Separate the questions
A projected score is a summary of a forecast, not a promise that the game will finish at those exact numbers. A win probability asks which team finishes ahead. A spread probability asks whether its margin clears a specific handicap. A total probability asks about the sum of both scores. Those are different events even when calculated from the same distribution.
Suppose a hypothetical home favorite wins 24–21. It covers a home spread of -2.5, pushes at -3, and fails to cover -3.5. The score did not change; the contract did. That is why a forecast cannot be evaluated independently of the actual selected line.
Keep pushes separate
For an integer line, some distribution mass can land exactly on the handicap or total. That push probability is distinct from win and loss probability. A sound comparison carries all three where appropriate and uses the actual payout for the selected side. Calling every non-loss a win exaggerates performance.
In a purely illustrative distribution with 50 winning outcomes, 5 pushes and 45 losses per 100 equally weighted outcomes, the unconditional win probability is 50%, not 55%. Excluding pushes creates a different denominator. A report must identify which convention it uses before comparing percentages.
Why iteration counts are not a quality certificate
Monte Carlo draws approximate a distribution conditional on the assumptions supplied. Additional draws can reduce numerical sampling variation, but they cannot repair an incorrect starting quarterback, a stale quote or a poor distributional assumption. A model can become consistently wrong at very high numerical precision.
Pick6Labs has multiple model and simulation components. Their budgets are not interchangeable: the inspected production path caps its possession-simulation diagnostic separately from other sampling. We do not claim that every request executes 10,000 full play-by-play games. The visible methods and coverage are more informative than a large marketing number.
Read a game page in this order
Check the game identity and kickoff first, then the source freshness and selected bookmaker. Compare the projected result with the exact spread and total shown. Finally inspect readiness and uncertainty. A WATCH or PASS label is meaningful: it is not automatically a recommendation just because a model favors a team.
Historical evaluation should use predictions recorded before the event, with a consistent grading rule. Live updates and new market lines answer later questions and must not silently replace the earlier prediction being scored. See the methodology page for the distinction between prepared views and frozen history.