How our football predictions work

1. Where the inputs come from

For each upcoming match we collect several independent forecasts. Each forecast proposes picks on the markets below with a confidence score from 0 to 100, and may quote a price. Fixtures, live scores, final results and bookmaker odds come from established sports-data providers. Forecasts that fail or that cover markets we don't publish (correct score, for example) are dropped.

2. Which markets we cover

Match result (1X2), double chance, both teams to score, over/under goals at the 1.5, 2.5, 3.5 lines and Asian handicap. Each over/under line and each handicap line is a market of its own, and a market gets at most one pick per match.

3. How a pick is selected

• Best selection per market. For every market we take each possible selection (e.g. Over 2.5 or Under 2.5) and give it the highest confidence any forecast gives it. The selection with the highest confidence wins; a tie goes to the selection more forecasts agree on, then to the higher price.
• Confidence threshold. Nothing is published below 60% confidence. Picks at 65% or more carry the “High confidence” badge and are the only ones eligible for Top predictions.
• Contradiction guard. If two opposite selections in the same market both reach 60%, the stronger one is published only when it leads by at least 10 points; otherwise that market gets no pick. The same rule applies across markets when a match-result pick contradicts the double-chance pick (e.g. “away win” vs “home or draw”): within 10 points both are dropped, otherwise the weaker one goes.
• Odds limits. The price is the best one we find for the selection among the bookmakers in our odds comparison; if none of them quotes it, the forecast's own price is used. Picks priced below 1.30 or above 5.00 are not published: very short prices carry too little return, very long ones too much noise.

Until kick-off a pick can be revised (new price, new confidence) or withdrawn when it no longer passes these rules. A withdrawn pick never had a published result and is not counted. At kick-off the current version is frozen, and that is the version that gets settled.

4. How picks are settled

About two hours after kick-off, once our live-score data marks the match as finished, every frozen pick is settled from the final score:

• Won: profit = published odds − 1 unit. Lost: −1 unit.
• Push (void): the stake is returned, 0 units, e.g. over/under 2 goals with exactly 2 goals, or a handicap that ends level.
• Quarter lines (e.g. −0.75) are settled as two half-stakes on the neighbouring lines, so a pick can be half won or half lost.

5. How hit rate and ROI are calculated

Every pick is a flat 1-unit stake at the price frozen at kick-off. No staking plan, no doubling up.

• Units = the sum of every settled pick's profit or loss.
• Hit rate = wins ÷ (wins + losses). Half-won picks count as wins, half-lost ones as losses, pushes are left out.
• ROI = units ÷ number of settled picks × 100 %. A pick that is not settled yet counts in neither.

The track record shows these figures overall, per market (over/under split by line) and per league. A league gets its own row from 10 settled picks; smaller samples are grouped as “Other leagues”. Each closed month with at least 30 settled picks gets its own monthly report; a pick belongs to the month its match kicked off in. Losing picks are never deleted.

6. What the confidence number means, and its limits

The confidence (e.g. 68%) is the strongest score the forecasts give that selection. It ranks picks against each other and decides what clears the 60% bar. It is not a calibrated probability: a 68% pick does not win 68% of the time, and we don't claim it does. Our own track record by market is the honest measure of how good the picks have been.

Compare it with the odds: a price of 2.00 means the bookmaker rates the outcome at about 50% (1 ÷ 2.00), before its margin. When our confidence sits well above that implied figure, the pick note says so, but that is a signal to check, not proof of value. Past results don't predict future ones, and short samples (a week, one league) swing a lot.

7. The correct-score estimates

The correct-score page uses a separate, simple model and is not part of the track record. Each team's expected goals come from this season's league results: home attack against away defence for the home side, and the mirror image for the away side, scaled to the league's average goals at home and away. Small samples are pulled towards the league average (4 pseudo-games), so three early games can't produce an extreme figure. Each side's goals are then treated as independent Poisson variables, which gives a probability for every scoreline. Known limits: this approach slightly under-rates draws (0–0, 1–1) and knows nothing about injuries, line-ups or motivation.

8. Who is behind it

Picks are produced by the automated process above, not hand-picked. The short note on each pick is written by Nito, our AI-assisted editorial persona, from the same numbers. More on how we work in our editorial policy and on the about page.

9. Bet responsibly

Our picks are information, not financial advice, and Tipnito doesn't take bets. Only stake what you can afford to lose. If betting stops being fun, take a break and see our responsible gambling resources. 18+.