Record › Method
How are the percentages measured?
The model predicts two numbers for every match: how many goals each team will score. Every percentage on screen is derived from those two numbers, none of them is predicted separately.
How often do the percentages come true?
Picks in the record are grouped in 5-point steps by the percentage the model gave them. For each group, the table shows how often those picks actually landed. When the percentages are honest, the two numbers stay close.
Measured over the last 29 days. When the model says 80%, it actually happens 77% of the time. We're still a little optimistic in the high bands and we're correcting it.
How is the pick chosen at each level?
Each level has a target: 78% for Safe, 69% for Balanced and 42% for Bold. For every match, the market whose percentage is closest to the target becomes that level's pick, as long as it is no more than 7 points away. If two markets are equally close, the higher percentage wins. If none is close enough, that level has no pick for the match.
Which markets can be picked?
A level's pick always comes from this list:
- Home win
- Draw
- Away win
- Home or draw
- Home or away
- Draw or away
- Home · draw void
- Away · draw void
- Both score
- Not both score
- Over 0.5
- Over 1.5
- Over 2.5
- Over 3.5
- Over 4.5
- Over 5.5
- Over 6.5
- Home over 1.5
- Away over 1.5
- Home clean sheet
- Away clean sheet
When does a pick count as landed?
A pick counts as landed only when the market it named actually happened. Safe, Balanced and Bold are measured separately. Days are settled the morning after. Only finished matches count: a match that is postponed or still being played does not go into the record. A pick that is void on a draw counts neither as landed nor as missed if the match ends level.
What does the model start from?
For each match it uses the best data available, in this order:
- Standings · form · goal patterns
- Team statistics · form · goal patterns
- Recent matches · goal patterns
- League averages
The percentages are then adjusted and calibrated against past results. When only league averages are available, the match stays on the site but is left out of the record, because those picks say nothing about the two teams.
Why doesn't the record go back further?
On 22 August 2026 a calibration step was added to the model: its percentages are now matched to how often similar percentages came true in past matches. That changed the percentages at the root, so the days before belong to a different model and are not added to the record.
Where the data comes from
Fixtures, results, standings and team statistics come from a commercial football data service. The model, the picks and the record are our own. About ›
No inside information. We don't take tips from anyone. We don't cherry-pick the ones that landed: the misses are in the record too.