Freeze the half-time state
| Field group | Minimum record |
|---|---|
| Match | Competition, fixture, venue and date |
| Period | First-half end timestamp and actual elapsed minutes |
| State | Score, player count, substitutions and unresolved reviews |
| Events | Provider-defined shots, chances, corners, cards and possession |
| Personnel | Starting line-ups, changes and known availability |
| Forecast | Pre-match probability and half-time update |
| Market | Exact second-half or full-match contract, quote and acceptance |
The IFAB duration law defines periods and allowance for time lost. Use actual event timestamps rather than assuming every first half has identical exposure.
Statistics require provider definitions
A first-half shot, possession share or defensive action is meaningful only under a stable data contract. Opta's football definitions demonstrate that event and derived metrics have provider-specific definitions. Store provider, version, correction time and denominators alongside the values.
Compare update against baseline
The research question is not whether a first-half number looks high. It is whether a predeclared model that adds available first-half information improves later-match probability forecasts compared with the same pre-match model. The peer-reviewed Bayesian in-play football model is one example of conditional updating and posterior checking, not a ready-made half-time rule.
Illustrative decision table
| Observation | Valid use | Invalid shortcut |
|---|---|---|
| 0-0 score | Input to a remaining-outcome model | Assume goals are due |
| Home red card | Re-estimate with player count and timing | Apply one universal percentage |
| High shot count | Use provider-defined shot features | Treat every shot as equal quality |
| Substitution | Update only if role and availability are modelled | Infer intention as fact |
| Price change | Compare timestamped executable quote | Treat movement as proof of team strength |
Line-up and substitution state
Sportmonks' line-up documentation distinguishes predicted and confirmed line-ups and exposes formation and player-status fields for that provider. At half-time, preserve what was known then; do not use a later substitution or injury confirmation in the earlier feature set.
Handle missing and corrected fields
Define a fallback before the match for a delayed line-up, missing event feed, corrected card, or clock disagreement. A robust record may carry a missingness flag, use only the smaller verified feature set, or declare no action. Silently replacing an unavailable half-time field with its post-match value introduces future information.
Market and settlement match
Second-half result, second-half goals and full-match result are different targets. Betfair's football rules provide one current operator example of period and football settlement. Confirm whether the half market resets the score, which added time counts, and how abandonment or awarded results are treated.
Evaluation protocol
- Define one half-time cutoff and data-latency tolerance.
- Keep every eligible match, including missing-data states.
- Fit and tune only on earlier matches.
- Compare with pre-match and score-time baselines.
- Report calibration and proper scoring by competition and state.
- Join the decision-time quote and accepted receipt.
- Report no action, suspensions and unavailable prices.
Continue with Second-Half Goal Statistics for exposure-aware period analysis.
Continue learning
- Next guide: How Football Odds Move During a Match
- Related guide: In-Play Over/Under Goals
Assumptions and limitations
Half-time provides a convenient cutoff, not an information advantage. Data corrections, added time, competition differences and small state samples can distort results. A forecast improvement does not establish a favourable live price or future profit.
GambleAware
GamCare
Gambling Therapy
GAMSTOP