See beyond the score.
Understand what is really happening.

GrabbleStats turns football data into clear, usable signals. Our models compute, the AI puts it in context — it explains the number, it never produces it. You see the trends, the gaps and the events that deserve your attention, without anyone pretending the data says more than it can.

Less noise. More context. Better readings of the game.

10K+Matches AnalyzedEvery Season
50+Expert ModelsBuilt In-House
100%Data-DrivenNo Guesswork
Strong signal20:00
Illustration
PSG
Marseille
61%PSG win▲ +4 pp
Superior transition play
Fragile on set pieces

Illustrative screenshots. The fixtures, probabilities and results shown on this page are demonstration examples, not live data.

The toolstwelve pages, one logic: explain the number
RadarThe matches that deserve your attention, one at a time, with the why and the why not.
Match IntelligenceAnalysis, market, context and history on a single sheet, five tabs.
Line-upsThe expected eleven placed on the pitch, each player's status, the effect on the probability.
LiveThe model recalculates on every event, with the trajectory and projections to the 90th minute.
And the analysis tools
Signal engine
Market Intelligence
Player scouting
Comparator
Manager DNA
Referee DNA
Laboratories
Continuous verification
Markets and models

40 markets,six models,one calibration.

Coverage says nothing about the quality of an analysis. What matters is what we actually model, with which models, and how well they hold up. One family is fitted and tested today; the others say so.

Goals modelDixon-Coles: two Poisson intensities per side with a low-score correction, fitted per competition on historical goals with time decay, then tested on a season it never saw.
Result modelThe same joint distribution, read as a result: 1X2, double chance and draw-no-bet are summations over the score matrix, so they cannot contradict each other.
Set-piece modelCorners as a negative binomial over a shared total, split between the sides. The law is written and validated against measured dispersion and correlation; it is not yet fitted per team, so it is not published.
Discipline modelNot built. Cards do not fit the possession-sharing structure that works for corners — measured correlation between sides is positive, the structure predicts negative. A shared match-intensity model is needed first.
Player modelNot built. Per-player, per-match statistics exist back to 2016 and are not yet ingested.
Live modelThe remaining-time Poisson: the score is a fact, what is left follows the pre-match intensities scaled to the minutes remaining. It reacts to the score and the clock, to nothing else.
What we modelthe five main families of 40
Match result
4 marketsEnsemble
Which markets
1X2Double chanceHalf-time winner
CalibrationBrier0.187
Goals
8 marketsDixon-Coles
Which markets
Over / under goalsBoth teams to scoreCorrect scoreGoals per half
CalibrationBrier0.196
Set pieces
3 marketsNegative binomial
Which markets
CornersSet piecesCorners per team
CalibrationBrier0.213
Discipline
5 marketsReferee-conditioned
Which markets
Yellow cardsCards per teamPenalty awarded
CalibrationBrier0.224
Players
3 marketsPlayer model
Which markets
GoalscorerAssist provider
CalibrationBrier0.241

The Brier score measures accuracy: the lower it is, the better calibrated the model. Below 0.20 is considered good in football. We publish ours market by market, including where it is weakest.

Illustrative figures. This block shows the format in which calibration is published per market family; the scores are demonstration examples, not a measured result.

The Radar

One question,one matchat a time.

Most platforms hand you every match and leave you to sort them. We do the opposite: out of the hundreds of fixtures available each day, a few clear the analytical threshold and are shown to you one at a time, with what supports the figure and what could prove it wrong.

The threshold is explicit. A match enters the Radar when the model's conviction, the data quality and the gap to the market all three clear their threshold.
Every figure carries its reasons. Two columns under the probability: what supports it, what could mislead us. Never one without the other.
The list view stays available. If you would rather see everything at once, the way out is one tap away. Focus mode is a default, not a constraint.
grabblestats.com/radarDemo
Decision queue3 of 13 fixtures clear the threshold
FixtureCompetitionModelGapLevel
PSGMarseilleLigue 161%+7.0Strong
LiverpoolArsenalPremier League53%+4.5Watch
BarcelonaValenciaLaLiga57%+3.2Watch
What supports the figure61% · 24% · 15%

Demonstration screenshot. PSG — Marseille and its figures are an example, not a match in progress.

Verification

We publishour mistakes.

An analysis platform that only shows its successes is worth nothing. Every analysis is reviewed after the match: what we announced, what happened, which reasons held and which were proved wrong. Failures are displayed at the same rank as successes.

10 704
out-of-sample matches tested
0.222
median Brier across 39 competitions
14
models retired for failing to beat the baseline
Read the methodology

The two cards opposite illustrate the verification format. The fixtures and figures are demonstration examples.

Analysis confirmedExample18 Jan
Napoli — Roma1 – 1
Our analysis
44%
Draw
The consensus
27%
gap +17 pp

The biggest gap to consensus this month. A single creative player's absence moves the probability more than the market prices in.

Analysis proved wrongExample18 Jan
Arsenal — Brighton0 – 2
Our analysis
63%
Arsenal win
The consensus
66%
gap −3 pp

We overvalued possession as an indicator of chances. This match fed a correction to the weight of possession in the creation index.

What wearenot.

Football attracts promises. We would rather state up front where our work stops.

Not a betting site
No advice, no selections to follow, no link to an operator. Odds serve as comparison, they do not enter the calculation.
Not a promise of a result
A probability of 61% means the opposite happens in 39% of cases. We do not claim to predict a given match.
Not a sports outlet
No news, no rumours, no opinion. Measurements, their context, and their verification.
Not a black box
The methodology is public, the sources listed, the calibration open, and every calculation carries a verifiable identifier.

Start by understanding.

The free plan gives access to the Radar and the full methodology. Paid plans open up the explanation, the line-ups, the market gaps and the verified history.

Free
€0
To understand what the model does
Today's Radar, one signal a day
Probabilities for Europe's top 5
Results and fixtures
Follow fixtures and teams
Full methodology
Verified analysis history
Model Health and calibration
Start for free
Intelligence ProMost chosen
€15/ month
Billed €180 per year · 14-day trial
Everything in Free, plus:
Every signal, every covered competition
Why / What could mislead us
Probable line-ups and their impact
Model vs market gaps
Live with the model's trajectory
Alerts when a probability moves
Comparator, five entity types
Surprise Lab, Time Machine, Hypothesis Lab
Full profiles: teams, players, referees, coaches
Start the trial
Expert
On request
Clubs, media, analysis departments
Everything in Intelligence Pro, plus:
Terminal and CSV / JSON exportSoon
API access and webhooksSoon
Multiple seats and rolesSoon
Competitions on demandSoon
Methodological support
Talk to the team
Mobile app

The Radar in your pocket.

The mobile app carries the whole product: decision queue, match sheets, line-ups, live and alerts. Also installable as a web app straight from the browser, without going through a store.

Coming to iPhone and iPadiOSSoonComing to phone and tabletAndroidSoonFrom the browser, todayInstall the PWA

Native apps are in preparation: only the web-app install is available today. The official App Store and Google Play badges will replace these buttons at launch — we do not redraw Apple's or Google's marks.

Notifications when a probability moves sharply
Offline access to your latest analyses
Line-ups as soon as they are officially published
Same account, same saved items, on every device

See beyond the score.

Fourteen days of Intelligence Pro, no card required. After that the free plan continues, asking nothing of you.