Skip to content
MatchDaySync

How the prototypereads a fixture.

The free predictions come from one small, disclosed model. This page is all of it: the inputs, the arithmetic, the limits, and what has not been proven yet.

Restored from the 11 October backup. Current updates are not connected.

Today's predictions
Prototype model, not yet validated
Model
Independent Poisson
Version
poisson-proto-1
Inputs
Stored season goals, read before kick-off
Minimum sample
5 matches per club, 40 per league
Result evidence
Retrospective reconstruction available

Five steps, nothing hidden.

  1. League averages

    Average home goals and away goals per match across every finished league match this season, plus the overall average per side.

  2. Attack and defence

    Each club's goals scored per match divided by the league average gives its attack; goals conceded per match divided by the same average gives its defence.

  3. Expected goals

    Home expected goals = home attack × away defence × league home average. Away expected goals = away attack × home defence × league away average.

  4. Every score

    Each side's goals follow a Poisson distribution around its expected goals, independently. Multiplying the two gives the chance of every score from 0-0 to 7-7, scaled so they add to one.

  5. The outlook

    Home win, draw and away win sum the scores where home leads, they level, or away leads. Each club scoring at least once sums every grid cell where its own score is positive. These figures, three or more goals and both clubs scoring all use the same normalised zero-to-seven goal grid.

A worked example.

Example figures in round numbers, not a real fixture, so the arithmetic is easy to follow.

A league of 100 matches with 150 home and 120 away goals averages 1.50 home and 1.20 away goals a match, 1.35 per side.

The home club scored 18 and conceded 9 in 10. The away club scored 12 and conceded 14 in 10.

That gives the home side 2.07 expected goals and the away side 0.71. The most likely single score is 2-0, at only 13%: even the likeliest score usually does not happen.

Expected goals, Home
2.07
Expected goals, Away
0.71
Most likely score
2-0 (13%)
Three or more goals
53%
Both clubs score
44%
Home score at least once
87%
Away score at least once
51%
Season sample
10 and 10 matches

Play it a thousand times

Each run draws a score from the two expected-goals figures. It shows how widely one estimate can land, and adds nothing the model does not already assume.

Home win–
Draw–
Away win–

Nothing has run yet.

Chance of each score from zero to seven goals per club, normalised within this grid. Rows are home goals, columns are away goals. The outlined cell is the most likely.
01234567
06%4%2%under 1%under 1%under 1%under 1%under 1%
113%9%3%1%under 1%under 1%under 1%under 1%
213%9%3%1%under 1%under 1%under 1%under 1%
39%7%2%1%under 1%under 1%under 1%under 1%
45%3%1%under 1%under 1%under 1%under 1%under 1%
52%1%under 1%under 1%under 1%under 1%under 1%under 1%
61%under 1%under 1%under 1%under 1%under 1%under 1%under 1%
7under 1%under 1%under 1%under 1%under 1%under 1%under 1%under 1%

What it leaves out

  • Line-ups, injuries and suspensions.
  • Rest days, travel and fixture congestion.
  • Shot quality and expected-goals data from shots.
  • Earlier seasons, cups and other competitions.
  • Any dependence between the two sides' goals, which tends to make real draws a little more common than this model says.

Why it is called a prototype

A forecast earns trust from its record: figures frozen before kick-off, then checked against what happened, with calibration published. That record does not exist for this model yet, so every figure says so. Until it does, read the estimates as a tidy summary of the season's scoring, not as a judgement on the match.

A historical reconstruction now scores the model against stored results, with a same-fixture baseline and calibration bands. It uses the current snapshot and cannot establish what data was available when the match was played.

Statistics only, not advice. The season rates on match pages are separate from these estimates: they count what happened, while the model estimates what might happen.

Read the evidence

  • Model evaluation: A historical reconstruction that scores the club prototype against stored results.
  • Archived accuracy ledger: Every stored World Cup line checked against its stored result, misses included.

Where the estimates appear

  • Predictions: Prototype club predictions for eligible fixtures, labelled as not yet validated.
  • Leagues: Tables, fixtures, results and statistics for each stored competition.
  • Fixtures: Stored fixtures by date.

Learn the vocabulary

Other methods and questions

  • Archive methodology: How the archived World Cup 2026 projections were built: sample, filters, confidence and limits.
  • Club ratings method: The Elo and strength formulas behind the club ratings pages.
  • FAQ: Short answers about projections, past purchases and our record.
  • Contact: Methodology questions, corrections, partnerships and press.