Probability vs Confidence

A NextGoalAI card can show two percentages, but they should not be read as two independent probabilities of the same event. Probability describes the selected next-goal outcome. Confidence is a separate assessment of support for the prediction. The difference matters when comparing published selections.

The two questions

• Probability: How likely does the model estimate that the selected HOME, AWAY or NO GOAL outcome is?
• Confidence: How strongly does the system support that selection, given the available evidence and its confidence assessment?

Confidence is not a second prediction of the match result. For example, 80% Confidence is not a statement that exactly eight out of ten such picks will succeed. Whether either indicator is well calibrated is a separate empirical question that requires historical evaluation.

Same estimated probability, different support

Two hypothetical illustrations — not real published cards

• Card A: HOME; Probability 67%; Confidence 74%.
• Card B: AWAY; Probability 67%; Confidence 82%.
• Both cards express the same estimated chance for their selected outcomes, but the confidence assessments differ.

These values are hypothetical, not real published cards or a statement about current production thresholds. 74% Confidence does not mean a 74% guaranteed success rate. You cannot conclude from these two scores that Card B will win, that either card offers betting value, or that the two matches are otherwise comparable.

A useful reading order

1. Identify the selection — Start with HOME, AWAY or NO GOAL: what exactly is being forecast?
2. Check the prediction time — Read the match minute and score that provide the context for the published selection.
3. Interpret Probability — Treat the percentage as the model’s outcome estimate, not a promise.
4. Interpret Confidence separately — Treat the score as a separate strength-of-support indicator; avoid treating it as an extra probability to multiply or add.
5. Verify the outcome later — Use History for the confirmed result, and compare performance across a meaningful sample rather than a single match.

What these numbers cannot tell you

• Neither score by itself proves that a model is perfectly calibrated.
• Neither score predicts the final match winner unless the selected next-goal event also determines it.
• Neither number is a bookmaker price, a financial return estimate or advice to place a wager.
• An empty Live list means no selection is currently being shown; it is not itself a NO GOAL forecast.

NextGoalAI publishes only qualifying predictions. Publication rules and confidence safeguards can change; this article describes how to read the displayed fields rather than claiming a fixed internal formula.

Related Insights

• Understanding Next Goal Probabilities
• Understanding NO GOAL Predictions
• How NextGoalAI Evaluates Prediction Accuracy

All Insights · How It Works · History · Live