GeorgeFloydOverdosed said: I think you said you used the original version 'FM26 Player Scoring System' is derived from but worked out how to read direct from memory. Expand Yeah version 1 was using their same penalties and goals rating system (or whatever the non-standard mode is called) except without the need to export stuff. Then I switched to your blended weights.
I'll make it possible to edit the weights used in my app too so people can use it to find players and you can use it to experiment.
GeorgeFloydOverdosed said: Unfortunately there is no proper substitute for Genie Scout. I've been considering FMRTE, but the free version doesn't allow you to save weight changes, and only some people will be willing to pay money for FMRTE. Expand I've just bought a house and after I get myself situated I'll have access to my Windows PC again so I can port the tool I've been developing on my laptop. It'll let you read your team from in-game memory and search the player database based on a number of constraints.
@GeorgeFloydOverdosed I've been thinking recently about an experiment to measure the importance of attributes for a given role (e.g. Striker); could you tell me if you think this is worthwhile?
Create a test league (maybe 10 or so teams) with identical coaching staffs (Assistant Manager is probably sufficient, since he'll be the one managing games), and identical players in all positions. That is, all the teams' goalkeepers are identical (copied from a real player), as are their fullbacks, their centrebacks and so on. None of the players ought to be able to play the position you're testing (e.g. Striker).
All teams are set up to play the same formation (if this can't be done in the pre-game editor, it would require ten human managers on holiday, hence identical Assistant Managers). All players have their attributes and conditions frozen.
The experiment is taking real-world strikers (e.g. Haaland, Mbappe, Vini, Goncalo Ramos) and putting multiple copies of them on to different teams, meaning all teams are identical except for their different strikers. If they're all playing the same tactics with the same coaches, the only variation will come from their strikers.
And I don't even think you necessarily need to look at the league table at the end of the experiment to determine which striker is better, since games can be won and lost by other players too. I'd look at the goals scored by the strikers, their assist count and possibly their average match ratings (or other things like key passes, xG, if you don't trust how match ratings are generated).
From here, you would perform a linear regression to assess what attributes / personality values are important in affecting the things we care about. I have to think that if was useful, someone would have done it by now but I can't see what's wrong with it.
GeorgeFloydOverdosed said: If you go by just pace/acc, this guy is nothing special, and yet he's averaging 7.58 rating. That's less than Mbappe's 7.90 at same club, which means pace/acc still mogs, but low pace/acc shouldn't be automatically discounted. Expand Does this lad get 90 minutes or come off the bench?
I'm interested to know if you found the location in memory where shortlisted or filtered players are stored. Genie Scout has a way of creating a list of players for you in the app, who come from the list of players you currently have on-screen in the game.
I can't find this list of players in memory (I assume it's an array of person or player addresses but I have a sinking feeling it's more complicated than that).
FYI I'm building a tool for myself on Linux and that's the last nice-to-have feature I'm missing. It's such a ball ache clicking through them one by one right now to see their ratings.
Yarema said: I mean, if you want to win Premier league with 1 CA players you kind of need every possible advantage you can get. Or any other feat that significantly defies odds. But even with the top tactics you can go right or wrong with player selection and that's what is being tested. Fairly sure that with an average tactic you stand no chance. Expand
True but I'm interested in learning what attributes are important to the match engine. For that, I'd think that all teams should probably be playing with the same tactic so the only influence comes from attribute variance.
GeorgeFloydOverdosed said: I myself always use Knap's EF 424 IF HP V2 P101 AC tactic. My reasoning is that's popular and it's one of the most successful tactics, so why bother tailoring to say a more defensive tactic. I have been considering trying an 'underdog' tactic out for a change for my low pace/acc testing though, to see if it makes it more viable. I'll report later on how that goes. Expand My favourite tactic. Is there a concern that your low-CA teams actually aren't good enough to finish fourth on their merits but through the strength of the tactics? I don't want to tell you or any of the other hardworking testers your business but it might be more instructive not to use a tactic that could see even the worst team finish in the top half.
Now it makes sense why that team wasn't comparable in terms of rating to the team who finished fourth in my save.
GeorgeFloydOverdosed said: Now is everyone lying, or making serious errors here? I don't think it's either. I think that the process changes at different levels of attributes, in ways that are too difficult to predict and require brute force testing to deduce. Expand Messy, messy, messy.
Out of interest, do you and others run these tests using the same tactics? If you're getting conflicting reports of what attributes matter, it might come down to player roles, which is what the game tells us.
GeorgeFloydOverdosed said: you are essentially pointing out is that the team of players I said achieve 4th, are not in line with the 'Blended' ratings file I posted, and that your own Newcastle result of 4th (same position) yet with players with significantly higher % rating is further evidence that the Blended file is out of whack. Expand Yeah, precisely. It would have been lovely if the weighted formula produced similar ratings for your artificial player and a real-world example but the real team is clearly regarded as much better.
If you're seeing a certain synergy between some attributes or significant performance gains/drop-off after crossing attribute thresholds, it would suggest that the actual formula is more complicated than just summing up weighted values. It sounds obvious but I wouldn't put it past Sports Interactive given how many corners they've cut.
I think Harvest Green in one of his spreadsheets had a formula which changed weights depending on how high or low the values were; I might try to implement that and compare the players that way.
Rain said: What is this screenshot from? Expand An analysis tool I'm working on. It's like those websites that ask you to export all your players as a webpage for analysis, except this reads data straight from the game (because who's got time for that?)
This makes it more like Genie Scout, but without the ludicrous startup time. Since then I've added some editing tools that make it similar to FMRTE too, except mine is free.
It's Linux only at the moment though (because that's all I use and the only machine I have to develop on, so I can't test the memory reading code on anything else).
GeorgeFloydOverdosed said: A team of these (hiddens ~8-13; DC jump 17) achieves 4th-5th in the Premier League: Expand
This is surprising to me. I looked at how effective a player like this would be using your blended weights and they don't appear to be anywhere near good enough to achieve a top-four finish.
I created a player using the same attribute values you listed (see screenshot) and calculated his rating at all outfield positions (I gave him 17 jumping reach when calculating centre-back rating). Here's what I got:
| Position | Rating | |----|---------| | FB | 74.4% | | CB | 76.98% | | WB | 73.97% | | DM | 67.56% | | W | 65.42% | | MC | 68.73% | | AM | 69.26% | | ST | 62.37% |
Don't give too much thought to the actual values because my weights are arbitrary but here's the rating of Newcastle's best XI on May 4th, 2025 in my current save (the season is over, they've finished in 4th place):
| Player | Position | Rating | |----|--------|--------| | Nick Pope | GK | 75.65% | | Ferdi Kadıoğlu | DL | 85.35% | | Gianluca Mancini | DC | 77.97% | | Sven Botman | DC | 78.32% | | Tino Livramento | DR | 78.93% | | Alexander Isak | ML | 79.31% | | Joe Willock | MC | 74.17% | | Sandro Tonali | MC | 76.27% | | Joelinton | MR | 74.78% | | Bryan Mbeumo | ST | 69.2% | | Anthony Gordon | ST | 68.2% |
The low-CA player you posted is at least 3-4% worse off than this real team that finished 4th in my game (in 5th place was an even stronger Manchester United). If I make Newcastle play with a defensive midfielder, it picks Tonali and Willock who score 74% each.
If you holidayed with a tactic which isn't considered overpowered, I see this as a bit of an indictment of the weights we've been talking about.
Of course, if you've crunched the numbers yourself and actually find your team is considered much stronger than your closest competition (ideally without the use of Genie Scout thumbing the scale mysteriously) then I'll hold my hands up. This has left me confused though.
EDIT: It strikes me that this might be what you meant when you wrote this:
GeorgeFloydOverdosed said: Now I have tried using HarvestGreen, Orion, my own, etc. data to try and optimize for positional differences. But everything I tried just made the result worse, even gentle adjustments. Expand
DOUBLE EDIT: Ignore the CA in the Lewdandowski screenshot. Ignore everything except the ability values and the overall rating. I took the real player and modified his attributes just to see what the formula would spit out.
GeorgeFloydOverdosed said: It is concerning though that you find Salah to be ahead of Mbappe. That's a 7.5% difference. Not sure what's going on with that. Expand
It could be a combination of the reduced impact of personality attributes in Genie Scout + the importance of having two strong feet (Mbappe has 100 and 50 for his strong and weak feet; Salah has 100 and 40). I think this is why Vini is considered so impressive by Genie Scout too (he's 100 and 60).
GeorgeFloydOverdosed said: This is all a genie scout problem, so I would think your own calculated player rankings are correct and you should use them. Expand Well that's not the conclusion I expected. Yay me, I guess.
GeorgeFloydOverdosed said: Personality attributes are valued at 20% of physical/mental/technical. This unfortunately means attributes like 'pressure' will have to be hard-capped at 20%. EDIT: It is unfortunately more complicated than this. Expand Is this 20% scale still the case? If so then I think it's not a coincidence that mental/physical/technical/goalkeeper attributes are internally stored as values 1-100 and personality attributes are 1-20.
If Genie Scout doesn't normalise these values then it's multiplying the personality weights against values that are one-fifth those of M/P/T/G attributes.
Apologies if I've missed this somewhere @GeorgeFloydOverdosed, but at what point (if any) do we start to penalise a player for having reduced competency in a position for which they're being rated (e.g. I think Vini has 16/20 for Striker but is considered to be the second best in the world by your blended metric)?
I will have more dumb questions later as soon as I organise them.
EDIT: I might as well ask this one now, while it's fresh. When you rated players according to various weights and formulae in this post, where did the player data come from?
I ask because I tried recreating the calculations manually (using, for example, the Genie weights you attached via screenshot earlier) and got a completely different order of players when looking for the best strikers in the world.
I created a new game in FM 24, July 3rd 2023, and got the following when calculating using your Fast Striker weights:
1. Erling Haaland 2. Mohamed Salah 3. Kylian Mbappé 4. Lautaro Martínez 5. Marcus Rashford 6. Victor Osimhen 7. Robert Lewandowski 8. Lionel Messi 9. Randal Kolo Muani 10. Romelu Lukaku
I did get it working when using HarvestGreen's new GK weights but either we're using different data for strikers or I'm incorrectly guessing how Genie Scout calculates ratings via weights (it is the sum of all attributes multiplied by their respective weights, right?)
GeorgeFloydOverdosed said: High PA (realistically 130 min for premier league standard. 200 ideal.) High CA to PA difference (the more the better) 20 training facilities (17 max required) 25-30 matches/season (28 ideal. 30 safe. 36 max required. 15 min. 2520 min exactly. 36matches@70min.) 10+ ambition (6 min. 10-13 ideal. 16 max required.) 10+ determination (6 min. 13 ideal. 18 max required.) 20 professionalism (13 min. 15+ ideal.) Expand
Sorry if this seems like a dumb question but what do you mean by "max required"? Do you mean that the positive impact from an attribute is capped at this value?
Sorry to pile on with demands @GeorgeFloydOverdosed, but can you provide these weights as raw data? That is, I don't think Genie Scout runs on Linux but I'd be interested in seeing what you've come up with.
GeorgeFloydOverdosed said: Obvious example would be finishing, where I found you can win the Premier League with 1 CA players and a ST with 1 finishing, it's just that the bulk of the goalscoring gets shifted to the other position that has the highest finishing. Expand
Well this is the thought I had. People explain the match engine the way they think it should work (e.g. "the game sees the opposing winger running to the byline, so it looks at your fullback's positioning, composure and decision making to determine whether he can track the runner and make himself available to the tackle" ) but my thought was: what if it just looks at the offensive capabilities of your team, your defensive capabilities (irrespective of what positions the players who have these attributes are in) and just simulates the results based off that, generating highlights to fit?
Hence my idea for the experiment, effectively swap your attackers' and defenders' attributes (but not their positional familiarities) and see what happens.
Idea for an experiment (which I'd run if I had a more performant PC, but I'm at the mercy of you kind strangers): what if you assembled a squad of players, who were capable of winning the league, but then randomly shuffled each player's set of attributes amongst each other?
For example, your LB has your MC's attributes, your ST has your RB's attributes, your DM has your ST's attributes and so on (leaving goalkeepers alone). Naturally each player's role analysis would be in the toilet but if experiments have revealed that tactics are less impactful than attributes alone, it might also follow that the sum talent of your team is more important than who plays where (assuming you're 1. using a tactic that makes sense and 2. players aren't playing out of position, just wildly out of role).
DOUBLE EDIT: I get it now but I've added my latest thought at the bottom.
@Possebrew Do you mind explaining the logic behind the values for the outfield and goalkeeper scaling? I read your initial post but couldn't understand what they're for.
It seems to me that they exist only to take the final rating and amplify it, but by drastically different amounts. If I've understood the formula correctly, it seems like the perfect outfield player would have a rating (using the goal weights/penalties formula) of 1016.49 and the perfect goalkeeper would be 92.75.
I don't understand why. Why not normalise the values so the best player, regardless of position, is 100? Or 1000 if you want to give more scope for variety (given that the average player is far, far, far away from perfect)?
EDIT: setting the weights to outfield:23.4631 and goalkeeper:37.7358 would ensure the perfect players are both 1,000 (but I'm not sure if that conflicts with what you were after re: the combined player contributions?)
DOUBLE EDIT: Okay, I understand now. You divide the contribution for all stats by 10 for outfield players to reflect the fact that there are 10 players contributing to the points/goals delta, and then multiply the final answer by a number to amplify the result again.
These two operations cancel each other out though, making one of them redundant.
To use my normalised example above, if you forego the division by 10 for outfield players and then multiply the final answer by 2.34631 (instead of 23.4631) then the final rating comes out to the same thing.
Yeah version 1 was using their same penalties and goals rating system (or whatever the non-standard mode is called) except without the need to export stuff. Then I switched to your blended weights.
I'll make it possible to edit the weights used in my app too so people can use it to find players and you can use it to experiment.
I've just bought a house and after I get myself situated I'll have access to my Windows PC again so I can port the tool I've been developing on my laptop. It'll let you read your team from in-game memory and search the player database based on a number of constraints.
No fee, no weird start-up time.
It'll just take a little bit longer.
Create a test league (maybe 10 or so teams) with identical coaching staffs (Assistant Manager is probably sufficient, since he'll be the one managing games), and identical players in all positions. That is, all the teams' goalkeepers are identical (copied from a real player), as are their fullbacks, their centrebacks and so on. None of the players ought to be able to play the position you're testing (e.g. Striker).
All teams are set up to play the same formation (if this can't be done in the pre-game editor, it would require ten human managers on holiday, hence identical Assistant Managers). All players have their attributes and conditions frozen.
The experiment is taking real-world strikers (e.g. Haaland, Mbappe, Vini, Goncalo Ramos) and putting multiple copies of them on to different teams, meaning all teams are identical except for their different strikers. If they're all playing the same tactics with the same coaches, the only variation will come from their strikers.
And I don't even think you necessarily need to look at the league table at the end of the experiment to determine which striker is better, since games can be won and lost by other players too. I'd look at the goals scored by the strikers, their assist count and possibly their average match ratings (or other things like key passes, xG, if you don't trust how match ratings are generated).
From here, you would perform a linear regression to assess what attributes / personality values are important in affecting the things we care about. I have to think that if was useful, someone would have done it by now but I can't see what's wrong with it.
Does this lad get 90 minutes or come off the bench?
I can't find this list of players in memory (I assume it's an array of person or player addresses but I have a sinking feeling it's more complicated than that).
FYI I'm building a tool for myself on Linux and that's the last nice-to-have feature I'm missing. It's such a ball ache clicking through them one by one right now to see their ratings.
@Gengar1001
True but I'm interested in learning what attributes are important to the match engine. For that, I'd think that all teams should probably be playing with the same tactic so the only influence comes from attribute variance.
My favourite tactic. Is there a concern that your low-CA teams actually aren't good enough to finish fourth on their merits but through the strength of the tactics? I don't want to tell you or any of the other hardworking testers your business but it might be more instructive not to use a tactic that could see even the worst team finish in the top half.
Now it makes sense why that team wasn't comparable in terms of rating to the team who finished fourth in my save.
Messy, messy, messy.
Out of interest, do you and others run these tests using the same tactics? If you're getting conflicting reports of what attributes matter, it might come down to player roles, which is what the game tells us.
Yeah, precisely. It would have been lovely if the weighted formula produced similar ratings for your artificial player and a real-world example but the real team is clearly regarded as much better.
If you're seeing a certain synergy between some attributes or significant performance gains/drop-off after crossing attribute thresholds, it would suggest that the actual formula is more complicated than just summing up weighted values. It sounds obvious but I wouldn't put it past Sports Interactive given how many corners they've cut.
I think Harvest Green in one of his spreadsheets had a formula which changed weights depending on how high or low the values were; I might try to implement that and compare the players that way.
An analysis tool I'm working on. It's like those websites that ask you to export all your players as a webpage for analysis, except this reads data straight from the game (because who's got time for that?)
This makes it more like Genie Scout, but without the ludicrous startup time. Since then I've added some editing tools that make it similar to FMRTE too, except mine is free.
It's Linux only at the moment though (because that's all I use and the only machine I have to develop on, so I can't test the memory reading code on anything else).
This is surprising to me. I looked at how effective a player like this would be using your blended weights and they don't appear to be anywhere near good enough to achieve a top-four finish.
I created a player using the same attribute values you listed (see screenshot) and calculated his rating at all outfield positions (I gave him 17 jumping reach when calculating centre-back rating). Here's what I got:
| Position | Rating |
|----|---------|
| FB | 74.4% |
| CB | 76.98% |
| WB | 73.97% |
| DM | 67.56% |
| W | 65.42% |
| MC | 68.73% |
| AM | 69.26% |
| ST | 62.37% |
Don't give too much thought to the actual values because my weights are arbitrary but here's the rating of Newcastle's best XI on May 4th, 2025 in my current save (the season is over, they've finished in 4th place):
| Player | Position | Rating |
|----|--------|--------|
| Nick Pope | GK | 75.65% |
| Ferdi Kadıoğlu | DL | 85.35% |
| Gianluca Mancini | DC | 77.97% |
| Sven Botman | DC | 78.32% |
| Tino Livramento | DR | 78.93% |
| Alexander Isak | ML | 79.31% |
| Joe Willock | MC | 74.17% |
| Sandro Tonali | MC | 76.27% |
| Joelinton | MR | 74.78% |
| Bryan Mbeumo | ST | 69.2% |
| Anthony Gordon | ST | 68.2% |
The low-CA player you posted is at least 3-4% worse off than this real team that finished 4th in my game (in 5th place was an even stronger Manchester United). If I make Newcastle play with a defensive midfielder, it picks Tonali and Willock who score 74% each.
If you holidayed with a tactic which isn't considered overpowered, I see this as a bit of an indictment of the weights we've been talking about.
Of course, if you've crunched the numbers yourself and actually find your team is considered much stronger than your closest competition (ideally without the use of Genie Scout thumbing the scale mysteriously) then I'll hold my hands up. This has left me confused though.
EDIT: It strikes me that this might be what you meant when you wrote this:
GeorgeFloydOverdosed said: Now I have tried using HarvestGreen, Orion, my own, etc. data to try and optimize for positional differences. But everything I tried just made the result worse, even gentle adjustments.
DOUBLE EDIT: Ignore the CA in the Lewdandowski screenshot. Ignore everything except the ability values and the overall rating. I took the real player and modified his attributes just to see what the formula would spit out.
It could be a combination of the reduced impact of personality attributes in Genie Scout + the importance of having two strong feet (Mbappe has 100 and 50 for his strong and weak feet; Salah has 100 and 40). I think this is why Vini is considered so impressive by Genie Scout too (he's 100 and 60).
Well that's not the conclusion I expected. Yay me, I guess.
GeorgeFloydOverdosed said: Personality attributes are valued at 20% of physical/mental/technical. This unfortunately means attributes like 'pressure' will have to be hard-capped at 20%. EDIT: It is unfortunately more complicated than this.
Is this 20% scale still the case? If so then I think it's not a coincidence that mental/physical/technical/goalkeeper attributes are internally stored as values 1-100 and personality attributes are 1-20.
If Genie Scout doesn't normalise these values then it's multiplying the personality weights against values that are one-fifth those of M/P/T/G attributes.
I will have more dumb questions later as soon as I organise them.
EDIT: I might as well ask this one now, while it's fresh. When you rated players according to various weights and formulae in this post, where did the player data come from?
I ask because I tried recreating the calculations manually (using, for example, the Genie weights you attached via screenshot earlier) and got a completely different order of players when looking for the best strikers in the world.
I created a new game in FM 24, July 3rd 2023, and got the following when calculating using your Fast Striker weights:
1. Erling Haaland
2. Mohamed Salah
3. Kylian Mbappé
4. Lautaro Martínez
5. Marcus Rashford
6. Victor Osimhen
7. Robert Lewandowski
8. Lionel Messi
9. Randal Kolo Muani
10. Romelu Lukaku
I did get it working when using HarvestGreen's new GK weights but either we're using different data for strikers or I'm incorrectly guessing how Genie Scout calculates ratings via weights (it is the sum of all attributes multiplied by their respective weights, right?)
High CA to PA difference (the more the better)
20 training facilities (17 max required)
25-30 matches/season (28 ideal. 30 safe. 36 max required. 15 min. 2520 min exactly. 36matches@70min.)
10+ ambition (6 min. 10-13 ideal. 16 max required.)
10+ determination (6 min. 13 ideal. 18 max required.)
20 professionalism (13 min. 15+ ideal.)
Sorry if this seems like a dumb question but what do you mean by "max required"? Do you mean that the positive impact from an attribute is capped at this value?
(No rush, whenever you have time).
Well this is the thought I had. People explain the match engine the way they think it should work (e.g. "the game sees the opposing winger running to the byline, so it looks at your fullback's positioning, composure and decision making to determine whether he can track the runner and make himself available to the tackle" ) but my thought was: what if it just looks at the offensive capabilities of your team, your defensive capabilities (irrespective of what positions the players who have these attributes are in) and just simulates the results based off that, generating highlights to fit?
Hence my idea for the experiment, effectively swap your attackers' and defenders' attributes (but not their positional familiarities) and see what happens.
For example, your LB has your MC's attributes, your ST has your RB's attributes, your DM has your ST's attributes and so on (leaving goalkeepers alone). Naturally each player's role analysis would be in the toilet but if experiments have revealed that tactics are less impactful than attributes alone, it might also follow that the sum talent of your team is more important than who plays where (assuming you're 1. using a tactic that makes sense and 2. players aren't playing out of position, just wildly out of role).
@Possebrew Do you mind explaining the logic behind the values for the outfield and goalkeeper scaling? I read your initial post but couldn't understand what they're for.
It seems to me that they exist only to take the final rating and amplify it, but by drastically different amounts. If I've understood the formula correctly, it seems like the perfect outfield player would have a rating (using the goal weights/penalties formula) of 1016.49 and the perfect goalkeeper would be 92.75.
I don't understand why. Why not normalise the values so the best player, regardless of position, is 100? Or 1000 if you want to give more scope for variety (given that the average player is far, far, far away from perfect)?
EDIT: setting the weights to outfield: 23.4631 and goalkeeper: 37.7358 would ensure the perfect players are both 1,000 (but I'm not sure if that conflicts with what you were after re: the combined player contributions?)
DOUBLE EDIT: Okay, I understand now. You divide the contribution for all stats by 10 for outfield players to reflect the fact that there are 10 players contributing to the points/goals delta, and then multiply the final answer by a number to amplify the result again.
These two operations cancel each other out though, making one of them redundant.
To use my normalised example above, if you forego the division by 10 for outfield players and then multiply the final answer by 2.34631 (instead of 23.4631) then the final rating comes out to the same thing.
Check my maths on that but I think it's right.