How OddUni predictions are made

A prediction on OddUni is not a tip someone typed. It is the output of a model reading data about a fixture, published on a schedule, and then graded against what actually happened. This explains each of those three stages, and what each one does and does not claim.

What goes in

For a given fixture the model reads what is available about it:

• Market prices from licensed bookmakers, used as reference data. A price is a crowd's estimate of likelihood, and it is usually the single most informative number about a match.
• Recent form for both sides — results, goals scored and conceded, and how recent those games are.
• Home and away splits, which are frequently a different story from overall form.
• Head-to-head history, where enough meetings exist for it to mean anything.
• League standings, where the competition has a table.

Coverage varies by competition, and that matters more than it sounds. A top European league has all of the above. A lower division may have almost none of it. Where a figure is missing, OddUni says it is missing rather than substituting an estimate — which is why some fixture pages carry a full read and others say plainly that there is not enough to go on.

The moment a prediction is fixed

This is the part worth understanding, because it is what makes everything downstream honest.

A day's predictions are written once, at publication, and read back unchanged from then on. They are never regenerated.

If a prediction were recomputed when you opened the page after kick-off, it would quietly become a better prediction than the one actually made — and every accuracy figure built on it would be fiction. The price shown against a selection is the price at publication, not one fetched later. A game that has already started cannot improve its own forecast.

How it is scored

Every published prediction is settled against the final score and counted. Wins and losses alike.

Nothing is dropped for being embarrassing, and there is no "void" bucket used to hide a miss — voids exist only where a fixture was abandoned or a market genuinely could not be resolved. The running total, with the sample size behind every figure, is published on the results page.

When a data provider corrects a box score after the fact — which happens more often than most people expect — the affected predictions are re-graded and the record moves with them. The record is meant to be right, not stable.

What a prediction is not

It is not a guarantee, and OddUni does not publish the language of guarantees: no locks, no sure bets, no "cannot lose". A model that was never wrong would be a model that had memorised the past.

It is also not advice. OddUni is not a sportsbook — it accepts no wagers, holds no funds and settles no bets. What you do with an analysis is your decision. If that decision is becoming a problem, please read responsible gambling.

Where to look next

• What a confidence rating actually tells you
• Why a 70% prediction is wrong three times in ten
• The full methodology, and the editorial policy covering how everything here is produced and corrected.