Fantasy Matchup Edge
Blog

← Back to Blog

How it works

How our model works

Fantasy Matchup Edge is a proprietary football analytics model, not a generative-AI prediction system. It evaluates player scoring profiles against opponent tendencies and the expected scoring environment of each fixture.

At its simplest, the model is trying to answer a question that traditional fantasy averages and fixture difficulty charts often miss:

What is this particular match likely to ask a player to do, and how well suited is that player to those opportunities?

A player's recent average score is useful, but it does not tell the whole story. The same midfielder can face one opponent that allows huge amounts of possession and passing through his area of the pitch, then another that forces the game into completely different spaces. A centre-back may have very little defensive work in one fixture and spend the next match clearing crosses and contesting duels for 90 minutes.

Fixture difficulty charts are also often too crude and outdated. A "difficult" away fixture may still create specific action routes and opportunities certain players can capitalize on. It's not as simple as "red=hard" or "green+difficult". Every team is different in what opportunities they allow, and every player is different, too.

Fantasy Matchup Edge tries to model those differences.

1. We build a statistical profile for every player

The starting point is the player's underlying football actions rather than simply their previous fantasy scores.

We track a broad range of events including:

This creates what we think of as a player's action fingerprint.

Two players can have similar fantasy averages while getting there in completely different ways.

One midfielder may depend heavily on goals and assists. Another may score consistently through passing, defensive actions and all-round involvement. Two defenders may have the same average score even though one depends on clean sheets and the other regularly accumulates clearances, tackles and interceptions.

Those differences become important once the opponent changes.

2. Role and area of the pitch matter

We do not treat every midfielder, defender or forward as interchangeable.

Where possible, players are associated with the roles and areas in which they actually operate.

A central attacking midfielder, wide forward, full-back and holding midfielder can experience completely different versions of the same match.

That allows the model to ask questions such as:

What does this opponent normally allow to players operating in this role or channel?

rather than simply:

What does this opponent allow to midfielders?

When there is not enough role-specific evidence, the model progressively falls back to broader samples. This prevents a tiny amount of data from being treated as more reliable than it really is.

3. We model what each team tends to produce

Player opportunity is partly determined by the team around them.

A player cannot create five big chances if his team barely gets into the final third. Likewise, a centre-back playing for a dominant possession side may have fewer opportunities for clearances than an equally capable defender playing under sustained pressure.

We therefore model the amount of each action that a team tends to generate.

Think of this as the supply side of the matchup.

If a team is expected to produce unusually high attacking volume, its attacking players have a larger pool of opportunities to benefit from. If the expected match involves long periods without the ball, defensive actions may become more important instead.

4. We also model what the opponent tends to allow

The other half of the equation is the opposition.

Teams do not concede fantasy-relevant actions evenly.

One opponent may allow:

Another team may suppress those exact actions.

Fantasy Matchup Edge compares these opponent tendencies with the player's own action fingerprint.

This is where many of the most interesting matchup edges appear.

If an opponent allows an unusually high amount of a particular action and the player is unusually good at producing that same action, the matchup becomes more interesting than either statistic on its own.

5. The expected match environment changes everything

Not all fixtures should be compared equally.

Manchester City playing as a strong favourite is a very different environment from the same team playing away against another elite side.

The model therefore uses betting-market information to estimate the expected shape of the fixture.

This includes things such as:

This gives the model a view of the expected scoring environment before the match begins.

Historical matches that looked more like the upcoming fixture are given more relevance than matches played in completely different circumstances.

Importantly, this is done softly rather than using rigid buckets.

A projected 3–1 type of environment is not compared only with previous matches that finished exactly 3–1. A 2–1, 3–0, 4–1 or 4–2 game may also contain useful information, with its influence gradually decreasing as the match becomes less similar.

Football is not made of neat categories, so the model tries not to force it into them.

6. Current-season evidence is added carefully

The model also learns from the current season as new matches are completed.

However, one or two early-season games should not completely overwrite several seasons of evidence.

Current-season information is therefore treated as an additional source of evidence rather than as an immediate replacement for the player's established profile.

As more relevant matches are played, that new evidence becomes increasingly useful.

This helps the model adapt to things such as:

without wildly overreacting to one unusual match.

7. We calculate a player-specific expectation for the fixture

Once those pieces are combined, the model produces expected rates for the player's underlying actions.

Conceptually, this combines:

Player ability and tendencies

with

Current role

with

Team supply

with

Opponent allowance

with

Expected match environment

with

Current-season evidence

The important point is that there is no single giant "good fixture" multiplier.

Different actions can move in different directions.

A fixture could be excellent for a defender's clearance volume while being poor for his clean-sheet chances. A midfielder might receive a strong chance-creation matchup but a weaker defensive-action environment.

The final projection is the result of many smaller components rather than one generic fixture difficulty score.

8. We simulate the match thousands of times

An average projection is useful, but fantasy football is highly variable.

A player projected for 6.0 points will not actually score exactly six points every week.

To understand that uncertainty, Fantasy Matchup Edge runs large numbers of match simulations.

Each simulated match begins with a coherent scoreline drawn from the expected market distribution.

For example, one simulation might produce:

Manchester City 3–0 Crystal Palace

Another might produce:

Manchester City 1–1 Crystal Palace

Another might occasionally produce a rare 5–1 or 6–0 result.

The model then changes the expected action environment depending on the simulated state.

A team winning heavily may generate very different actions from a team chasing the game. Defenders protecting a lead, midfielders dominating possession and goalkeepers facing sustained pressure can all experience different statistical environments.

Again, these effects are learned softly from comparable historical match states rather than from rigid exact-score categories.

9. Goals and assists remain tied to the simulated scoreline

This is an important detail.

If a simulated team wins 3–1, the model does not independently generate five goals for its players.

There are exactly three team goals available in that simulated match.

Individual players receive scoring opportunities according to their own goal propensity, role and matchup, but their outcomes remain constrained by the actual simulated team score.

Assists are treated in a similar way.

This helps keep the fantasy simulations connected to a believable football match rather than creating unrelated player outcomes.

10. The simulation produces probabilities, not just averages

Once we simulate a fixture many thousands of times, we get an entire distribution of possible outcomes.

For FPL, that lets us show statistics such as:

For Sorare, we can estimate:

This allows us to distinguish between two players with similar averages but very different ceilings.

One may be relatively consistent.

Another may have a much greater chance of producing a gameweek-winning score.

11. The scoring model is platform-specific

The football underneath the model is shared, but FPL and Sorare reward different things.

That means the same matchup can produce different conclusions depending on the fantasy platform.

FPL

The simulation translates football events into FPL scoring, including:

Sorare

Sorare rewards a much wider range of football actions, so passing, defensive work, possession actions, chance creation and other all-around contributions play a larger role.

A player can therefore be an excellent Sorare matchup without necessarily being one of the strongest FPL plays, and vice versa.

That distinction is intentional.

12. Our projections are currently "if the player starts"

Unless explicitly stated otherwise, Fantasy Matchup Edge projections answer:

What do we expect if this player starts the match?

We do not currently multiply projections by an artificial start probability or predicted number of minutes.

That is important when interpreting the rankings.

A rotation player can have an excellent projection if selected, even if he is unlikely to make the starting XI.

For our public rankings we therefore separately identify players who have been starting, players with rotation risk and players who have not yet started this season.

This keeps two separate questions separate:

How good is the matchup if he plays?

and

How likely is he to play?

13. Why a player projects well is as important as the number

One of the main goals of Fantasy Matchup Edge is not simply to publish a list of projections.

We want to explain why the model likes or dislikes a player.

For example, a player's strong projection might be driven by:

In other cases, the model may simply conclude:

This is an excellent player whose projection is mostly driven by his normal level of production. The matchup itself adds very little.

We think that distinction matters.

A high projection is not automatically a "matchup edge."

Sometimes the player is simply very good.

What Fantasy Matchup Edge is trying to do

Fantasy football projections will never remove uncertainty from football.

That is not the aim.

The aim is to make better use of the information available before a match.

Rather than asking only:

How many points has this player averaged?

we want to ask:

What kind of football is this fixture likely to produce?

Where are the opportunities likely to appear?

What does the opponent tend to allow?

Which players are best equipped to exploit those opportunities?

And finally:

How often does that combination turn into a meaningful fantasy score?

That is the idea behind Fantasy Matchup Edge.