SportSignals Prediction Methodology and Attribution

Inputs, validation and responsibility map

The documented input is a provider probability tied to a covered fixture. SportSignals validation at the product boundary checks fixture and outcome mapping, eligible price context, publication timing, missing coverage and settled-result reporting; validation of the provider's training process remains the provider's responsibility. Sportmonks documents the supplied probability types and predictability metadata, while the SportSignals methodology records the maintained integration boundary.

Sportmonks documents probability outputs, including available prediction types and predictability metadata. The maintained SportSignals methodology states how those outputs are attributed and processed in the current production path.

How a probability reaches the product

1. A covered fixture is matched through provider identifiers.
2. Available provider prediction types are read for that fixture.
3. Supported outputs are mapped to SportSignals market and outcome labels.
4. Eligible current prices are processed for the corresponding selection.
5. The prediction, comparison context and timestamps are published when requirements pass.
6. After settlement, the public results record is updated using the declared outcome rules.

Coverage can fail at several stages. A fixture may lack a provider prediction, a supported market mapping or an eligible price. Missing output should not be interpreted as a zero probability or a low-confidence prediction.

Interpreting a displayed difference

For decimal odds d, raw implied probability is 1 / d. At 2.50, raw implied probability is 0.40, or 40%. If a supplied model probability is 0.46, the arithmetic difference is six percentage points. OpenStax defines probability terminology; the example is illustrative.

That difference is not a guarantee of value or profit. Raw bookmaker probabilities can include margin, prices can change, availability and settlement matter, and the supplied probability may be wrong or miscalibrated.

How performance should be read

Current figures belong on the maintained SportSignals accuracy page, with settled sample sizes and update timing. Static resource articles do not copy a changing accuracy percentage. Probability quality also requires calibration: scikit-learn's calibration guide explains why forecasts near a stated probability should be assessed across many comparable cases.

What SportSignals does not claim here

• That it trains the current provider prediction architecture.
• That the provider uses a particular undisclosed algorithm, feature set or update schedule.
• That every fixture or market receives a prediction.
• That a probability difference guarantees a correct outcome or return.
• That historical settled results establish future performance.

Relationship to team ratings

SportSignals also exposes separate team-rating views in parts of the product. Those ratings are not evidence that they generate the provider-supplied fixture probabilities. The maintained SportSignals methodology separates the rating view from the provider probability path, while the team-rating methodology guide documents the verified read-side behavior.

Audit the provider-to-product handoff

For one published fixture, the handoff record should identify the provider fixture, provider prediction type, supplied probability, receipt time, mapped SportSignals market and outcome, eligible price record, publication time and later settlement. This makes each responsibility testable without claiming access to the provider's private training system.

Sportmonks' prediction documentation supports the provider output and predictability boundary. The SportSignals methodology supports the maintained integration and reporting boundary.

Version the public explanation

When provider, mappings or settled-result logic change materially, update the methodology as a substantive change and preserve the effective date. Do not rewrite historical attribution to make an older record look as though it used the new process.

Keep current performance on the maintained results page. Static educational copy can explain how to read the record without freezing a changing percentage into search results.

Continue learning

• Next guide: Training vs Testing Football Prediction Models
• Related guide: What Data Football AI Models Use

Assumptions and limitations

This description is current as of 14 July 2026. Provider coverage, product mappings and presentation can change. Material changes should update the public methodology and this page together.