A plain-statistics dictionary.

Projection

A modelled estimate of a statistical total for a single match (e.g. projected throw-ins, projected corners). A projection is a number with uncertainty attached, not a prediction of a single outcome.

Probability

The model's estimated chance, expressed as a percentage, that a statistical threshold will be reached. A 62% probability means roughly 62 such matches in every 100 would clear that threshold if the model is well calibrated.

Confidence rating

Our High / Medium / Low label on every projection. It reflects historical reliability of the statistic, agreement between the 2018 and 2022 sub-samples, and how many contextual filters fired.

Expected total

The model's central estimate of a statistical total (for example, expected total throw-ins in a match). The number you would expect on average across many repetitions of the same matchup.

Statistical threshold (over / under)

A line drawn through the expected total, for example Over 33.5 throw-ins. "Over" means the model estimates the total will land above that line; "under" means below. We publish thresholds as statistical reference points, not as recommendations.

Over 33.5 throw-ins

A working example of a statistical threshold: the model's probability that combined throw-ins in a match clear 33.5. We use throw-ins because the historical sample is unusually consistent between 2018 and 2022.

Correlation

The degree to which two statistics move together. Corners and shots on target are positively correlated in our sample; throw-ins and goals are largely independent.

Independent vs correlated combinations

When two statistics are independent, combined probability is roughly the product of each. When they are correlated, multiplying them overstates or understates the true joint probability. Our combined-probability calculator flags this explicitly.

Poisson (to-score)

A discrete probability distribution we use to translate an expected goals figure into a probability that a given team scores at least one. Useful for low-count events; less useful when the underlying rate shifts dramatically.

Reliability

How well a statistic's historical signal holds up across both calibration tournaments. High reliability means the 2018 and 2022 sub-samples agree closely. Throw-ins currently lead the reliability ranking.

Consistency

A related but distinct measure: how stable a statistic is from match to match within a tournament. A small consistency gap means the statistic does not swing wildly between fixtures.

Breakeven rate

A statistics concept that converts a price into the implied probability you would need to be right on, for the call to break even over many repetitions. We surface it as a reference point for understanding implied probability, not as a recommendation.

We use statistics vocabulary throughout: probability, projection, threshold, confidence. Our terminology is statistical throughout. The model is a tool for understanding a match, not a recommendation engine.

• The sample behind the rate: Denominators, valid zeros and missing values.
• A model estimate needs a separate test: The club prototype, its minimum samples and score-grid limits.
• Threshold explorer: Observed goal, corner and card frequencies in a chosen sample.
• Club prediction method: The five steps behind the free club predictions, with a worked example.

• Archive methodology: How the archived World Cup 2026 projections were built: sample, filters, confidence and limits.
• Club ratings method: The Elo and strength formulas behind the club ratings pages.
• Archived accuracy ledger: Every stored World Cup line checked against its stored result, misses included.
• FAQ: Short answers about projections, past purchases and our record.