Any app can claim “high accuracy”; counting it by rules anyone can check is harder. This page explains, one by one, the rules behind the count on the aiALYSIS Live aialysis.live/accuracy page. The in-app AI Performance screen counts by the same rules.
In one sentence
Accuracy = analyses counted as correct ÷ resolved analyses. A resolved analysis is one marked correct or wrong against the final match data. Analyses that are still open or voided do not enter the denominator.
Which analyses are counted?
Only analyses generated and published by the aiALYSIS system. Members’ own predictions in the community game are a separate pool and are never mixed in. Pre-match and live analyses are shown both together and separately.
When is an analysis correct or wrong?
Automatically, once the final match data is in. Example: a “Total goals: 2.5 OVER” analysis, and the match ends 2-1 → 3 goals → correct. Had it ended 1-0 → 1 goal → wrong. For live analyses, if the line is passed during the match (e.g. the third goal), the analysis is resolved at that moment. Nothing is resolved by hand.
What is not counted?
- Analyses that are still open (not yet resolved).
- Analyses for postponed, abandoned or cancelled matches.
- Analyses for matches whose required statistic (e.g. corners or cards) was never published.
- These are counted neither way: not correct, not wrong.
Windows and breakdowns
“Last 7 / 30 / 90 days” are measured by match date. “All-time” starts with the first resolved analysis (March 2026). The category breakdown shows only categories with at least 100 resolved analyses in the last 30 days; smaller samples would mislead. A weekly series for the last 12 weeks is published as well.
Nothing is deleted
Historical results are kept permanently. Wrong analyses are never removed, hidden or changed afterwards. You can see each day’s individual results marked “hit” or “miss” on the Today's AI match analysis page.
Why a percentage alone is not enough
- Sample size: a percentage from 20 analyses does not carry the same weight as one from 2,000; the page shows the number of analyses next to every percentage.
- Daily noise: day-to-day accuracy moves by a few points; the 30- and 90-day windows are the meaningful ones.
- Category differences: goal, corner and card analyses have different accuracy; an overall figure can hide that.
- Selection: only analyses above a confidence threshold are published; the rate is the accuracy of published analyses, not of every possible scenario.
How to check it yourself
The figures on the aialysis.live/accuracy page are recomputed every time the page is opened. Yesterday’s analyses with their individual results are on the Today's AI match analysis page, and each match page shows that match’s analyses and result. For how to read an analysis, see How to read an AI match analysis.