POINT PROPHET / OUR APPROACH
Evidence for your own decisions
Follow what players are doing, examine their projected finishes, and compare the evidence with the line you are considering.
Know which kind of evidence you are reading
Observed: Scores, player stats, plays, and sportsbook quotes reported by a source. A page refresh time is not necessarily the time a quote changed.
Historical: What happened in the available sample. A 5/5 record describes five past results; it does not establish certainty about the next game. Check participation notes, sample size, and whether the player's role has changed.
Projected: A model's estimate of a future total. STRONG, LEAN, and SKIP are selection signals based on the model's rules. They are not calibrated probabilities.
Validated: Performance measured against outcomes the model did not train on. Average error describes a group of forecasts; interval coverage measures how often a stated range contained the outcome. Neither is automatically an over/under win rate.
NFL training and test results
The NFL Game Center publishes its own model report: the training seasons, later evaluation seasons, live inputs, and measured error for each supported player stat. This is separate from the NBA model. Sportsbook odds are not required to produce NFL projected finishes.
See how the NFL model is trained and testedThe standard we are building toward
Start with historical expectations, update them as relevant game evidence arrives, and keep uncertainty visible. Pace, playing time, opportunities, lineup changes, and game state should influence a forecast only through methods we can test.
Calling a method Bayesian does not validate it. Each sport and prediction target needs its own evaluation on later games, with only the information available at prediction time. NFL historical results and live validation are tracked separately.
This is the shared development standard. The current pages contain different levels of evidence; a unified, calibrated probability engine is still work to complete.
From one selection to several
The cheat sheet's builder is a way to explore historical patterns. Its ordering does not establish the best expected return or the most likely winning ticket. Legs in the same game may depend on the same pace, playing time, and scoring opportunities.
A defensible ticket probability requires validated individual probabilities and a model of how the legs relate. Until those are available, the builder displays the underlying evidence without a combined win percentage.
Missing evidence stays visible
An unavailable feed, incomplete sample, or unvalidated forecast should remain identifiable. Check the source and update information, and compare the displayed quote with the sportsbook's current offer when making your selection.