How DNP and Void Rules Change What a Probability Means

Ask most people what a 62 percent probability on a rebound line means and you will get a reasonable answer: out of a hundred versions of tonight, the player clears the number in about 62 of them. Then ask what happens in the versions where she never leaves the bench. The answer decides whether that 62 percent was ever the right number, and almost nobody checks it first.

The mechanism is unglamorous. Many player prop products remove a pick from settlement when the participant does not appear, returning it rather than grading it. That single rule reshapes the estimation problem. The event being priced is no longer whether the player clears the line. It is whether the player clears the line given that the pick settles at all.

Two different questions wearing the same number

Write the two quantities side by side and the difference stops being philosophical. The unconditional probability asks how often the player clears the line across all futures, including the ones where an illness, a coaching decision, or a late scratch keeps him out entirely. The conditional probability asks how often he clears it among the futures where he actually plays. That is conditional probability in the textbook sense: a probability computed inside a restricted set of outcomes.

When a platform voids on non appearance, the conditional version is the one that matches the product. The futures where he sits are not losses for the over; they are futures that never existed as far as settlement is concerned. A model that folds those futures into a single number is deflating every over on the board by the probability of an absence that would have been refunded anyway.

Twelve points is not a rounding difference. It is larger than nearly any real edge anybody finds in a player prop market, which means a research process that gets the conditioning wrong is generating noise bigger than the signal it is hunting.

What this changes about your research

The first consequence is that availability research stops being a filter and becomes part of the definition of the number. Under a void rule you are no longer asking whether a player is likely to play; you are asking, if he plays, what his workload looks like. Those are different investigations. The second is about the shape of the distribution rather than its center.

• Full absence is a void, so it should be removed from the outcome set rather than counted as a zero. Counting it as a zero is the classic error, and it drags every projection down uniformly.
• Partial appearance is not a void anywhere. A player who plays four minutes before leaving settles at whatever he produced in those four minutes. That is the genuinely dangerous branch, because it is a real loss on an over and it lives inside the conditional set.
• Historical averages built from box scores usually already exclude absences, since a game he missed produces no row. Averages built from a season total divided by scheduled games do not, and mixing the two silently blends conditional and unconditional numbers.
• Grading records inherit the same rule. If voided picks are counted as losses in a published hit rate, that record is measuring availability forecasting rather than projection quality.

The partial appearance branch deserves the extra attention it rarely gets. Casual research treats availability as binary because news feeds do, but a return from a layoff, a minutes restriction, or an early exit all produce settled results well below a normal projection. Conditioning on appearance does not mean assuming a full workload, and a model that conditions on appearance while still projecting a healthy usage share has traded one error for a subtler one.

The shape of the problem changes by sport

Non appearance is a single idea with wildly different anatomy depending on what you are researching. The useful move is to identify, per sport, exactly which real world events fall on the void side of the line and which settle.

Soccer: appearance and substitution

Soccer has the most consequential version of this, because the lineup is genuinely unknown until roughly an hour before kickoff and because a bench player who enters in the eighty fifth minute has appeared. That player's pick settles on five minutes of football. Meanwhile the same player, unused, may void. So the distribution is honestly bimodal for anyone outside the certain starters, and the single most valuable piece of research in the sport is the posted lineup. Slateline's soccer engine models minutes before it models anything else, and its probabilities are appearance conditional by construction; the soccer guide walks through why minutes come first there.

Basketball: rest, load, and rotation

In WNBA research the void case is usually a clean did not play, and the interesting risk is the one that settles: a rotation shortened by foul trouble, a blowout that pulls starters, an unannounced minutes limit. Availability news matters enormously, but it is the conditional workload that decides the pick once the player is active. Our WNBA probabilities are conditional on appearing, which is why an availability caveat and a projection can appear on the same card without contradicting each other.

Tennis: retirement mid match

Tennis introduces a version that occurs after play has begun. A walkover before the first serve is straightforward. A retirement in the second set is not: the match started, statistics accrued, and whether the offer voids or settles on the partial match is a rules decision that varies. Because volume markets such as total games depend on a match reaching a certain length, our tennis probabilities are completion conditional, and the possibility of an unfinished match is treated as a structural feature rather than an afterthought. The tennis guide goes further into how format and completion interact.

MMA: cancellation versus an early finish

Combat sports separate the two cases more cleanly than any other sport, and getting them backwards is a common error. A bout that is cancelled, or a fighter who withdraws, is the void event. A bout that ends in ninety seconds is not: the fight happened, and volume props such as significant strikes settle on whatever was recorded before the finish. That is why duration modeling has to come before volume modeling in MMA. An under on a strike total is substantially a wager on the fight being short, whether or not anyone frames it that way. The fight duration article is the long version of that argument.

Esports: substitutions, remakes, and scope

Esports adds a wrinkle the traditional sports do not have: an offer can span several maps, so a player can appear for part of the priced scope and not the rest. A roster substitution between maps, a remade game, or a series that ends before the third map is played all interact with how the offer was scoped in the first place. An offer covering the first three maps of a series is not voided when the series ends in two; it settles on what was played. Scope is doing as much work as availability here, which is the subject of the map scope article.

How we handle it, and where that leaves you

Slateline publishes appearance conditional or completion conditional probabilities deliberately, per sport, because those are the events the products actually settle. Our grading follows the same rule: a leg where the player never appeared is voided out of the record rather than counted as a miss, so the published hit rates measure projection quality rather than availability guessing. Where a sport's void semantics are genuinely uncertain, the honest response is to skip the leg in grading rather than to guess a convention, and that is what we do.

This is not a claim of superiority so much as a claim of consistency. A conditional probability is only correct if everything downstream, the grade, the calibration curve, the record, is computed against the same condition. Mixing them, publishing conditional probabilities while grading them unconditionally, would produce a system that looks systematically overconfident for reasons that have nothing to do with the model. You can inspect the resulting per sport records in the Model Room.

For your own process, the practical version is short. Before trusting any probability, ask what the platform does when the participant does not appear, and ask whether the number in front of you was computed under that same assumption. If the two do not match, adjust it or discard it. Availability rules are also a reminder of how much of this sits outside anyone's control; research narrows uncertainty and never removes it. Keep the activity recreational, set limits ahead of time, and if it stops being recreational, help is available at 1 800 GAMBLER and through the National Council on Problem Gambling. When you want to see conditional probabilities attached to live offers, the Signal Board is where they land.