Estimate team strength
Separate attack and defense ratings, home advantage, rest, and time-weighted results. Optional inputs include pregame xG, special teams, lineups, and goalies.
October 9 matchups. Hard Rock Bet’s posted prices.
One selection to a 20-leg parlay. Start with a strategy.
Tap a price to add or remove a leg. One selection per market, maximum 20.
Forecast a score distribution. Derive every market from it.
A favorite is not automatically good value. This engine estimates win, loss, and push probabilities, then compares them with the price you supply.
Separate attack and defense ratings, home advantage, rest, and time-weighted results. Optional inputs include pregame xG, special teams, lineups, and goalies.
A shared game-pace factor connects both teams’ scoring. Learned late-game hazards respond to time remaining and the score.
Resolve regulation ties with an overtime/shootout winner model. Derive all market probabilities from the same final scores, including pushes.
Walk forward through historical games using only previously available data. Compare calibration and probability errors against simple and market benchmarks.
Authorized historical game data, genuinely pregame feature snapshots, and timestamped odds. A Python run creates the forecast files used here. This site displays those results and calculates price comparisons; it does not train the engine. Public-board observations are collected separately.
Simulation count is not an accuracy score. The displayed uncertainty ranges depend on model assumptions and do not guarantee outcomes. Bookmaker settlement rules must match the full-game convention.
Load the engine’s results into your workspace.
The dashboard checks the board every 60 seconds while open. An observation timestamp records when we retrieved it; the provider does not expose the age of each price. Florida account prices remain unverified.
Listed game times come from the sportsbook. This feed may omit games or markets. It is not a complete official NHL schedule.
Select these two files from the same prediction run. Owner uploads are saved to the public forecast archive with their actual receipt time.
Maximum 3 MB per file; combined upload must be under 4 MB. Synthetic and late forecasts are excluded from performance tracking.
View published model runs on any device. No sign-in required.
Requires real forecasts saved before verified game starts and final results. No verified returns yet.
Loading saved observations…
Only changes observed during successful checks can be recorded.
View public APIImported files retain their original timestamps and data-quality flags. Importing a forecast does not establish that its probabilities are accurate or profitable.
Owner administration · Public browsing does not require an account.