Contract before average
IFAB Law 17 defines when a corner kick is awarded under football law. Betting settlement still needs an accepted operator and provider definition. Betfair's football rules and DraftKings' soccer rules provide current examples.
Record match or team corners, full match or interval, over or under direction, whole or half line, push state, extra-time treatment, provider and correction window.
Build a line probability
A mean corner count does not directly answer P(total corners over 9.5). Estimate a complete count distribution under a documented model, then sum the outcomes above the line. Validate the resulting probabilities by range using scikit-learn's calibration guidance.
| Input group | Availability check |
|---|---|
| Team attacking and defending corners | Same provider and competition coverage |
| Score-state tendencies | No future match events in features |
| Venue and pitch | Known before decision |
| Lineups and tactical roles | Timestamped source |
| Opponent interaction | Modelled rather than added as a narrative |
Worked three-leg check
Suppose three illustrative over-line probabilities are 0.54, 0.57 and 0.51. If the match events are treated as independent:
Joint probability = 0.54 * 0.57 * 0.51 = 0.1570, or about 15.70%
The arithmetic does not show that the events are independent or correctly priced. It only combines the stated probabilities.
Dependence and settlement audit
Shared weather, competition format, team overlap, tactical inputs and provider corrections can connect separate forecasts. OpenStax's independence rule must be justified before multiplying marginals.
After the match, reconcile every logged corner event with the named provider and preserve corrections. Test boundary fixtures at exactly the whole line, one below and one above, plus abandonment and awarded-result cases.
Next step
Use Corner Betting Markets for the next part of this topic.
Build line-specific QA
For a whole line such as Over 9.0, retain three states: more than nine corners wins, exactly nine pushes under the assumed rule, and fewer than nine loses. For Over 9.5, there is no integer push state. Test both products even if the forecasting model is shared.
| Fixture | Over 9.0 | Over 9.5 |
|---|---|---|
| 8 corners | Loss | Loss |
| 9 corners | Push | Loss |
| 10 corners | Win | Win |
The table is a generic line-mapping illustration; verify the accepted operator contract. Record team versus match scope because Home Over 4.5 does not settle from the match total alone.
Run a provider reconciliation on at least one corrected event and preserve both provisional and final counts. If research and settlement providers disagree, do not silently edit the model dataset to match the bet result. Document the mapping decision and exclude unresolved records from performance claims.
Continue learning
- Next guide: Correlated Parlays
- Related guide: Goalscorer Accumulators
Assumptions and limitations
The probabilities are illustrative. Corner counts can be volatile, provider events can be corrected, and a model calibrated in one competition or line range may not transfer to another.

