Score the market before the selection
| Criterion | Pass question |
|---|---|
| Contract | Can the event, period, line and settlement source be written exactly? |
| Data | Does the provider cover every required event consistently? |
| Probability | Is there a documented transformation from inputs to a calibrated forecast? |
| Price | Is the accepted price retained at full precision? |
| Dependence | Are conditional links to other legs modelled? |
| Maintenance | Can rule and data changes be detected? |
Betfair's football rules and DraftKings' soccer rules illustrate how result, total, player, card and corner markets can use different periods, participation conditions and result sources. A market is not simple merely because its label is familiar.
Market decision map
| Market family | Main modelling object | Main failure check |
|---|---|---|
| 1X2 or double chance | Complete result probability vector | Draw and margin treatment |
| Goals and BTTS | Complete score or goal-event distribution | Tail coverage and dependence |
| Asian handicap | Multi-state split payoff | Push and quarter-line mapping |
| Cards and corners | Provider-defined event count | Definition and correction window |
| Player props | Participation and event model | Minutes, role and non-participant rule |
| Half or interval | Period-specific event distribution | Stoppage time and period mapping |
Provider definitions matter. Opta's definitions show that event metrics are not self-defining. Archive the provider and field version used by the model and settlement record.
Validation beats raw hit rate
Use chronological held-out data, as described by scikit-learn's TimeSeriesSplit documentation. Review reliability by probability band with calibration guidance, not only the percentage of selections that won.
A 70% hit rate can be poor if accepted prices require 80%, and a 30% hit rate can be consistent with well-calibrated 30% forecasts. Market comparisons need probability quality, net payoff and uncertainty under the same inclusion rules.
Inclusion record
For each candidate market, record a keep, research or exclude decision and one reason. Exclude when definitions drift, coverage is incomplete, prices cannot be archived, the joint event cannot be estimated, or settlement cannot be reproduced. Do not fill an accumulator with a second-choice market merely to reach a leg count.
Use OpenStax's expected-value method only after mapping complete probabilities and net payoffs.
Next step
Use Football Betting Markets for the next part of this topic.
Apply the framework to one candidate
As an illustrative decision, suppose a reader is choosing between a regulation-time 1X2 leg and a player-shot prop. The current operator rules cited on this page show why player participation and match-result settlement need different fields. If the reader's 1X2 record is complete while player minutes and prices are missing, the framework excludes the prop from that workflow.
That illustrative outcome is a data-readiness decision, not a claim that 1X2 is generally superior. Under the same operator-defined contract distinction, a reader with reliable player-event data and a validated participation model can reach a different eligibility decision.
Use a four-state output:
| Decision | Meaning |
|---|---|
| Eligible | Contract, data, model, price and settlement all pass |
| Research | A named missing field can be resolved before decision time |
| Exclude | A required field or validation is unavailable |
| Retire | A former market workflow no longer has maintainable data or rules |
Record the reason and review date so exclusions do not silently return when a market label or promotional interface changes.
Continue learning
- Next guide: Compare Accumulator Bookmakers in 2026
- Related guide: Asian Handicap Accumulators
Assumptions and limitations
The framework does not rank current operators or promise that any market is profitable. Data quality, prices, rules and model performance can change, and a strong marginal model may still produce a poor joint accumulator model.
GambleAware
GamCare
Gambling Therapy
GAMSTOP