From context to a candidate
Start with player and game context
NBA points is our first model vertical. Player role, minutes, scoring opportunities, and game context help frame a points projection. Availability and role changes introduce uncertainty; stale or incomplete information can undermine the analysis.
Estimate a range of points outcomes
A projection is an estimate, not a promised total. Outcome modeling considers a range of possible scores rather than treating a single average as certain. Simulated scenarios depend on their inputs and assumptions; more simulations alone do not prove accuracy.
Evaluate a specific over or under
A probability estimate applies to a particular points line and direction. Over 22.5 and over 24.5 are different events. The estimate can be wrong even when the projection appears reasonable, especially when a player's role or minutes change.
Compare the estimate with an available price
Expected value depends on the odds you can actually obtain, not just a model probability. A changed line, worse price, fee, or unavailable quote can remove a calculated advantage. Comparing market prices is useful context, not proof that the model beats the market.
Separate delivery from evaluation
Model picks must come from the model pipeline before they can be published. Current customer delivery is paused. Evaluating predictive value requires eligible evidence; market-tool rows and the retained historical replay cannot fill that gap.
What the replay can tell you
The public record covers NBA points from January 1–June 13, 2026. It lets you inspect retained historical results and their disclosures. It does not establish a current win rate, calibrated probabilities, or a predictive advantage for future bets.
The replay used simulation inputs that depended on the same game's outcome. That makes it ineligible to demonstrate predictive edge. A real prediction must be evaluated using information available before the outcome is known.
Explore the historical recordAn estimate is not evidence
A model probability is an estimate of how often an outcome might occur. Observed evidence is what happened in an eligible evaluation. Those are different things: a model saying “55%” does not mean it has been shown to win 55% of the time.
Calibration asks whether outcomes occur at roughly the rates the model estimates across enough comparable predictions. We do not claim that the retained replay proves calibration or predictive edge. Inputs, assumptions, and changing player roles can all introduce error.
Expected value, with assumptions
Expected value (EV) is the probability-weighted net result of a wager. For a win-or-loss bet, it is the estimated win probability times net profit if it wins, minus the estimated loss probability times the stake. Any fees or other costs must also be accounted for.
Hypothetical example · no fees or pushes
At -110 odds, a $100 stake earns about $90.91 net profit on a win and loses $100 on a loss. If the estimated win probability is 55%:
EV = (0.55 × $90.91) − (0.45 × $100) ≈ +$5.00
The break-even probability is about 52.38%. If the actual win probability is 50%, the same bet has EV of about −$4.55. Estimation error can erase the apparent advantage.
Positive estimated EV is conditional on the probability and price being right. It cannot guarantee a winning bet, a profitable sequence, or that you should wager. Skipping a bet is always an option.
Model picks and market tools
Model picks are outputs from PropPilot's model pipeline. The first model vertical is NBA points, and current customer delivery is paused. Additional model markets and sports are future work.
Terminal market comparisons, calculators, money signals, and research pages provide context. Market-math tools calculate from supplied prices and assumptions; they are not proprietary model predictions or substitutes for paused model picks.
Removing a bookmaker's margin from odds produces a market-based estimate, not a known true probability. Agreement between books also does not establish a model's accuracy.
A displayed price is not execution
Quotes can be delayed, partial, changed, or unavailable to you. Check the current line, odds, stake limit, fees, and settlement rules at the venue before making any decision. Source timestamps describe the quote's age; they do not reserve a price.
Exchange comparison quotes do not establish that fees are included. An arbitrage calculation can also fail in practice if one side moves, is rejected, or settles under different rules. It is not risk-free. Promotional terms can change the effective payout as well.
PropPilot provides analysis. It does not execute trades, accept wagers, hold betting funds, or operate as a sportsbook.
Limits before calculations
Sports betting can result in losing all the money you allocate to it. Treat any gambling spend as an entertainment expense you can afford to lose, separate from rent, bills, savings, and debt payments. Do not borrow money or chase losses.
Set a loss limit and a time limit before starting. Stop when either is reached, take regular breaks, and do not increase a limit to recover money. A smaller stake reduces the amount at risk on one bet; it does not make an uncertain model profitable or prevent gambling harm.
No sizing formula guarantees growth
Bankroll and Kelly calculations depend on probabilities that may be wrong. A suggested stake is not a safety control or a promise of compounding returns. PropPilot does not enforce wagering, deposit, or time limits at a sportsbook.
Past performance does not guarantee future results. The retained historical replay is ineligible to establish predictive edge. If gambling affects your finances, relationships, or wellbeing, stepping away and seeking support matters more than any calculation.
Find practical limits and gambling supportRead the record in context.
Historical results describe a replay, not what a future model will earn.
