Inside the FM27 match engine: how it actually works

AreaDice it scales
In possession shapeFirst Touch, and a team cohesion die
Passing styleDecisions, Composure
Build upTechnique, Vision
Chance creationAnticipation, Off the Ball
Out of possession shapeConcentration, Marking
Line of engagementWork Rate, Aggression
Pressing intensityAnticipation, Tackling
Pressing trapDecisions, Positioning

Using this part we can iterate back to the tactical understand / tactical effectiveness stuff, right? And put the training sessions in that most impact the dice that matter most for the match engine? And leave out some of the others presumably.
 
Using this part we can iterate back to the tactical understand / tactical effectiveness stuff, right? And put the training sessions in that most impact the dice that matter most for the match engine? And leave out some of the others presumably.
Yes. But I think it is also testable with the regular game engine tests. i.e. we change each area just before the match and see how that makes a different over thousands of simulated matches. I'm actually running a formal experiment to test that and will report when it comes back. It also serves as a validation of our understanding of the game engine code.
 
Hey there! Thanks for all your hard work. Apologies for all the questions, but I wanted to ask how roles work in the match engine. I noticed your studies focus on attributes, so I'm curious what part positional roles play. Do they change the weight of each attribute during a match? Are there synergies between roles, e.g. is an inside forward better with an inverted full-back or an overlapping full-back? Or does each player just do his own thing, with any synergy coming from their decisions happening to work together? What happens to attributes that are relevant to a role but not listed as key for it? And do roles mostly just influence what kind of decisions a player makes, like traits do?
 
Hey there! Thanks for all your hard work. Apologies for all the questions, but I wanted to ask how roles work in the match engine. I noticed your studies focus on attributes, so I'm curious what part positional roles play. Do they change the weight of each attribute during a match? Are there synergies between roles, e.g. is an inside forward better with an inverted full-back or an overlapping full-back? Or does each player just do his own thing, with any synergy coming from their decisions happening to work together? What happens to attributes that are relevant to a role but not listed as key for it? And do roles mostly just influence what kind of decisions a player makes, like traits do?
My AI agent and I have also been digging into the match engine, and we’ve looked into roles too. Here’s what we’ve found so far:

Broadly, the engine uses a role in two ways. First, it combines it with the player’s position, team instructions and individual instructions to build his working behaviour: where to move, how much risk to take, whether to favour dribbling, crossing, shooting, etc. Then, during play, some routines also check specific roles directly - for example, when deciding whether a defender moves inside or how attractive a playmaker is as a passing option. So a role is partly a behaviour preset, with additional rules that depend on the situation and teammates.

On your specific questions:
  • Attribute weights: we haven’t established a general system where choosing a role changes the match weights of its highlighted attributes. Roles can change which attributes become useful by changing the situations and actions a player gets involved in.
  • Unhighlighted attributes: they can still matter whenever the relevant action uses them. The highlighted list shouldn’t be read as an exclusive list of what the engine checks.
  • Synergy: there’s both complementary movement and explicit interaction. Inverting defenders consider teammates’ roles and occupied positions. Certain playmakers get specific treatment when teammates evaluate passes to them. Players aren’t just running independent scripts.
  • Inside forward + full-back: we haven’t established a universally better partner. An overlapping player can provide complementary width, but whether that helps more than an inverting player depends on the rest of the shape. We haven’t found a blanket bonus for that pairing.
  • Like traits? Partly: roles influence action preferences. But they also shape positioning and can affect teammates’ decisions, so their influence goes beyond the player’s own choices.
For me, a big part of this is how roles shape the team’s positioning. Well, some roles, anyway. :) In my testing, the “battle of shapes” - how your structure matches up against the opponent’s - has stood out more than individual team/player instruction tweaks (most of them didn't influence the results in any way; but changing roles did). It also partly explains why 5-2-2-1 is so good right now — because it does a great job to counter 3-2-5 attacking shapes. And in general it is pretty solid defensively.

And honestly, it makes me happy. :) I feel like the game is on the right track. It feels close to the problems real coaches solve by moving players around, rather than just shouting “shorter passing!” :) Terzić’s adjustment against Atlético is a great example: Dortmund switched from a narrow 3–2 build-up to a wider 4–1, making Atlético cover much greater distances to press.
 
This is amazing. Thanks!

Though I still have fundamental questions -
Is the score calculated/decided at the start of the game (assuming I change nothing during the game and, for instance, watch only key highlights) or the score is the result of all those dynamic calculations during the match? If dynamic calculations, how would it be posdible with such complexity and huge number of outcomes every second to ensure a realistic score (e.g. 2:1)? If the score is calculated at the start of the game, what are static parameters that define it and what are then the dynamic calculations for?… just visualization with some limitations to lead to the already calculated score?
 
Last edited:
My AI agent and I have also been digging into the match engine, and we’ve looked into roles too. Here’s what we’ve found so far:

Broadly, the engine uses a role in two ways. First, it combines it with the player’s position, team instructions and individual instructions to build his working behaviour: where to move, how much risk to take, whether to favour dribbling, crossing, shooting, etc. Then, during play, some routines also check specific roles directly - for example, when deciding whether a defender moves inside or how attractive a playmaker is as a passing option. So a role is partly a behaviour preset, with additional rules that depend on the situation and teammates.

On your specific questions:
  • Attribute weights: we haven’t established a general system where choosing a role changes the match weights of its highlighted attributes. Roles can change which attributes become useful by changing the situations and actions a player gets involved in.
  • Unhighlighted attributes: they can still matter whenever the relevant action uses them. The highlighted list shouldn’t be read as an exclusive list of what the engine checks.
  • Synergy: there’s both complementary movement and explicit interaction. Inverting defenders consider teammates’ roles and occupied positions. Certain playmakers get specific treatment when teammates evaluate passes to them. Players aren’t just running independent scripts.
  • Inside forward + full-back: we haven’t established a universally better partner. An overlapping player can provide complementary width, but whether that helps more than an inverting player depends on the rest of the shape. We haven’t found a blanket bonus for that pairing.
  • Like traits? Partly: roles influence action preferences. But they also shape positioning and can affect teammates’ decisions, so their influence goes beyond the player’s own choices.
For me, a big part of this is how roles shape the team’s positioning. Well, some roles, anyway. :) In my testing, the “battle of shapes” - how your structure matches up against the opponent’s - has stood out more than individual team/player instruction tweaks (most of them didn't influence the results in any way; but changing roles did). It also partly explains why 5-2-2-1 is so good right now — because it does a great job to counter 3-2-5 attacking shapes. And in general it is pretty solid defensively.

And honestly, it makes me happy. :) I feel like the game is on the right track. It feels close to the problems real coaches solve by moving players around, rather than just shouting “shorter passing!” :) Terzić’s adjustment against Atlético is a great example: Dortmund switched from a narrow 3–2 build-up to a wider 4–1, making Atlético cover much greater distances to press.
What absolute AI dribble, write it up yourself
 
Using this part we can iterate back to the tactical understand / tactical effectiveness stuff, right? And put the training sessions in that most impact the dice that matter most for the match engine? And leave out some of the others presumably.

It is confirmed. Based on your idea, I expected chance creation and pressing intensity to matter more than the others in terms of tactical understand / tactical effectiveness. This is simple because they affect the attributes that matter most. And it turns out it is true.

I've documented these here: Fm-Match-Lab/docs/studies/24-familiarity-areas.md at main · caiograco289/Fm-Match-Lab

Here are the actual results:

Expected points per game; t in brackets.

AreaGegenpress: lower to 20raise to 80Control Possession: lower to 20raise to 80
Chance creation−0.079 (−6.9)+0.021 (+1.9)−0.082 (−7.2)+0.011 (+1.0)
Pressing intensity−0.081 (−6.9)+0.019 (+1.7)−0.066 (−5.6)+0.005 (+0.4)
Out-of-possession shape−0.024 (−2.0)+0.001 (+0.1)−0.029 (−2.6)+0.017 (+1.5)
Passing style−0.023 (−2.0)−0.002 (−0.1)−0.022 (−1.9)+0.004 (+0.4)
Build-up−0.019 (−1.6)+0.005 (+0.5)−0.012 (−1.1)−0.012 (−1.0)
In-possession shape−0.015 (−1.3)−0.010 (−0.9)−0.008 (−0.7)−0.017 (−1.5)
Pressing trap−0.004 (−0.3)+0.005 (+0.5)+0.006 (+0.6)−0.003 (−0.2)
Line of engagement+0.002 (+0.2)+0.005 (+0.4)+0.009 (+0.8)+0.004 (+0.3)
 
Back
Top