Choose the event before the price
A view about a team is not yet a view about a bet. Match winner, map winner, a map handicap and a player total pay in different states. The best contract for an opinion depends on which states you can model and which price is actually available. No market family is inherently profitable. Lower margin does not prove an edge, and a higher success rate does not prove a better return. Begin by writing the precise winning condition, then price that condition rather than the team name.
A compact map of the contracts
The match-winner contract concerns the official series result under its rules. A single-map contract concerns just the named map. In a completed ordinary Bo3, Team A −1.5 maps means A wins 2-0; A +1.5 loses only if A loses 0-2. In a Bo5, A −1.5 wins on 3-0 or 3-1, while A +1.5 wins on any series victory or a 2-3 defeat. A +2.5 wins unless A loses 0-3. These definitions assume no starting map advantage or exceptional settlement clause.
Totals require another random variable: rounds played, kills recorded or a named statistic. A half-integer line distinguishes neighbouring integer counts; an integer line may introduce a push according to the rules. For a regulation first-to-13 example, 13-7 contains 20 rounds and 13-8 contains 21. Thus 20.5 separates those scores. Check whether overtime contributes to the total and whether a prop covers one map, two specified maps or the whole series. The title of a prop is not a complete settlement specification.
Measure the quote without inventing a margin ladder
Record both sides of a complete two-outcome market at the same timestamp. Let S = 1/dA + 1/dB. Overround is S − 1, while 1 − 1/S is a theoretical hold for balanced liabilities, or an expected loss under proportional margin loading and correct underlying probabilities. It is not the realised loss on every ticket. At 1.90 on both sides, overround is 5.263158% and theoretical hold is 5%, not 4.76%. At 1.95 on both sides they are 2.564103% and 2.5%.
Dividing each reciprocal price by S gives a proportional no-vig reference. Additive removal distributes the excess equally across outcomes. Different removal methods can produce different estimates from the same quotes; none observes the true chance of winning. A reference operator is a comparison source, not an oracle. Compare several independent sources where available, and record the method. A disagreement may reflect different rules, stale timestamps, liability or information missing from your model. Do not automatically label the longest quote a trading error.
Link maps to a series conditionally
A series model makes the relationships explicit. Suppose A has hypothetical map-win probabilities p1 = 0.62, p2 = 0.45 and p3 = 0.55, with independent outcomes under these inputs. The four terminal scorelines exhaust a completed Bo3. They give a series chance of 0.5606 for A and 0.4394 for B. This is not inferred from two map prices alone: the decider input is essential. A market mismatch may reflect a different decider estimate or path dependence rather than an executable arbitrage.
Hypothetical independent-map Bo3, A perspective
P(2-0) = .62*.45 = .2790
P(2-1) = (.62*.55 + .38*.45)*.55 = .2816
P(1-2) = (.62*.55 + .38*.45)*.45 = .2304
P(0-2) = .38*.55 = .2090
sum = 1; P(A series) = .5606
P(A +1.5) = 1-.209 = .791
P(Over2.5) = .2816+.2304 = .512
EV(B series at2.45) = .4394*2.45-1 = .07653
EV(B +1.5 at1.48) = .721*1.48-1 = .06708The example also explains why probability disagreement and EV are different quantities. A model can exceed a no-vig reference yet still fall below the break-even probability of the actual ticket. Calculate p × d − 1 separately for every offered contract. Do not subtract the market overround again from that EV: the posted price already contains the margin. Subtract only additional cash costs appropriate to the transaction. Keep rounding for presentation until after the calculation, especially when the estimated advantage is small.
Capacity and uncertainty constrain the choice
A hypothetical 2% edge on an accepted stake of 1000 has expected profit 20; a 6% edge on 100 has expected profit 6. These are conditional amounts, not earnings forecasts. The larger number does not automatically win the comparison: probability uncertainty, correlated losses, costs and available capital still matter. Record actual accepted capacity rather than publishing universal limits for a market family. No reliable number of bets before an account restriction can be inferred from this arithmetic, and camouflage betting does not improve the underlying contract.
For count props, test the distribution rather than declaring Poisson or negative binomial universally correct. A Poisson model sets variance equal to mean. A negative-binomial model permits overdispersion, but its extra parameter still needs data and validation. Random game length can mix several count distributions. An apparently precise tail probability can be driven mainly by an uncertain number of rounds or maps. Compare calibration on held-out observations and retain an explicit scenario for incomplete participation, replayed maps and substitutions under the actual rules.
Live contracts add an information timestamp and an acceptance process. A pause or map break does not prove your screen is current or the bookmaker has stopped receiving information. Outrights add a tournament path, a larger outcome set and a longer holding period. Neither feature sets a universal margin or makes the product automatically negative EV. Use observed quotes and the event rulebook. If you cannot establish the contract or align the timestamps, mark the candidate unevaluable rather than inserting a guessed latency or a supposed standard fee.
Finish the comparison with a decision log: contract, source time, full quotes, modelling assumptions, probability range, net EV, accepted capacity and all related open positions. Use the same information cutoff for model and benchmark. Review forecasts that produced no bet as well as settled tickets, or selection bias will hide failures. A closing-price comparison is evidence about price, not proof of true EV. The next action is either a sized position within a prewritten loss budget or a recorded pass. Market selection is complete even when every candidate is rejected.
For an uncertainty exercise, let a quote be 1.90 and let the plausible probability range be 0.45 to 0.60. Conditional EV then ranges from −14.5% to +14%, while break-even is 52.631579%. A central estimate of 0.55 would display +4.5%, but that does not make the whole range positive. An arbitrary minimum EV rule cannot repair an unsupported range. Investigate which assumption moves the forecast most, and decide whether the available evidence can narrow it before committing money. If not, record a pass rather than a spurious precise stake. This exercise can change the preferred product: a particular map may be better understood than the full series, or the reverse. State what information would reverse the decision and retain the quote time, so a later outcome cannot rewrite what was knowable before entry.
