- Data source
- API-Football — fixtures, form, tables, expected goals, lineups, injuries and odds from multiple bookmakers
- Settled on
- The 90-minute score. Extra time and penalties never count
- Coverage
- Senior first-team competitions across more than 25 leagues
- The record
- Public, on our results page and accuracy tracker
The pipeline, step by step
Collect the data
For every fixture on the day we pull the numbers from API-Football: the fixture itself, recent results and form, the league table, head-to-head history, team statistics including expected goals, confirmed lineups and injury news where they are published, and odds from multiple bookmakers for each market.
Filter the competitions
A quality gate checks every competition name and every team name. Youth, reserve, development and under-age football is excluded, as are women's competitions, so the board carries senior first-team matches only. The gate runs on the website, the Telegram channels and the live centre alike.
Model each market
The model scores the markets for the fixture — goals lines, both teams to score, match result and more — from recent form, goals scored and conceded, home and away performance, table position, head-to-head results and the extra fixture load on clubs playing in Europe.
Judge each league on its own history
A historical database of more than 55,000 completed matches tells us how often each market lands in each competition. A goals call in a low-scoring league has to clear a higher bar than the same call in a league where goals flow, so a league is never read off a global average.
Calibrate the confidence
Model output is calibrated against our own settled record, league by league and market by market. The confidence rating you see reflects how often calls like it have actually landed, not how strongly the model leans.
Check the price
A probability is only half the answer. The pick has to be offered at a price above its break-even before it is worth publishing. The next section explains exactly how that is measured.
Publish, then settle
Picks go out before kick-off, on the site's daily board and on our Telegram channels, and a game leaves the board once it starts. After the final whistle the pick is settled from the official 90-minute score and joins the record.
The break-even rule
For every type of pick at every price range, our settled record shows how often it has landed. The break-even price is 100 divided by that hit rate. As an example, a type of pick that lands seven times in ten needs a price above 1.43 to be worth publishing — at 1.43 or below, being right that often still does not pay its way.
A number that only looks good in one slice of the data is not a real edge, so a market-and-price combination has to hold up under every one of these tests before it is treated as sound:
Volume
Enough settled picks in the combination to mean something. Thin samples are stories, not evidence.
Wide window
It comes out ahead across a broad price band, so a hand-picked window cannot flatter it.
Narrow window
It still comes out ahead in a tight window around the actual price on offer.
Tail
It still comes out ahead with its two biggest winners removed, so one long shot cannot carry it.
Walk-forward
It holds in both the older half and the newer half of the record, so it is not a pattern that only fits the past.
The weakest result is the verdict, never the strongest. For the Telegram channels this check runs at the moment of sending: the desk must run a full analysis of the fixture first, and a pick in a combination our record rates as a loser is stopped before it reaches a channel. A pick is refused because a number says so, never because of a hunch.
Markets we call
Over 1.5 goals, Over and Under 2.5 goals, both teams to score, match result, double chance, handicap and corners, each with its own page on the site. Correct-score calls are long shots, so the accuracy tracker keeps them out of its headline strike rate.
How a pick is settled
Every market is graded on the official full-time score: 90 minutes plus stoppage time. Extra time and penalty shoot-outs never count, so a cup tie that is 1-1 after 90 minutes settles as a 1-1 draw, whoever goes through.
Settlement reads that score straight from API-Football by fixture. A market the system cannot grade with confidence is left open for checking rather than guessed. Every pick we send to our Telegram channels appears on the results page, win or lose.
A prediction is a probability, not a promise. Even a well-built model loses calls, and bookmaker prices are a strong forecast in their own right — which is exactly why every pick is measured against them. The record shows the losses beside the wins so you can judge the method on evidence. 18+, and please play responsibly.
Common questions
Where does Pundit Kings get its data?
From API-Football: fixtures, results and form, league tables, head-to-head records, team statistics including expected goals, lineups, injury news and odds from multiple bookmakers.
Do extra time and penalties count?
No. Every pick is settled on the score after 90 minutes plus stoppage time. Extra time and penalty shoot-outs never count.
Do you publish losing picks?
Yes. Every pick we send to our Telegram channels appears on the results page at punditkings.com/results.html, win or lose.
Is a Pundit Kings prediction ever certain?
No. A prediction is a calibrated probability. Football has variance, and even the strongest calls lose sometimes.
Who writes the reasoning and the articles?
Predictions come from the statistical model. Articles, match reports and live commentary are written with AI writing tools from the same live data, under strict rules against inventing facts.
Ask the desk
Want to know how a specific pick was made, or think one was settled wrongly? Tell us the match and we will look at it.
Talk to us →