The veto changes the matchup, not the laws of value
Map selection turns an overall team opinion into a set of particular matchups. That can materially change a forecast, but it does not establish that bookmakers ignore map pools or that every underdog pick offers value. A map-win record describes past opponents and conditions. A future probability needs the actual roster, map version, opponent and possible starting side. Begin with the tournament procedure, then build possible legal veto paths. Treat an unannounced choice as uncertain rather than selecting the path that makes your preferred bet look best.
An abstract seven-map procedure
Here is a hypothetical Bo3 procedure using mapsA throughG. The teams alternate an initial ban, each picks one map, then each bans another; the remaining map is the decider. This is a chosen example, not a claim that every event uses this order. Confirm who acts first, how side selection works and what happens if a map is unavailable. In the example, A bans mapA and B bans mapB; they pickC andD, removeE andF and leaveG. Count the remaining maps after each action to catch an impossible tree.
Hypothetical pool={A,B,C,D,E,F,G}
TeamA bans mapA: remaining6
TeamB bans mapB: remaining5
TeamA picks mapC: unassigned4
TeamB picks mapD: unassigned3
TeamA bans mapE: unassigned2
TeamB bans mapF: unassigned1
MapG is decider
Played order=[C,D,G], with G conditional on1-1
Side-selection procedure must be specified separately.A frequent ban is a tendency, not a permanent law. Report its numerator and denominator within the current roster and rule context. Zero picks does not necessarily mean a team refuses the map: the opponent may have removed it first. Nor does a surprising ban reveal private scrim results with certainty. It may reflect preparation, side choice, an opponent-specific plan or sampling noise. Record the action and possible explanations instead of assigning an automatic three-to-five-point adjustment to the opponent.
Weight paths before the maps are known
Suppose two legal veto scenarios have hypothetical weights 0.60 and 0.40. For TeamA, scenarioX has map probabilities 0.65,0.40,0.55; scenarioY has 0.55,0.48,0.50. Assume independent outcomes within each scenario. Calculate the series distribution in each branch, then average those distributions. Averaging map probabilities first can change the result because the series formula is nonlinear. The weights need evidence in a real application; they are chosen teaching values here. After the actual veto is confirmed, condition on that path rather than continuing to use the pre-veto mixture.
Hypothetical independent-map Bo3 within each veto scenario
F(p1,p2,p3)=p1*p2+[p1*(1-p2)+(1-p1)*p2]*p3
F(X)=F(.65,.40,.55)=.5515
F(Y)=F(.55,.48,.50)=.515
P(Aseries)=.60*.5515+.40*.515=.5369
P(A+1.5|X)=1-.35*.60=.79
P(A+1.5|Y)=1-.45*.52=.766
P(A+1.5)=.60*.79+.40*.766=.7804
Mean-map shortcut: F(.61,.432,.53)=.5364488
Mixture and shortcut differ by .0004512.The handicap complement must respect dependence too. In scenarioX, replacing unconditional second-map loss 0.60 with a hypothetical conditional loss 0.70 after losing Map 1 changes A+1.5 from 0.79 to 1−0.35×0.70=0.755. That 3.5-point change follows from the chosen conditional input, not a measured universal correlation penalty. The decider probability has no role in Bo3+1.5 after both picks are fixed: reaching 1-1 already covers. It does matter to the series winner, so the two markets respond to different uncertainties.
Pool depth requires comparable evidence
Do not rank pool depth solely by raw map win rates. Inspect opponent strength, roster changes, version changes, sample selection and how often the map was available. Round differential adds information but is not automatically superior or calibrated as a probability. A hand-built depth index with venue weights and a fixed shrinkage constant remains an unvalidated score until tested. Multiplying every signed round margin by the opponent’s rating can even reward and penalise results in unintended ways. Define what the metric is meant to predict before choosing its transformation.
A small Bayesian example makes the assumption visible. Choose an illustrative Beta(5,5) prior for an exchangeable map-win probability. Six wins and four losses produce Beta(11,9), with mean 0.55, rather than the raw 0.60. Two wins without losses produce mean 7/12≈0.583333, not certainty. The prior is not fitted esports evidence, and exchangeability may fail across opponents or rosters. This demonstrates shrinkage, not a universal minimum sample of 10 or 15 maps. Validation must match the target matchup and information time.
A decider is simply the remaining map under the selected procedure. It need not have 28% higher variance or a higher overtime rate. A normal curve over final round margin is also only a model: legal final scores are discrete and constrained by stopping rules. Estimate score distributions with the correct regulation and overtime format. Do not assume a universal overtime starting balance or that the economy resets every round. A known map pool does not remove the need to check the rules and the cash state.
From the veto tree to a decision
Record legal branches, weights, map inputs, confirmed choices, complete odds and source timestamps. Compute EV at the accepted price after updating the information, and keep all series-related exposure visible. A change in a ban can invalidate the earlier model without implying an operator error. Do not enforce an arbitrary 55% decider-confidence or 65% pick threshold as a licence to bet. A forecast can be well structured and still too uncertain for a stake. Retain no-bet forecasts so later review can test the entire process rather than a highlight collection of correct picks.
Suppose a team selected mapC in 12 of 15 past legal opportunities. That is an 80% historical selection frequency under that sample definition. It is not automatically an 80% forecast against a new opponent. Check which maps were available in those 15 cases, whether the opponent’s ban removed alternatives and whether the roster was the same. Count an unavailable map separately from a declined pick. If a new rule changes the order of choices, earlier frequencies may no longer correspond to the same decision. Pool depth is partly about alternatives that were available, not merely the maps eventually played.
Keep two distinct forecasts when testing the process: a pre-veto mixture and a post-veto conditional forecast. Score each using information available at its own cutoff. Replacing predicted maps with confirmed ones in the earlier row creates look-ahead bias. Retain the original scenario weights even when the realised path was unlikely. Evaluate whether repeated forecasts allocate sensible mass to outcomes, not only whether the most likely decider was guessed. A method that predicts the decider well can still estimate team strength poorly, so assess veto prediction and match calibration separately.
Finally, check how the offered market numbers maps. A team’s own pick need not be Map 1, and the label “Map 3” is conditional on that map being required. A market tied to the first map played is not the same as a market tied to a particular named map. Save the contract wording with the veto tree. If a match is postponed or replayed, reassess which version of the tree the ticket refers to instead of assuming the old probability survives unchanged.
A scenario log should identify its source, observation time, roster, legal alternatives and reason for its weight. If the weight is only an analyst’s assumption, label it that way. Comparing several plausible weight sets is more informative than giving a single decider prediction a false air of precision. Keep the accepted price alongside the final model version so later review does not combine a favourable old quote with information learned only after the veto.
