CS2 prediction markets: map vetoes, lineups and series
How CS2 map vetoes, best-of series and roster rules shape prediction market outcomes, with a worked best-of-3 example and resolution checks.
In this guide
Why a CS2 series is not one strength number
A CS2 series is played as best-of-one, best-of-3 or best-of-5. The winner takes the required map count, so no single team-strength number decides the match. Each map in the pool carries its own probability, and the veto decides which maps appear. A market titled after a match can resolve on the series winner, a map handicap, total maps, or a specific map. Read the rules before pricing anything: ordinary binary winning shares resolve to 1 and losing shares to 0, while rare unknown or 50-50 resolutions can pay 0.50 per share instead of refunding. Rules should name the data source, the deadline and the edge cases; the title alone is not enough.
Separate the event from the contract. An organizer publishes the format, map pool, veto order and roster rules in its rulebook. Valve tournament requirements define a roster as 5 specific athletes, separate from organization branding, and direct invites require at least 3 invited athletes to participate. Those are eligibility facts, not win probabilities. The same document does not supply one universal map veto procedure, so check the current rulebook of the event you are pricing.
What the veto removes
A veto is a pick/ban process: teams alternately ban maps they do not want and pick maps they do want until the required number remains. A best-of-3 typically leaves 3 maps, a best-of-5 5. This matters because a team can be strong on its permaban-proof map and weak elsewhere, and the veto can hide or expose that map. An overall series win rate does not tell you which maps produced it, and you cannot reconstruct the veto from the final score.
Map-specific samples are also small. A team may have played one map a handful of times on the current patch, against mixed opponents, with side choices that differ from the coming match. Treating a 4-1 map record as a stable 80 percent probability is an assumption, not a measurement. Use it as an input, label it uncertain, and compare it with the veto order the organizer publishes. The veto constrains which probabilities matter; it does not rank teams.
A worked best-of-3 example
The numbers below are an authored hypothetical illustration, not observed team data. Suppose the veto leaves 3 maps and the first team wins them with probabilities 0.7, 0.4 and 0.5. If map results are independent, the series probability is the chance of winning at least 2 of 3 maps. That is p1p2 + p1p3 + p2p3 minus twice the all-3 product, because the all-3 case is counted 3 times instead of once. Plugging in: 0.7*0.4 + 0.7*0.5 + 0.4*0.5 = 0.28 + 0.35 + 0.20 = 0.83; the triple product is 0.7*0.4*0.5 = 0.14, twice that is 0.28, so the series probability is 0.83 - 0.28 = 0.55.
Running the same arithmetic for the opponent, who wins those maps with probability 0.3, 0.6 and 0.5, gives 0.3*0.6 + 0.3*0.5 + 0.6*0.5 = 0.18 + 0.15 + 0.30 = 0.63, minus 2*0.3*0.6*0.5 = 0.18, so 0.45. The 2 outputs sum to 1.00 because they are complementary under independence. Real CS2 maps are correlated through map pool overlap, veto strategy, side selection and momentum, so 0.55 and 0.45 are model outputs, not facts about either team.
| p1 | p2 | p3 | p1p2 + p1p3 + p2p3 - 2p1p2p3 |
|---|---|---|---|
| 0.7 | 0.4 | 0.5 | 0.55 |
Turning the model into a decision
Use the model as a comparison device. Put your map probabilities in one column, compute the series number, then change one input at a time to see where the decision flips. If a market offers 0.60 for Team A to win the series and your model says 0.55, the gap is 5 percentage points before fees, spread and liquidity. That is a reason to pass or to re-examine the map inputs, not proof the market is wrong. If the market offers 0.50, your model implies a small modeled edge, and you still have to judge whether the inputs are strong enough to act on.
Losing cases matter. Suppose you back Team A at 0.50, the maps are fully independent, and your inputs are right: the outcome still lands against you about 45 times in 100. If your inputs come from a 4-1 map record, the true probability might be nearer 0.60 or 0.40, and the modeled edge disappears. Correlated maps make it worse because one side's dominant map can swing the whole series. Treat any positive expected value as conditional on assumptions you can name, not as a recommendation.
| Market price for Team A | Model series probability | Illustrative reading |
|---|---|---|
| 0.60 | 0.55 | Price above model output by 5 points; under these assumptions, negative edge for buying Team A. |
| 0.55 | 0.55 | Price equals model output; no edge under these assumptions. |
| 0.50 | 0.55 | Price below model output by 5 points; small modeled edge, conditional on input quality. |
Roster changes, sides and format
Valve requirements define a roster as 5 specific athletes and treat organization branding separately, so a market on an organization name can diverge from a market on a specific 5-player lineup. Substitutions, stand-ins, coaches filling in and mid-event changes can make old map records unrepresentative of the current team. Check who is actually playing, then check whether the market rules define the team by organization, by 5 named athletes, or by something else. A lineup change widens your uncertainty; it does not let you reuse an old probability unchanged.
Side selection and format also move the number. Counter-Strike is asymmetric: map layout and economy interact differently with the starting side, and some teams are better on one side. Best-of-one reduces the pool to one vetoed map and removes the series arithmetic entirely, so it needs a different model. Best-of-5 extends the same logic with more terms. The resolution rule is unchanged either way: get the organizer's current map pool, veto order and format from the rulebook.
Rankings are invitations, not probabilities
Valve Regional Standings (VRS) is an official performance ranking used for invitations. An invitation ranking does not itself establish a calibrated map-win probability. That distinction matters when a market price roughly matches a ranking order: agreement is not evidence that the ranking predicts map outcomes. If you want to use a rating, treat it as one input among veto structure, map pool, current lineup and side choice, and state the assumption explicitly.
A rating that updates across many matches can also lag roster changes and patches. A standing may reflect results from a previous lineup or an older map pool while the coming series is played on the current one. Before trusting a rating, ask when it was last updated relative to the roster and patch. If you cannot answer that, you are carrying hidden uncertainty into the price.
Reading resolution rules before you trade
Resolution rules should specify the data source, the deadline and how edge cases are handled, including postponements, forfeits, replacements and abandoned maps. Ordinary binary markets resolve winning shares at 1 and losing shares at 0. Rare unknown or 50-50 resolutions can pay 0.50 per share instead of refunding, so a market that looks like a simple team-vs-team bet may pay differently when the event becomes ambiguous. Verify the actual market rules for the specific contract rather than generalizing from one platform's documentation. If you want to work through the resolution mechanics in general terms, the linked guide to resolution is the place to start.
When you compare prices across venues, keep unit conventions straight with the linked odds guide, and use the linked Elo guide only as a model reference, not as a claim about CS2 accuracy. A map-level model and a series-level market are different objects, so the same 0.55 is not automatically a price. The broader esports guide covers how this cluster fits together, while every number above stays conditional on the veto, the lineup and the format you actually checked.
Sources & verification
Sources checked
Sources checked
PolyZeno. Automated review with DeepSeek V4.1 Flash.