Settlement rules and hit definitions

Why do we need to define it first?

If a sports quantitative report does not have ex-ante settlement rules, it can easily become arbitrary interpretation afterwards. PitchCopilot will confirm the hit definition of each core configuration and secondary configuration before release, and settle according to the rules after the game. This allows public performance to be checked and also lets readers know that hit rates are not simply rewritten after the fact.

Settlement rules include result sources, time ranges, valid conditions, postponement or cancellation of events, and statistical methods at different configuration levels. If you are just starting to read, it is recommended to read /guides/reading-performance-dashboard first, and then use this article to understand the numbers of /performance.

How to determine hit

The hit is not a vague "direction is about the same". If it is reported that the core configuration corresponds to a certain match result, goal range or team direction, a pre-defined check will be used during settlement. If the event is cancelled, the data source is abnormal or the result cannot be confirmed, the platform should mark it as unsettled or excluded instead of forcibly classifying it.

PitchCopilot's manual review publishing process checks these definitions before the report goes live. Administrators need to confirm that the statements output by the model can be converted into billable items and avoid using overly vague descriptions. This is why the report distinguishes between core configurations, secondary configurations and observations: not every insight is suitable for settlement.

Samples, windows and versions

When reading the hit rate, please also look at the number of samples, time window and model version. Short-term hit rates are easily affected by extreme results; long-term samples are more reflective of calibration quality. If there are major changes to the version, the old samples and the new samples should be interpreted separately. Further reading is available at /guides/model-version-transparency.

Brier points and field goal percentage complement each other. The hit rate answers "whether the definition is met", and the Brier score answers "whether the probability confidence is appropriate." If the model has many low-confidence hits but performs poorly in the high-confidence interval, it means that the model ranking may need to be adjusted. Please refer to /guides/brier-calibration for this part.

How to settle virtual points

Virtual point studies convert different configurations into standardized units, making it easier to compare long-term performance. It does not represent real assets and does not constitute any promise of income. During settlement, the system will update the virtual fund pool record according to the allocation level, weight and prior rules, and present the results in the public performance.

The value of this approach is to allow readers to see whether the model can maintain consistency between risk and confidence. If the core configuration fluctuates too much for a long time, or the second-choice configuration is more stable, the team needs to return to the methodology to check the weight and release threshold.

How to deal with gray situations

Gray situations often occur in sports events, such as game postponement, early suspension, post-game official corrections, time differences in data sources, or major lineup changes after the report is released. The principle of PitchCopilot is to deal with the rules beforehand, rather than changing due to favorable or unfavorable results. If the rule cannot cover it, it should be marked and explained to avoid shoehorning uncertain samples into the hit rate.

Manual review also checks whether statements are settleable before publishing. Narratives such as "could play very open" can serve as context, but to move into performance, they must be translated into clear goalscoring areas or configuration definitions. This keeps the report both readable and inspectable.

Readers should also be aware of sample exclusions when seeing performance figures. Exclusions do not necessarily mean hidden results, sometimes because event conditions no longer meet the assumptions at the time of publication. Trusted platforms should clearly state the reasons for exclusions and give hits, misses, and non-settlement their respective places. This kind of transparency is more important than simply pursuing a beautiful hit rate.

Additional reading reminder

Settlement rules also allow teams to learn. If a certain type of configuration often falls into gray situations, it means that the release statement may be too vague; if there are too many excluded samples of a certain type, it means that the data pipeline or event selection needs to be improved. Settlement is not the end, but the starting point for the next method revision.

Last check

The best habit for readers is to jot down your own hit definition before the game. Compare the public performance after the game, and re-explain only when the result can be avoided. This is also where quantitative research is more checkable than subjective reviews.

FAQ

Will observations be included in the hit rate?

Usually not. Observation items are research reminders and do not necessarily have clear settlement conditions.

How to deal with the postponement of the event?

It should be marked as unsettled or awaiting result according to the prior rules, and should not be arbitrarily changed to hit or miss.

Why should the hit rate be divided into core and secondary?

Because the confidence levels are different, mixing them together will blur the quality of the model.

This translation is for convenience only; the Traditional Chinese version prevails.