Before the percentage: check the evidence

01

Choose the exact question

Open Fixtures, choose the match and define the selection: result, goals, corners or another supported market. Keep its line and settlement period visible. Over 2.5 match goals and home-team Over 1.5 goals require different evidence and prices. A market for qualification may include extra time; a regulation-time 1X2 forecast does not answer that question.

The readiness indicators separate results, goals, corners, cards and players. Open “Why this status?” for the actual blockers. An available result forecast does not make a player-card market ready. Free Fixtures and Bet Audit are a useful place to begin; deeper match research and model comparisons require All Access.

02

Read the estimate’s identity

Published forecasts preserve the provider’s probabilities and model version. The xG Poisson estimate uses valid provider expected goals in a separate calculation, so it shares inputs with that provider forecast. Comparing them is useful, but agreement is not two independent votes.

Elo + Dixon–Coles reconstructs strengths from covered completed results. Current structural defaults are research estimates, require explicit opt-in and at least 20 covered results per team, and are not established as calibrated forecasts. Meeting the floor is a minimum input rule, not evidence of predictive accuracy.

The corners count curve uses provider-derived corner inputs. It is distinct from published probabilities at supported lines and is not an independently trained team-corner model. Fixed structural/market blends are comparison benchmarks with visible weights. Fitted movement or mixture output remains withheld until its actual training requirements are met.

03

Put a denominator beside every rate

A historical rate describes the covered observations in the selected window and venue. Read the eligible-match count and date range alongside it. Missing observations are excluded rather than treated as zero. Two successes in two covered games do not carry the same evidence as twenty in thirty.

Home/away filters can shrink the sample sharply. Shared fixtures, incomplete feeds, old records and mixed competitions also change what a rate means. Inspect the underlying matches before treating a headline rate as representative. A frequency interval describes sample uncertainty; it is not a calibrated confidence interval for the next match.

04

Check four different clocks

Generation time tells you when the provider made its forecast. Receipt time tells you when FootyDesk obtained it. The underlying-data cutoff, when supplied, describes which evidence could be included. Quote observation time belongs to the bookmaker price. They answer separate questions.

A forecast received this afternoon may have been generated days ago. A refresh does not make that forecast current. Current published-model eligibility applies the existing freshness rule; unsupported or stale inputs remain excluded even when the raw record is available. If a source does not supply its data cutoff, the site must say so.

05

Make unavailable inputs part of the decision

A withheld model or empty ranking is a useful result. It tells you that the required inputs did not support that comparison. Try a supported market or another fixture rather than substituting a different line, lowering the history floor unnoticed or interpreting a missing value as zero.

Save the original source, sample, estimate class and timestamps with your reasoning when saved research is available. Return after the result to review the assumptions separately from whether the selection happened to win. One correct result does not validate a model.