Method
How AI match analysis works
What actually happens, not what is claimed
Four steps in order. The interesting part happens to the numbers before the model ever sees them.
- 01
It collects everything on the match
For the match Sharkline takes team and player statistics, form, head-to-head history and line-ups. And how the same match was assessed by several dozen sources.
One assessment is one company’s opinion. When there are dozens of them, you can see what the sources as a whole assess. And where the sources argue with each other, the most interesting part is hiding.
- 02
It puts the sources on one scale
Dozens of independent assessments never line up on their own. Sharkline puts them on the same scale and weighs each by how reliable it is, so the result reflects the match, not the loudest source.
The point is not to copy anyone’s number, but to reach one comparable probability the model can build on — while keeping the places where the sources disagree, because that is often where the signal hides.
- 03
It works out which outcome is likelier
On the cleaned numbers the model works out the probability of each outcome, with an eye on form, line-ups, head-to-head meetings and the competition.
It leans only on real data. A rare league, few matches, almost no data: Sharkline says so plainly rather than slotting in a plausible figure.
- 04
It shows how confident it is
Every analysis has a confidence percentage. It is higher when there is plenty of data and it agrees, and lower when there is little of it or it conflicts.
Beside it stands one reason the conclusion might not work out. And when data is really scarce, the honest answer is: skip it. That is what sets Sharkline apart from those who are paid for a confident answer every day.
What the AI does not know about a match
- No outcome is guaranteed. A well-calibrated 70% is still wrong three times out of ten — that is the point of it.
- Thin coverage means thin data. Lower divisions are covered by fewer sources, the consensus is weaker, and the confidence level reflects that.
- An analysis reflects what is known at the moment it is made. A piece of news about the match that comes out afterwards will not be in that analysis.