Criteria guide
Odds Analysis Sites: How Should You Choose One?
This page is a list of criteria, not a list of sites. Let us say why up front: we are one of the parties doing this work, so if we published a list ranking our competitors we could not be a neutral source. Instead, below are ten criteria that help you judge for yourself whether an odds analysis tool is reliable.
Under each criterion we have also described how we do it ourselves — so you can judge us by the same criteria.
Is the sample size shown next to the percentage?
Saying a price "hit 100%" means nothing on its own. If 3 out of 3 matches and 300 out of 300 matches are described with the same sentence, the reader is being misled.
How to check · Look at any percentage and search for "out of how many matches" next to it. If it is missing, that percentage cannot be verified.
On our site · The sample is written next to every percentage, and if the sample is below three matches the percentage is not shown at all.
Example: the Over 2.5 market page →Are predictions settled automatically, or can they be changed afterwards?
If results are entered by hand, a losing prediction can be quietly deleted or corrected. In that case the published success rate is not a measurement but a selection.
How to check · Look at the site's past predictions: are the losing ones still there? Is it written how settlement is done?
On our site · All seven systems are settled automatically from the real score of the match. A settled prediction cannot be changed a second time — this constraint lives at database level.
Our prediction performance — losses included →Are the data source and date range disclosed?
A sentence like "15 years of data" is just a claim if it cannot be verified. The first and last day of the archive should be stated.
How to check · Look for three numbers on the site: how many matches, from which date to which date, how many leagues.
On our site · The archive figures are read live on every page: 168,344 matches, 970 leagues. The numbers are not hard-coded.
Our data foundation →Are opening odds or closing odds used — and is it stated?
The two measure different things. Opening odds are the first price; closing odds are the final price before betting closes. If it is not stated which one is used, the result cannot be interpreted.
How to check · Search the page for the words "opening" or "closing". If neither appears, there is a problem.
On our site · The two phases are kept on separate pages and never mixed; there are 118,405 odds sets with opening odds.
The closing side has its own page →How is missing data handled?
Counting a match without a score as "did not happen" systematically lowers the percentage; counting it as "happened" raises it. Both distort the measurement.
How to check · Check whether the site puts missing data into the denominator. If it is not written down, it has probably not been considered.
On our site · A match without a result is not included in the denominator and is counted separately. A postponed match is not counted as "lost".
Is the base rate given?
"Over 0.5 happened 93% of the time" looks impressive, but unless you are told that this outcome happens in almost every match anyway, it is decoration rather than information. The real information is the deviation from the base rate.
How to check · Is there a comparison next to a percentage showing "how often this outcome happens in general"?
On our site · Every row carries its own base rate and its deviation from that base; ranking is done by confidence-adjusted deviation, not by raw percentage.
Filters — base rate and difference columns →Is the track record public, or behind a membership?
A claim of success only means something if it can be verified. A success rate behind a membership cannot be checked without buying.
How to check · Can you see past results without logging in?
On our site · The count of settled predictions does not require login; losing ones are on the list too.
Performance page — no login required →Is "banker", "guaranteed" or "sure win" language used?
No data source can guarantee the outcome of a match. A site that uses this language is telling you what it wants to sell, not what its data says.
How to check · Search the home page and the prediction pages for these words.
On our site · These expressions are explicitly banned in our editorial principles and are checked automatically in the content production pipeline.
Our editorial principles →Is the method written down?
If it is not written how a number is calculated, that number cannot be discussed. Hiding the method usually shows that the method is weak.
How to check · Look for a heading such as "how it is calculated", "method" or "methodology"; is the formula given?
On our site · Every analysis page has a "what this number says and does not say" section, and the formula is written out clearly.
Opening Odds Analysis — method section →Is the coverage really broad, or only popular leagues?
An archive that covers only the five big leagues is useless for the rest of the bulletin. Coverage should be given as numbers.
How to check · How many leagues, how many matches? If there are no figures, there is no coverage.
On our site · 168,344 matches and 970 leagues, including lower divisions and women's leagues.
Coverage figures →
Short summary
If you only have time to check one thing, look at the sample size. A percentage that does not state how many matches it was calculated from cannot be verified, and a number that cannot be verified is not analysis. Second comes results being settled automatically and irreversibly: a source that also shows its losing predictions is always more reliable than one that does not.
Frequently asked questions
Why is there no list of sites here?
Because we are one of the parties doing this work ourselves; a list ranking our own competitors cannot be neutral. Instead we give you the criteria — you can build the list yourself and judge us by the same criteria.
Which of these criteria matters most?
The sample size. A percentage that does not say how many matches it was calculated from cannot be verified; everything else comes after that.
Are free odds analysis sites reliable?
There is no direct link between price and reliability. What decides it are the criteria above: is the sample shown, are results settled automatically, is the method written down? If a paid site does not do these things, what you are paying for is a claim, not data.
Does odds analysis predict the result of a match?
No. Odds analysis counts what happened in the past at similar odds; it does not tell you the result of the next match. Be sceptical of any source that claims it does.