Compare the market contract first
| Market family | State to model | Main data requirement | Frequent contract question |
|---|---|---|---|
| Match result | Score, time, team strength, player count | Reliable clock, score and personnel | Regulation result or qualification? |
| Goal totals | Remaining goal count | Score, time and chance process | Which line and qualifying period? |
| Next goal | Identity of next scorer or no more goals | Team-specific event intensity | Is no goal an explicit outcome? |
| Half market | Events within one period | Period clock and half-specific state | Does added time count? |
| Corners | Remaining qualifying corners | Provider corner events and territory context | Awarded or taken? |
| Cards | Remaining qualifying sanctions | Recipient and sanction-level events | Which people and card types count? |
Betfair's football rules and DraftKings' soccer rules illustrate that period, result, corner, card and player-market settlement can be product specific. The comparison must use the rule attached to the actual market, not a generic label.
Use a six-part acceptance test
Score each candidate as pass, fail, or unknown:
- Contract: can every win, loss, push, void and correction state be written down?
- Observation: are score, time, player count and event states available at a known cutoff?
- Definition: do historical and live providers define the event consistently?
- Validation: can the forecasting method be tested chronologically on unseen matches?
- Execution: are accepted price, available size, suspension and delay preserved?
- Exposure: does the maximum loss fit the pre-set session limit without relying on cash out?
If any essential item is unknown, adding more live statistics does not repair the contract.
Data-rich does not mean decision-ready
Provider definitions can differ. The Opta football definitions show how event and derived metrics require explicit definitions. A model trained on one corner, shot, possession, or disciplinary-event convention should not be connected to another live feed without a mapping and reconciliation test.
Execution can reverse a paper comparison
Suppose two illustrative models are equally calibrated. Market A shows a smaller theoretical pricing difference but has timestamped size and a stable rule. Market B shows a larger difference but is suspended around material events and lacks an accepted-price record. Market B is not automatically preferable because its paper estimate is larger.
The Betfair historical feed specification provides one example of market status, publish time, available ladders and traded volume. Those records help distinguish a model quote from a price that could be executed.
Reader decision table
| Situation | Appropriate route |
|---|---|
| Need a transparent count model | In-play over/under |
| Need a team-specific next-event model | Next-goal betting |
| Need event-definition checks | In-play corners and cards |
| Material incident under review | VAR and in-play markets |
| Cannot timestamp the live state | No analysis and no action |
The Gambling Commission's in-play standard specifically warns that live information may be delayed. Latency is therefore part of market selection, not a footnote.
Continue learning
- Next guide: In-Play vs Pre-Match Betting
- Related guide: Football Momentum
Assumptions and limitations
This framework compares analytical tractability, not expected profitability. Market availability, rules, data quality, delay, limits, liquidity and settlement vary by operator and jurisdiction. A market can pass the framework and still have no favourable price or reliable forecast.
GambleAware
GamCare
Gambling Therapy
GAMSTOP