilities compared with actual home wins
Forecast range
Games
Average forecast
Actual home wins
Home, away and tie probabilities are saved once before kickoff. Actual winners and final scores are added when verified; pending games remain ungraded. Times are shown in your local time zone.
Waiting for the first saved NFL forecast.
Eligible upcoming games will appear after a successful scheduled model update.
NFL forecasts saved before kickoff and actual final outcomes
Game / saved forecast
Home chance
Away chance
Tie chance
Actual outcome
These results measure forecast quality. A small sample cannot establish a dependable advantage.
MLB · System health
Checking updates…
Loading the latest model and tracking status.
Checked
—
Last successful update
—
Model data as of
—
Model refresh due
—
Forward tracking
MLB live scorecard
Loading tracking period…
This record starts when live tracking is enabled. It does not fold historical backtests into live results.
Recorded—Pregame forecasts saved
Pending—Awaiting a final result
Graded—Confirmed final games
Excluded—Voided or invalid games
Probability quality
Do the numbers hold up?
Awaiting results
If the model gives a team a 60% chance, teams with similar forecasts should win about 60% of the time over many games. Small samples can swing widely.
No graded results yet.
Saved forecasts stay pending until an official final score is available. Accuracy and calibration appear here as those games finish.
Brier score—
Average squared probability error. 50/50 benchmark: 0.2500. Lower is better.
Log loss—
Penalizes confident misses more. Lower is better.
Favorite win rate—
How often the higher-probability side won. Not a measure of profit.
Home-team probability calibration
Final games only
Predicted home-team probabilities compared with actual home wins
Forecast range
Games
Average forecast
Actual home wins
Saved before the outcome
Forecast history
Latest 50 forecasts
One locked forecast per game and model version, saved 5 minutes to 24 hours before first pitch. Times are shown in your local time zone.
Waiting for the first saved forecast.
Eligible upcoming games will appear after a successful scheduled model update.
Partner with PlayDelta