Define what counts
IFAB Law 12 defines cautions and send-offs in the Laws of the Game, but a betting market can apply additional counting and settlement conventions. Betfair's football rules and DraftKings' soccer rules are current operator examples.
| Field | Questions to answer |
|---|---|
| Unit | Cards, card points or named-player event? |
| People | Players only, substitutes, staff or everyone? |
| Second yellow | One yellow plus red, one red, or another scheme? |
| Period | Regulation, stoppage, extra time, after final whistle? |
| Participation | Must a named player start or appear? |
| Provider | Which event feed and correction window settles? |
Never copy booking-points weights from another operator. Store the accepted point scheme with the receipt.
Data and probability workflow
Use provider-defined event data. Opta's definitions demonstrate why event fields need a named source. Freeze data before the decision and validate on later matches using a chronological split, consistent with scikit-learn's TimeSeriesSplit guidance.
Possible inputs include team style, opponent, expected possession state, referee history, player role, minutes and competition, but each needs an availability timestamp and held-out test. Do not publish a fixed referee or league threshold as universally predictive.
Accumulator record
For every leg, retain the contract and provider evidence required by the operator examples and Opta's event definitions:
- Exact line and accepted price.
- Event and qualifying period.
- Named player and participation rule where relevant.
- Provider definition and last update.
- Probability estimate and uncertainty range.
- Links to other card legs through referee, team, competition or model.
- Final event log and any correction.
Different matches can share refereeing policy, competition context and model error. OpenStax's independence rule therefore requires evidence before marginal probabilities are multiplied.
Verification fixtures
Test one ordinary yellow, second-yellow dismissal, straight red, substitute card, staff card, post-whistle card, abandoned match and provider correction against the market mapper. Every fixture should produce one documented settlement state.
Next step
Use Card Betting Football for the next part of this topic.
Model a count or event distribution
OpenStax's expected-value framework requires the probabilities and net payoffs of the relevant states. For a total-cards line, retain every count needed for win, push and loss; for a player-card event, include participation, card, no-card and product-specific void states.
| Model check | Failure it catches |
|---|---|
| Count probabilities sum to one | Truncated or duplicated states |
| Whole-line equality retained | Push removed from expected value |
| Player non-participation present | Biased sample from deleting voids |
| Second-yellow path tested | Incorrect booking-points mapping |
| Provider correction retained | Outcome overwritten after settlement |
IFAB Law 12 supplies the underlying caution and send-off framework, while the accepted operator rule supplies market counting. Keep those layers separate.
For cross-match accumulation, compare low, central and high joint probabilities. Sparse player-card forecasts can have wide uncertainty even when a point estimate appears precise.
Continue learning
- Next guide: Champions League Accumulators
- Related guide: Championship Accumulators
Assumptions and limitations
Card events are relatively sparse and sensitive to definitions, game state and officiating. Historical rates can change, and operator and provider records may differ. No cards market is described as inherently profitable.
GambleAware
GamCare
Gambling Therapy
GAMSTOP