Although I don't agree with the reasoning given as critique of my weightings, I do believe the underlying idea that the precise rankings may be somewhat impaired.
I attempted to address this with weighting the weights, but it produced worse results, or at least no better.
I've tried comparative deduction by comparing several players match ratings & goal tallies with their attributes. Some key problems with this approach:
- Match ratings have been demonstrated to favor presence of technical/mental ability over actual wins - Goals scored can simply reflect finishing, and poor finishing has minimal effect on the team win rate (it simply means someone else with higher finishing in the team scores the goals) - I actually managed to get the players to 'line up' with the ratings/goals a lot better, but it took a very strange set of GS weightings to get there. Either I'm on the wrong track, or even if it's right, I would be building a house without being able to know the foundations.
I concluded that what's actually needed is more result-based data, but on individual players. It's more tedious, and will require a fair amount of samples for good accuracy, but I think it's the best approach.
Method
I use my meta team at Man City, but replace a position with duplicates of a particular player. If I'm testing ST, I remove proficiency in other positions so as not to mess up the results, as what I'm most interested in is deducing what constitutes the precise ranking of STs. Full detail is used for all first team competitions.
Results
Here's some data I've collected so far (ranked by team performance with player as ST):
The reason I picked Le Fondre to include is because I saw him compared in a Youtube video, and even though he's a 36-year-old with 9 pace/acc, he managed the results you see here. A takeaway from this for me is that pace/acc isn't strictly necessary. Using the comparative deduction I mentioned before, I found that anticipation and composure are more critical than pace/acc.. as I said, these are strange conclusions, so rather than make the facts fit the theory, I'm going to be getting the facts straight first and then seeing what theory fits the data.
I'll probably do about 5-6 samples each of ~10 players for ST and then see if I can deduce consistent patterns to create accurate weightings.
iezzex said: So heres my question If i still win, but i wanna find a keeper that actually saves, what i shall look for Rn Chevalier looks good for me, best keeper for me yet, but he's old in my save, so i wanna find or make a good regen, waht to look for? Expand I may have a look into it later, but I suspect that it's fundamentally your tactic and overall level of quality of your players, not the attributes of your GK, that determine the number of goals he concedes.
We can influence who scores goals in a team, but that's because at the end of the day a goal is going to be scored. But you can't reduce the number of goals your GK concedes, because he's the only one who can concede them.
iezzex said: Is it all about team overall performance or not? Cuz what i've seen on my long real save, as example, Courtouis were bad (with my high def line at least) Yeah i won alot, and gd was insane, but he's avg rating was like 6.8 or something like that haha Expand To clarify, my template, and therefore the weightings, are based on team position at end of season, not match ratings.
keithb said: Here you go. It will make the game very easy if used with a top tactic and set pieces which score a lot. Thats one of the main things about high jumping reach is those players will score a lot from set pieces. Expand Interesting that you give all hidden attributes 0 weighting even though you claim Genie Scout isn't bugged.
According to those weights, Donnarumma is ranked 9th, while Muric is 202nd.
Meanwhile, actual Premier League test results in reality:
Muric - 3rd 6.81, Sacked March (10th) 6.67, Sacked Febuary (10th) 6.78 = Average 7.666th 6.75 Oblak - 6th 6.87, 7th 6.76, 1st 7.03 = Average 4.666th 6.88 Donnarumma - Sacked November (16th) 6.51, Sacked December (7th) 6.83, 5th 6.91 = Average 9.333th 6.75
For the outfield, the big problems come up where he strays from convention. For instance, he gives 0 weighting to concentration for STs, resulting in this player being rank 40th ST (classed alongside Kane and de Bruyne):
He is limited by his low concentration. I have him ranked 114th alongside Danny Ings at West Ham, rather than Kane at Bayern.
He has Mbappe rated higher than Haaland. You can see on Youtube or my own test results that Haaland does significantly better than Mbappe when Mbappe is put in the same division.
And this is whom he reckons is the 7th best DC in the world:
This is what you get when you weight dribbling at 0, composure at 5, and strength at 50.. or conversely, when you put pace/acc to 100 and don't factor in other attributes properly.
For comparison, this is what my weightings say is the 8th best DC in the world (John Stones was 7th; a little too old to do):
Zouma isn't bad, I have him at 65th. But he's not an absolute top player.. he's a West Ham player.
Yarema said: You can absolutely win PL without a single player having 18 pace or acceleration, as long as their CA is high enough. And that is the entire point. CA isn't worthless, it may not be best distributed but it is very very hard for a 120 CA player to outperform 160. Even with engineered perfect attribute distribution on 120 it's close and we know those don't actually exist ingame.
Not to mention that somehow you claim that under 18 pace you can't win the league and at the same time it's weighting is almost the same as 12 other attributes. So the scoring only works once you reach the minimum thresholds or what? Expand I'm pretty sure I could devise a team of players where they have 18 pace/acc and high CA and yet still finish towards the bottom. But of course it is generally true that a team of high CA players will dominate, but if that is sufficient and viable for you, then why bother with GS ratings at all, and instead just sort by CA? The truth is a lot of CA used is useless, and expensive to boot. Muric did better than ter Stegen, even though ter Stegen has much higher CA and cost/reputation.
The perfect attribute combination to win the English Premier League is 1 CA. The minimum realistic attribute distribution is ~85 CA as shown with my player templates. Obviously real players will have some extra padding on top of that in places, but we have been talking about GKs I favor have CA as low as ~100-110. I have specifically tailored my template to be realistic.
How I found that 18 pace/acc alone is insufficient is through my 1 CA testing. Even 20 pace + 20 acc + 20 drib cannot win.
Here are some quotes from the thread to show you I am not just making this up on the spot now:
GeorgeFloydOverdosed said: Mid-table results were easy with just 20 pace/acc/drib (not sure if I included jump), but it required careful tweaks — closer to Orion’s data than HarvestGreen’s — to push into the top four.
- Pace+Acc alone = 0 points, 0-7 loss typical - Jump+Acc alone = a draw or two, 1-7 loss typical - Drib+Pace+Acc alone = mid-table Expand
But it doesn't take much extra if your player does have 18 pace/acc. Generally speaking, any real player with 18 pace/acc will dominate because his other attributes will all be at least 6-8.
However a 16 pace/acc player is not so straightforward. They do require at least moderate values in certain other attributes to win. The two most obvious are dribbling and jumping reach, but '6' concentration or pressure and he'll be a dud.
With my player templates, I try to leave at least a bit of leeway to account for times when players deficient in a certain key attribute pop up in your top ratings when they really shouldn't (the ter Stegen example explored earlier is a good example of this problem). So for instance, the real minimum for pace/acc to win the Premier League isn't 14, it's 13 from memory or even 12 in certain cases (think Harry Kane). But the requirements for other attributes are so rigid and high with that, so I left pace/acc at 14.
Another example is the less critical an attribute is, the lower I will generally put a threshold on it, even if you can gain an extra position from a single '1' boost. An example of this is agility. Agility '11' or '12' has decent benefit, but it's not strictly necessary. This is an aim for 5th or 3rd debate. Ultimately I keep at the back of my mind always that the more attributes you put high thresholds on, the less player options you end up with.
blackbird said: If thresholds are more important why not use filters instead? Expand Both have downsides. For instance, if you filtered using my template, ter Stegen would be excluded due to his 13 Aerial Reach, and there's also the classic problem of if you filter for more than ~3 attributes no one would qualify.
Obviously the best solution would be some combination of both. A lenient set of filters combined with attribute weightings. But the practical reality I think for most of us is that we're rarely bothered to use filters at all. I think most of us just want a list sorted by rating % to look at.
Yarema said: Bottom line is people are looking for who is a better player right now kind of rating, not who you can train to be better in 5 seasons. You severely undervalue pace and acceleration for example because you argue that it's easy to train it up, which might be true but it takes a lot of time and some luck.
Secondly if you compare 2 players with same "key" attributes, one is 120 CA the other is 160 CA the 2nd one will perform better. Probably even significantly, those 1% here and 1% there add up. Can a team of 120s finish 7th in PL? Sure, but most of people looking for ratings aren't after that. They simply want a scoring system that will tell them player X is better than player Y, and I think you've strayed further and further from the original intention with each iteration. Expand
treath said: I never said Coutouis is not the best player at the start of the game, i said he is the best player and in the rating he lost to Ter Stegen.
Oblak with 20ish reflexes 19 one-one, 18 handling and 15 aerieal reach lost to some dudes in Equador with 120-130ca just because he lost some pace (??????????) due this age.
There is no way he is not a better goalkeeper than dudes with 120-130ca just because they are taller tham him.
When i use genie scout i want to know who is the best players in the filter i selected (like just english players for example), not who is the best cost effective players.
People are misguiding using this rating because of that.
I get that yeah 1 speed is better to train than 5 passing because of Ca cost, i agree with you, but in the rating when you put ZERO evaluate in passing, is like it's useless in the match engine and that's not true at all. Expand I have conducted some tests using the following players:
Arijanet Muric - 128 CA, 69.23%, Burnley Jan Oblak - 171 CA, 69.13%, Real Madrid Gianluigi Donnarumma - 161 CA, 65.84%, Paris SG
In each test I have replaced the GKs at Manchester City with 2 duplicates of the player. First, dropping them into my team of meta players and using Knap tactic that would normally come ~5th:
Muric - 3rd 6.81, Sacked March (10th) 6.67, Sacked Febuary (10th) 6.78 = Average 7.666th 6.75 Oblak - 6th 6.87, 7th 6.76, 1st 7.03 = Average 4.666th 6.88 Donnarumma - Sacked November (16th) 6.51, Sacked December (7th) 6.83, 5th 6.91 = Average 9.333th 6.75
Next, putting them a default Manchester City team and selecting the default 424 gegenpress tactic:
So we can see that in normal play, you wouldn't be able to tell the difference between these three players. When we put them in a struggling team with less variation allowing us to isolate the GK differences more clearly, all three hold up decently, but clearly Oblak has an edge on Muric, and Muric has an edge on Donnarumma.
To clarify just how much of an edge Oblak has, I have also tested the highest rated GK (Marc-Andre ter Stegen, 175 CA, 75.17%, Barcelona) to compare:
ter Stegen - Sacked November (17th) 6.60, Sacked April (11th) 6.64, 5th 6.88, 4th 6.80, 6th 7.05 = Average 8.6th 6.79
Comparing ter Stegen to the template, the two areas he is deficient in are Aerial Reach and Pace. In fact, a close look at Oblak's stats shows he fits the template better. The reason ter Stegen is rated higher is because of some very high values for key mentals and generally higher physicals. As mentioned before, this is the inherent downside of weightings, it can't tell the difference between 1 pace 20 acc and 10 pace 10 acc.
Comparing ter Stegen and Muric, the only advantages Muric has are Aerial Reach (16 > 13) and Jumping Reach (18 > 16). ter Stegen absolutely dominates him in other areas. Yet we see that Muric performs as good or slightly better. For Oblak vs. ter Stegen it is largely the same story; ter Stegen mogs Oblak in everything except Aerial Reach and Handling.
The conclusion I draw is that Aerial Reach has been slightly underweighted.
It is worth noting that if we assessed by passing, first touch, and whatnot, then ter Stegen would be rated even higher above Oblak and Muric.
I know from testing that Aerial Reach is the least forgiving GK attribute, and you can see evidence of me believing this here where I give Aerial Reach the highest rating of 80 in a 'weighted weights' version of my file. You can see that by contrast I give Reflexes only 28 and Jumping Reach 32. So ter Stegen going under the 14 minimum is pretty clearly to me the cause of his poor performance. The problem with the weighted weights is that they simply didn't perform better in comparison testing - it's not the end of the story, I just haven't progressed further with it yet.
Now hopefully that settles the facts of the matter.
I would now like to address the theoretical part.
1) I see ongoing conflation of two separate things. Criticizing basing weights on what gets ~5th-7th instead of 1st is one thing, criticizing that certain attributes are weighted at zero is another. They are two separate things. In regards to weighting useless attributes at zero, I stand firm on this - there is no '1%' gain.. there is 0% gain, or negative gain. Now I hear you, you say 'just suppose it is 1%, what's the harm in giving it a little weight?' - the harm would be that if start weighting all these worthless attributes even a little, you would end up with 36 year olds being recommended as 1st-choice strikers.
2) I did not wake up and decide to aim for 5th place in my making my latest weightings file. As you'd expect, I aimed for 1st. But what I quickly came to realize is that coming 1st is only feasible if all your players have 18+ pace/acc, or otherwise simply spend a billion dollars on standard players that will win without using the speed exploit. Now you could say, well that's how it has to be. But the conclusion I arrived at is that GS ratings are kind of pointless if there aren't any players you can buy or even exist. That 18 pace/acc requirement I mentioned.. it's not just that, you need to pair it with say high drib and jumping reach and a few other things; high pace/acc alone is not enough. So I thought, ok what can we realistically get this pace/acc requirement down to.. and that turned out to be ~14, which so happened to manage ~5th. 13 = relegated, and 17 = no win. You could argue that I should pick 15 for 4th, or 16 for 3rd, that's a fair point of debate. But you are not aware of the nuance of it if you believe that calibrating weightings towards ensuring 5th means they ensure 5th more than they ensure 1st. Here are the parts of the puzzle you are missing:
- Generally speaking, '20' pace is worth no more than '18' pace in the Premier League. If you have 18, you win all games. If you have 20, you win all games. - I have tested every such threshold. As you can see with the GKs above, '18' concentration shows negligible benefit compared to '13'. With some attributes, such as dribbling, it's generally the higher the better (linear). With Aerial Reach, it has a concrete minimum ('14' ) but quickly diminishing returns above that minimum. - If I weight attribute X at '15', that does not mean it is weighted to get 5th. It is weighted to get 1st so long as the cost is not too great. In the end, only a few attributes are weighted below 1st. These are: pace, acc, jump, drib and perhaps another I have forgotten. - If you say 'well let's boost pace, acc, jump, drib to 100 weighting!' then you would end up with key technical and mentals falling out of favor, and the whole optimized structure falling apart. It is more important to get 12 concentration instead of 11 concentration, than 17 pace instead of 16 pace. Obviously this gets too complicated to fully think out, but I have a hunch it's why my 'weighted weightings' did worse than the standard weightings in my comparison test, and hence why it's given me current pause for thought on the matter. - If you say 'to hell with the 'structure'! give me copious amounts of pace, acc, jump and drib at 100 with some passing/tackling slop on top and be done with it.. keithb, give me your ratings file!', then you end up with bloated players on top of the list that fall short in some key regard.
treath said: mate i don't discredit your research, i think it's valid when we talk about cost effective relatively.
but people don't use the ratings for that, they use to find the best players, so when you put in your ratings ZERO for long shot, technique, first touch, etc. It will fall apart. It will not show the best players, but the more cost effective maybe.
Take a look at your picture here from GenieScout, there is a GK from KV Mechelen 112Ca better than Mamardashville 148ca..near as good as Diogo Costa and Verbruggen.
But he is not that good, am i right?
Your rating show that, well forGk with 110-120ca he is the best cost effective in the game but not the best player to play for my team now.
There is no way in hell Courtouis it's not the best Gk in the game, there is no way a Gk from Indepediente Del Valle and Burnley better than Oblak, just because the Aerial Reach and Pace (???) is a little bit better than Oblak (discredit for all other atributes).
So you misguiding people when we talk about find the best players to play for my team now.
But you help a lot, and helped me a lot when we talk about cost effective for training. Expand For many attributes, including all the ones you mention, there is simply no benefit to having them. They are in fact often a net negative because they take up CA, reducing available CA for useful attributes such as pace/acc and slowing development (CA-PA gap strongly influences growth).
That may sound hyperbolic, but consider the following example:
A player has 5 extra passing. +5 passing = Approx +0.96 win rate according to HarvestGreen's data I took a random 140 CA player and gave him +5 passing. His RCA went to 142 CA. So +2 CA for +5 passing. If I give him +1 acceleration instead, his RCA is 143. So +3 CA for +1 acceleration. +1 acceleration = Approx +4.78% win rate according to HarvestGreen's data. If we adjust the CA to match, it's +0.96 (passing) vs +3.186 (acceleration). So extra passing is just losing.
That's HarvestGreen's data.
For my own reference, I use my own data. I change an attribute from 1 to 20 for all players in a team, and if the team has an identical or worse position after a full season, I say that attribute is worthless, end of story, and mark it '1'. If going from 1 to 20 changes the position by 1 place, I put it in my weights - even if it costs a lot of CA. An example of this is strength.
A zero weighting does not mean they are penalized for having the attribute, it just doesn't give them an advantage for it.
Now as for saying Messi is obviously better than Joe Blow from Slough.. I have resigned myself to not making presumptions. First off, I'm not an expert on every great player in the footballing world, and my knowledge is probably a bit outdated. Second, it would obviously be a form of confirmation bias to judge results by name recognition.
The gold standard for me is results. I have compared a few top strikers, testing them in the same league as a control.
A good second option I believe is to see who the top clubs actually start with or buy. You say Courtois is not the best GK in the game. Well, he's the 1st choice GK for Real Madrid, and in a save I checked he averaged 7.21 rating over 5 years with 2 runner-up World Goalkeeper of the Year Awards and 3 third-place awards, which doesn't sound too shabby to me. I guess my question to you is.. on what basis do you know he is not one of the best GKs?
In regards to best players vs CA cost-effective, the current weightings I present do not take CA cost-effectiveness into account. I previously have, if you are familiar with the 'Blended' vs. 'Pure Performance' files. The only way it is currently coming into consideration is as follows:
Suppose I test X attribute '20' finishes 5th. '12' finishes 5th. '11' finishes 6th. '10' finishes 7th. '9' finishes 10th.
My thinking would be that I should choose somewhere between '10' and '12'. If attribute X is very CA costly, or more importantly difficult to find/train, then I will likely choose '11' or possibly '10'. Otherwise I will choose '12'.
How you should actually present your argument therefore is that I should target 1st, not 7th. But I think your typical player is taking Slough to the Champions League, not assembling the World's best XI in Season 1.
I mean honestly, what do you even suppose to propose? I can only imagine you would suggest weighting passing as 5 instead of 0 for example. But then you would have 36 year olds giving Div 2 strikers a run for their money, because their 16s in various dud technicals & mentals would make up for the difference in speed.
and also have tried to emphasize that Unique Nation ID matters a lot - as important as junior coaching or more.
But I've never discussed the randomness factor, let alone put a precise figure on it. And many people just handwave discussion of newgens away with 'it's all dominated by randomness anyway'.
So I think it is time to start determining and stating just how random it is exactly. After a brief thinkle I've put it at 50% and junior coaching at 13% for a comparison, but this is just a starting point for a more precise estimation.
treath said: not mention other positions that infer a player like a winger or midfield with 20 passing 20 first touch 20 long shoot 20 technic and 14 driblling are worst than a player with 20 driblling and 5 on the rest. Expand Passing, first touch, long shots, and technique have negligible benefit or even possibly negative correlation for outfield players.
If you don't want to take my word for it, refer to HarvestGreen's data on this:
You're also not attempting a fair example. No players are going to have 20 passing, first touch, long shots, and technique. Nor is a player going to have 20 dribbling and 5 in those other stats.
treath said: he blindly trust this prem 1.0 ratings that show pace is more importante for a GK than reflexes and his gk are a phisical monster.
the problem is people trust this ratings blindly i made tests with this rating with gsscout and it show goalkeepers with 100 or 110 pa better than great goalkeepers just because they are tall and have enough aerial reach and jumping reach to surpass lack of reflexes, 1-1, etc. Expand If you doubt it, you can edit an English Premier League's team in the editor to match these attributes and see if they come ~7th in combination with Knap tactic as I claim they do. For hiddens, convert the values here (simply divide by 4).
It requires a bit of work, but the difficult legwork is already done. It would take about 30 minutes, and then perhaps another 30 minutes to test.
I have tried to state clearly that while the attribute templates are very reliable (plug those figures in and you *will* get 7th on average and with not too much variation), weightings have the downside of being unable to distinguish between 1 acc/20 pace and 10 acc/10 pace.
You mention 'prem 1.0 ratings that show pace is more important for a GK than reflexes' and 'aerial reach and jumping reach to surpass lack of reflexes, 1vs1', which are examples of this inherent lack of discrimination. I have attempted a weighted weights version in attempt to try and compensate for this problem, however my comparison testing found that it did performed worse than the standard Premier League 1.0 weights under normal playing conditions (8.333 position result average vs 7.333 for standard Premier League 1.0 weights).
The comparison tests included a control of Orion's weights, which averaged 13.166 position, so my weightings are definitely worth using above any random set of values, and I daresay anyone else's weights presented so far.
You say that my weightings 'show goalkeepers with 100 or 110 pa better than great goalkeepers'. This irks me because this is just straight up disinfo. Now maybe what you mean to say is that sometimes a 100 CA goalkeeper can supposedly perform at the highest level, but cmon man, anyone can load up my ratings with the starting database in 3 minutes and see this:
bf3metro said: Thanks for your detailed reply, I really appreciate it.
I genuinely have this feeling that a goalkeeper with poor Reflexes is basically a walking sieve, haha! But at the same time, what you said makes perfect sense, because a goalkeeper isn't supposed to save everything on his own.
I'll post a screenshot so you can tell me what you think. This was my goalkeeper: he conceded 58 goals over the season. I replaced him the following year, but according to the ratings he's still considered the best goalkeeper in my save... despite having only 9 Reflexes. That just feels strange to me.
Anyway, thanks again. I also noticed you mentioned FMSuperScout. I've never used it before. Is it as accurate as Genie Scout when it comes to calculations and visible attributes? I see that FMSuperScout also takes hidden attributes into account, so it's definitely worth testing. I'll have a closer look at it. If you've already done some testing with it, I'd be really interested to hear your thoughts, especially regarding the reliability of its ratings and percentages.
One last thing: would it be possible to talk to you via DM instead of posting publicly here? Do you accept private messages? If not, no worries at all!
Thanks again, and speak soon! Expand The only issue with that player is the 9 reflexes. I would expect that player to do better than 6.7 rating in England Div 3. Could it be that the player has poor hiddens?
It's an inherent downside of weightings that it can't distinguish between (1 pace/20 acc) and (10 pace/10 acc). I think it can be mitigated in various ways though, similar to how there are perceptual tunings available in video encoding to improve the subjective quality.
I have attempted this by weighting the weightings, which you can find here. Here are the weighted weightings for GK, which I haven't posted before:
It's a trade off where determination for GK is hardly taken into account anymore, but on the other hand aerial reach will always get the priority it deserves. It should be an improvement, because it's worse to be deficient in aerial reach than in determination.
But in a controlled test of it, it finished position 8.5 on average, which was worse than 7.333 for the standard Premier League 1.0 file. I will still be going down this route further, but that is where things are at right now. You can try using these weighted weights if you like.
I respond to DMs but obviously I only have the bandwidth for a few people from time to time
Kamas1 said: Does it need to be changed? const META_W = { Pace: 6, Acceleration: 6, JumpingReach: 5.2, Dribbling: 5.2, Balance: 4.8, Concentration: 5.2, Anticipation: 5.2, Determination: 5.6, Agility: 4, Stamina: 5.2, Composure: 4.4, WorkRate: 4.4, Aggression: 4.4, ImportantMatches: 5.2, NaturalFitness: 4.8, Consistency: 5.2, Pressure: 5.2, Professionalism: 5.2, };
And why is the highest value 6 if the scale goes up to 20? Expand Does what need to be changed? Seems correct
My presumption is that the weights are relative to each other rather than absolute. The results appear similar to FM26 Scoring System, so I haven't bothered verifying. I've simply divided by my weightings by 10.
bf3metro said: Hi, first of all, thank you so much for your incredibly detailed reply. I hope I'm not bothering you, but I have a couple more questions that could really help improve my experience with the game.
Regarding the Premier League 1.0 ratings, I feel like the goalkeeper ratings are a bit off. Quite often, average or even poor goalkeepers end up with very high percentages. In fact, the highest-rated player in my save is a goalkeeper with only 9 Reflexes who conceded 58 goals during the season. Do you think there's any way to improve the goalkeeper ratings in the file?
You also mentioned the Genie Scout bug regarding hidden attributes. In your original FM26 ratings file, hidden attributes are taken into account. I'm completely willing to modify the Premier League 1.0 ratings file to include only the hidden attributes, but do you know what weights or values they should be given? Would it simply make sense to import the hidden attribute weights from the original FM26 ratings file into the Premier League 1.0 file?
Thanks again in advance for your reply. What you're doing is genuinely impressive, and I really appreciate the amount of work you've put into it. I'll definitely continue supporting your work in the future. Expand For reference, these are the optimal minimum set of attributes for GK derived from testing:
The proper version has 8s instead of 1s, but this makes it clearer what attributes matter and which don't.
I've had a double check and it seems to be working as intended. The top rated GKs are the ones you'd expect to be (1st GKs from the very top clubs), and if I filter for current rep 5000< or 3000< then the top players even if low in value or have weird looking distribution of attributes nonetheless are getting high ratings and are getting picked up by higher clubs which are strong indications it's on the right track.
Some examples:
Keep in mind that a GK alone is not going to stop goals. The results depend largely on the rest of your team. HarvestGreen concluded that a GK is worth 20% of an outfield player. My impression is that it's quite a bit more than that, but a GK alone certainly won't save a bad team. What I assess by is the improvement or drop in league position.
In regards to Hidden Attributes in Genie Scout, the problem is not that I can't work out what weighting to give them - you can find those weightings here. The problem is that GS does not calculate using those weightings correctly. It uses some complex formula that makes setting them accurately difficult if not impossible. Think of it as two curves that don't match up and therefore can't be integrated together even if you work out what those curves exactly are, this is the challenge I'm facing. The band-aid solution right now is to combine the weights with a basic filter for hiddens (i.e. min 8 pro, pressure, etc.).
Now if you use FM26 you may be in luck because there is now a program that can do what we're looking for. See my post below this.
Looks better than FM26 scoring system on the face of it
For anyone interested in using my weightings with it, what you need to do is the following:
1) Go to install folder 2) Go inside 'app' folder 3) Open 'app.js' with notepad 4) Find 'meta-score (FM-Arena attribute testing)' 5) replace the weightings with the following and save:
66connor66 said: Hi, haven't played in a while... did we figure the best file for ratings so far? Is it different doing a long term save where developing vs looking for instant success? Expand bf3metro said: Hi @GeorgeFloydOverdosed,
I noticed that some people have been criticizing your ratings file. I've been using your FM26 Genie Scout Ratings file since almost the beginning of FM26. Then I stopped playing around February and only came back in June. Since then, I've still been using the FM26 Genie Scout Ratings file, but I recently saw that you now recommend using the Premier League 1.0 Genie Scout Ratings file instead.
First of all, congratulations on all the work you've put into this. It's rare to see this level of dedication and effort.
I have three questions:
Are you still actively improving and refining your ratings file? If you had to estimate it, how reliable and accurate do you think the Premier League 1.0 ratings are, as a percentage? Finally, which file should we actually be using: the FM26 Genie Scout Ratings file or the Premier League 1.0 Genie Scout Ratings file?
That's all! I hope you'll have the time to reply. Thanks again for all your hard work, and have a great day! Expand I recommend the Premier League 1.0 file, while using filtering for hiddens. It is designed to 'save' you in your first season, but I think this is the right balance and does not impair min-maxxing.
It would be less accurate for FM26 most likely, but I suspect the difference would be small.
This question gets asked a lot, so I'll briefly explain why. Take long shots - it's ~3x importance in FM26 supposedly, ok that's 3x0 = 0 for me.. maybe that ends up being a 10 minimum for the attribute, which is a minor difference since most players will have ~6-8+ by default anyway. Going by HarvestGreen's data, there don't seem to be any paradigm-shifting changes, and the thing is, if you suppose balance now has say a '16' requirement, that's highly unlikely to be the case because very few players would have 16 balance in the game. If I had to make a rough guess, the differences between FM24 and FM26 might be a player going from rank 57th > 68th say. Not nothing, but nothing huge either.
The ratings file will continue to be improved, but there is no set timetable.
The main goals I would like to achieve are:
1) Re-integrate hidden stats again. Either with a program that comes along that can do so, or by essentially backporting the rankings FM26 scoring system gives and fiddling with the numbers until it largely fits the mold. 2) Create a FM26 version. Waiting until FM26 is free on Epic, or otherwise settle for working with HarvestGreen's FM26 data. It's not unfeasible, it's just a lot of effort for little reward whichever way I do it. 3) Further improve attribute value accuracy and weightings accuracy. Ideally I want to be able to say this is player is 5th, that player is 6th, etc. with high confidence.
If I had to put my impression of the accuracy of my GS rating files as a % based purely on intuition, I would put it like this:
88% - Premier League 1.0 (in FM24) 83% - Premier League 1.0 (in FM26) 82% - FM24 Pure Performance 80% - FM26 Pure Performance 75% - If you just plugged in pace/acc at 100 weight and added a few others as keithb suggests (basically every simple interpretation of HarvestGreen's data) 70% - File based on Orion's data 56% - Default Genie Scout
I can't say I even trust my own intuition on this though. I know the attribute figures are very close to accurate, translation to GS weights is another matter. Here is what the first control testing showed (number is average finishing position in Premier League):
6.5 - FM24 Pure Performance 7.333 - Premier League 1.0 13.166 - Orion's weights
So I think there's pretty clear evidence it works well, but my intuition about Premier League 1.0 being an improvement over the previous version may be wrong. I still think Premier League 1.0 is better, because it's based on solid data (my extensive attribute testing). It could just be that beyond a certain point, further precision is superfluous (i.e. if one file gives 73.54% and the other 72.88%, you're going to end up picking the same player anyway). I anticipated this and measured overlap of players. From memory the overlap was only around 50% for all three, so there is still substantial variation going on, which was actually quite surprising to me.
BrushlessPlaymaker said: Still on this topic: Like I said, I prefer may saves with generated players rather than real ones and, since you mentioned it, I'm noticing a lot of young and potentially good player with very low jumping. Is it worth it trying to train that up? I've switched a couple of my own players with decent pace/Acc from quickness to Strength Individual training (which supposedly improves strength and jumping reach) but I'm yet to notice improvements. Expand I assessed alternative training focuses and found it to be a bad idea. But if a player is particularly deficient in a certain key attribute, it is probably worthwhile training them up to ~8 in that attribute.
With jumping reach, would it be worth say training up further to 12 instead of pace 16 > 18? It's hard to say. Personally I would just get the high jump reach player to begin with, because it will take time anyway, and jumping reach is particularly unforgiving if you lack it. You can get away with 14 pace; you can't get away with DCs that have 14 jump.
This set of guidelines for Italian FM volunteer researchers for FM20 is something of a goldmine when you plug it into Google Translate. Most of the FM info dished out by SI staff is pure BS, but here we have some tidbits of proper info about how it works under the hood.
There are two interesting things in it on height specifically:
1) There are conversation tables for height to jumping reach (one for GK, one for outfield). 2) Pace - 'Very short players cannot have a very high Speed ​​value (at most they will have a high acceleration value)'
This means you can use minimum height to help filter players in the starting database, as the researchers have to follow these strict guidelines in the player data they're putting in.
We can deduce that a player under 1.95m won't ever reach 20 jump even with training. 1.87m would be the minimum to eventually reach 17 jump (ideal minimum for DC). 1.82m is the bare minimum for a DC. Under 1.72m would be too deficient in jumping reach in any position, and should probably be avoided.
We are told that short players are pace-constrained to boot as well.
Now that is the starting database, but if you think about it, they've probably programmed the game to generate new players according to the same guidelines.
Fraudiola01 said: @GeorgeFloydOverdosed With regards to the Match Engines on FM 24, what is the current consensus of the community? Does the FM Match Lab ME or other ME mods help with the game being more realistic/a wider variety of tactics being viable? Or is it just placebo? Expand I wouldn't know as I haven't used those mods, because I like playing the same game mostly everyone else is playing.
I suspect that problems may arise if you make meaningful changes in isolation. For instance if you somehow reduced the importance of physicals to winning, that would probably mess up the transfers side of the game.
Panneton0 said: So it is not a rating file that truly shows the "statistically best" player, but the "statistically most promising, considering pace/acc will be trained later". For instance, Haaland is way above Mbappe because Haaland could gain the 1-2 pace/acc that Mbappe has on him, while Mbappe could never gain the 10 jumping reach that Haaland has on him, even with perfect CA reallocation through Harvest's training regime.
So if I'm using the file to figure out the "best XI" of my team for a given tactic at a given date, your earlier "FM26 GS file" would make more sense, but for long-game recruiting purposes / understanding best promising youngster from my academy, your new "Premier League 1.0" file is prbly better? Expand I had to go back over my posts and have a few puffs of the pipe on this one.
I misremembered the reasoning for pace/acc being low. I've been away from it all for about a month.
The goal I had started with in my 3rd iteration was this: Get the pace/acc requirement as low as possible, because getting a full team of 18 pace/acc players your first season in the Premier League isn't realistic. It turned out that ~14-15 was feasible in combination with certain levels of other attributes - it wouldn't win the Premier League, but it would guarantee avoiding relegation and even potentially achieve a continental qualification finish.
Now it's true that I also adjusted the values for training later, however pace/acc remained at 15. And yes, the idea is to train them to ~18, but this is beyond what is strictly necessary to place highly. The idea is that growing pace/acc to 18+ will ensure finishing 1st in the 2nd season onwards.
The history of this can get very confusing, but I think this timeline makes it pretty simple:
Dec 15th - First GS ratings file released. May 7th - 'FM24 Pure Performance' is released. This is the "best XI" ratings file you have in mind. It is mostly based on HarvestGreen's data. May 14th - LightningFlik posts the discrepancy that leads me to finding Genie Scout is bugged. June 12th - 'Premier League 1.0' is released, which is mostly based on my own testing. June 21st - 'Premier League 1.0' tested in comparison to 'FM24 Pure Performance', with the latter edited to avoid causing the GS bugs. Position results were 7.333 and 6.5 respectively, with Orion's weights as a control being 13.166. I had also created another version of 'Premier League 1.0' which refined it according to the essentiality of the attributes and its result was 8.333. I think I planned for it to become 'Premier League 1.1' but either the result dissuaded me or I forgot about it and took a break.
So I recommend just using the 'Premier League 1.0' file.
Panneton0 said: Not sure what the point of that attitude is, but anyways.
I agree that speed and acceleration might be underweighted in the premier league GS file. Jumping reach having high weight is an interesting take, but can you remind me why it has actually higher weight that pace/acc? I feel like nothing should be higher, according to actual testing. Not saying that it's a bad idea, but I wonder the rational behind that choice.
Also, any pointers on how (according to your file, or of general understanding of positional proficiency) we should interpret GS ratings of sub-20 position for a player as per my messages before the interruption? Expand The reason why pace/acc is now seemingly under-weighted is because it is expected that you will train these up using the quickness focus in training. I have based the adjustment on real training results, not just the optimal possible. They still remain the highest weighted attributes.
Jumping Reach is only weighted higher in 2 positions, DC and ST target man, but generally it is up there with pace and acc. I can't remember all the nuances around it, but here are the key ones I can recall:
1) Jumping Reach is critical on DC and has a high minimum requirement. Something like the difference between 17 and 13 is 5th > Relegated say (you can probably find these results somewhere earlier in the thread). 2) It's not that you just need one or two tall players to do the job, and that having your MC be a bit taller makes no difference. Increases even at mid levels across the team result in significantly better performance. 3) Jumping Reach is harder to train than pace/acc, at least when we assume quickness focus is being used.
I read your post on position proficiency, but don't have anything to contribute to it really. I do make sure to search for players with only green position proficiency in Genie Scout. I've dabbled with position proficiency in various ways, but I don't think to find performance level at 1 proficiency say, and I'm too hazy on it now anyway.
What I will say is that I think certain dual positions that have no CA penalty (i.e. AML/AMR) are beneficial in a practical sense.
I vaguely recall experimenting with playing optimal players out of position to try and cut the CA cost and that it failed miserably. So it does have a big impact.
As I was thinking about this, it was lurking at the back of my mind that I'm sure someone had tested proficiency and I had seen the results once. It took a while, but I think I've found what I was recalling. HarvestGreen tested it:
He also did FM26 OOP proficiency and found the win rate difference between 4 OOP proficiency and 20 OOP proficiency to be 40.9% > 46.4%. Link to thread.
I attempted to address this with weighting the weights, but it produced worse results, or at least no better.
I've tried comparative deduction by comparing several players match ratings & goal tallies with their attributes. Some key problems with this approach:
- Match ratings have been demonstrated to favor presence of technical/mental ability over actual wins
- Goals scored can simply reflect finishing, and poor finishing has minimal effect on the team win rate (it simply means someone else with higher finishing in the team scores the goals)
- I actually managed to get the players to 'line up' with the ratings/goals a lot better, but it took a very strange set of GS weightings to get there. Either I'm on the wrong track, or even if it's right, I would be building a house without being able to know the foundations.
I concluded that what's actually needed is more result-based data, but on individual players. It's more tedious, and will require a fair amount of samples for good accuracy, but I think it's the best approach.
Method
I use my meta team at Man City, but replace a position with duplicates of a particular player. If I'm testing ST, I remove proficiency in other positions so as not to mess up the results, as what I'm most interested in is deducing what constitutes the precise ranking of STs. Full detail is used for all first team competitions.
Results
Here's some data I've collected so far (ranked by team performance with player as ST):
Erling Haaland đź‘Ť - 1st, 1st, 4th = 2nd position, 1.263 goals/match, 7.93 rating
Kylian Mbappe đź’© - 5th, 1st, 3rd = 3rd position, 0.708 goals/match, 7.54 rating
Daizen Maeda - 2nd, 3rd, 4th = 3rd position, 0.533 goals/match, 7.28 rating
Harry Kane - 5th, 2nd = 3.5th position, 0.429 goals/match, 7.25 rating
Victor Osimhen - 3rd, 3rd, 7th = 4.333th position, 0.699 goals/match, 7.27 rating
Adam Le Fondre - 8th, 11th (sacked), 5th = 8th position, 0.416 goals/match, 6.82 rating
The reason I picked Le Fondre to include is because I saw him compared in a Youtube video, and even though he's a 36-year-old with 9 pace/acc, he managed the results you see here. A takeaway from this for me is that pace/acc isn't strictly necessary. Using the comparative deduction I mentioned before, I found that anticipation and composure are more critical than pace/acc.. as I said, these are strange conclusions, so rather than make the facts fit the theory, I'm going to be getting the facts straight first and then seeing what theory fits the data.
I'll probably do about 5-6 samples each of ~10 players for ST and then see if I can deduce consistent patterns to create accurate weightings.
If i still win, but i wanna find a keeper that actually saves, what i shall look for
Rn Chevalier looks good for me, best keeper for me yet, but he's old in my save, so i wanna find or make a good regen, waht to look for?
I may have a look into it later, but I suspect that it's fundamentally your tactic and overall level of quality of your players, not the attributes of your GK, that determine the number of goals he concedes.
We can influence who scores goals in a team, but that's because at the end of the day a goal is going to be scored. But you can't reduce the number of goals your GK concedes, because he's the only one who can concede them.
Cuz what i've seen on my long real save, as example, Courtouis were bad (with my high def line at least)
Yeah i won alot, and gd was insane, but he's avg rating was like 6.8 or something like that haha
To clarify, my template, and therefore the weightings, are based on team position at end of season, not match ratings.
Interesting that you give all hidden attributes 0 weighting even though you claim Genie Scout isn't bugged.
According to those weights, Donnarumma is ranked 9th, while Muric is 202nd.
Meanwhile, actual Premier League test results in reality:
Muric - 3rd 6.81, Sacked March (10th) 6.67, Sacked Febuary (10th) 6.78 = Average 7.666th 6.75
Oblak - 6th 6.87, 7th 6.76, 1st 7.03 = Average 4.666th 6.88
Donnarumma - Sacked November (16th) 6.51, Sacked December (7th) 6.83, 5th 6.91 = Average 9.333th 6.75
For the outfield, the big problems come up where he strays from convention. For instance, he gives 0 weighting to concentration for STs, resulting in this player being rank 40th ST (classed alongside Kane and de Bruyne):
He is limited by his low concentration. I have him ranked 114th alongside Danny Ings at West Ham, rather than Kane at Bayern.
He has Mbappe rated higher than Haaland. You can see on Youtube or my own test results that Haaland does significantly better than Mbappe when Mbappe is put in the same division.
And this is whom he reckons is the 7th best DC in the world:
This is what you get when you weight dribbling at 0, composure at 5, and strength at 50.. or conversely, when you put pace/acc to 100 and don't factor in other attributes properly.
For comparison, this is what my weightings say is the 8th best DC in the world (John Stones was 7th; a little too old to do):
Zouma isn't bad, I have him at 65th. But he's not an absolute top player.. he's a West Ham player.
Not to mention that somehow you claim that under 18 pace you can't win the league and at the same time it's weighting is almost the same as 12 other attributes. So the scoring only works once you reach the minimum thresholds or what?
I'm pretty sure I could devise a team of players where they have 18 pace/acc and high CA and yet still finish towards the bottom. But of course it is generally true that a team of high CA players will dominate, but if that is sufficient and viable for you, then why bother with GS ratings at all, and instead just sort by CA? The truth is a lot of CA used is useless, and expensive to boot. Muric did better than ter Stegen, even though ter Stegen has much higher CA and cost/reputation.
The perfect attribute combination to win the English Premier League is 1 CA. The minimum realistic attribute distribution is ~85 CA as shown with my player templates. Obviously real players will have some extra padding on top of that in places, but we have been talking about GKs I favor have CA as low as ~100-110. I have specifically tailored my template to be realistic.
How I found that 18 pace/acc alone is insufficient is through my 1 CA testing. Even 20 pace + 20 acc + 20 drib cannot win.
Here are some quotes from the thread to show you I am not just making this up on the spot now:
GeorgeFloydOverdosed said: Mid-table results were easy with just 20 pace/acc/drib (not sure if I included jump), but it required careful tweaks — closer to Orion’s data than HarvestGreen’s — to push into the top four.
- Pace+Acc alone = 0 points, 0-7 loss typical
- Jump+Acc alone = a draw or two, 1-7 loss typical
- Drib+Pace+Acc alone = mid-table
But it doesn't take much extra if your player does have 18 pace/acc. Generally speaking, any real player with 18 pace/acc will dominate because his other attributes will all be at least 6-8.
However a 16 pace/acc player is not so straightforward. They do require at least moderate values in certain other attributes to win. The two most obvious are dribbling and jumping reach, but '6' concentration or pressure and he'll be a dud.
With my player templates, I try to leave at least a bit of leeway to account for times when players deficient in a certain key attribute pop up in your top ratings when they really shouldn't (the ter Stegen example explored earlier is a good example of this problem). So for instance, the real minimum for pace/acc to win the Premier League isn't 14, it's 13 from memory or even 12 in certain cases (think Harry Kane). But the requirements for other attributes are so rigid and high with that, so I left pace/acc at 14.
Another example is the less critical an attribute is, the lower I will generally put a threshold on it, even if you can gain an extra position from a single '1' boost. An example of this is agility. Agility '11' or '12' has decent benefit, but it's not strictly necessary. This is an aim for 5th or 3rd debate. Ultimately I keep at the back of my mind always that the more attributes you put high thresholds on, the less player options you end up with.
Both have downsides. For instance, if you filtered using my template, ter Stegen would be excluded due to his 13 Aerial Reach, and there's also the classic problem of if you filter for more than ~3 attributes no one would qualify.
Obviously the best solution would be some combination of both. A lenient set of filters combined with attribute weightings. But the practical reality I think for most of us is that we're rarely bothered to use filters at all. I think most of us just want a list sorted by rating % to look at.
Secondly if you compare 2 players with same "key" attributes, one is 120 CA the other is 160 CA the 2nd one will perform better. Probably even significantly, those 1% here and 1% there add up. Can a team of 120s finish 7th in PL? Sure, but most of people looking for ratings aren't after that. They simply want a scoring system that will tell them player X is better than player Y, and I think you've strayed further and further from the original intention with each iteration.
treath said: I never said Coutouis is not the best player at the start of the game, i said he is the best player and in the rating he lost to Ter Stegen.
Oblak with 20ish reflexes 19 one-one, 18 handling and 15 aerieal reach lost to some dudes in Equador with 120-130ca just because he lost some pace (??????????) due this age.
There is no way he is not a better goalkeeper than dudes with 120-130ca just because they are taller tham him.
When i use genie scout i want to know who is the best players in the filter i selected (like just english players for example), not who is the best cost effective players.
People are misguiding using this rating because of that.
I get that yeah 1 speed is better to train than 5 passing because of Ca cost, i agree with you, but in the rating when you put ZERO evaluate in passing, is like it's useless in the match engine and that's not true at all.
I have conducted some tests using the following players:
Arijanet Muric - 128 CA, 69.23%, Burnley
Jan Oblak - 171 CA, 69.13%, Real Madrid
Gianluigi Donnarumma - 161 CA, 65.84%, Paris SG
In each test I have replaced the GKs at Manchester City with 2 duplicates of the player. First, dropping them into my team of meta players and using Knap tactic that would normally come ~5th:
Muric - 3rd 6.81, Sacked March (10th) 6.67, Sacked Febuary (10th) 6.78 = Average 7.666th 6.75
Oblak - 6th 6.87, 7th 6.76, 1st 7.03 = Average 4.666th 6.88
Donnarumma - Sacked November (16th) 6.51, Sacked December (7th) 6.83, 5th 6.91 = Average 9.333th 6.75
Next, putting them a default Manchester City team and selecting the default 424 gegenpress tactic:
Jan Oblak - 1st 6.99, 2nd 7.12
Muric - 2nd 7.00, 1st 6.94
Donnarumma - 1st 7.07, 1st 6.97
So we can see that in normal play, you wouldn't be able to tell the difference between these three players. When we put them in a struggling team with less variation allowing us to isolate the GK differences more clearly, all three hold up decently, but clearly Oblak has an edge on Muric, and Muric has an edge on Donnarumma.
To clarify just how much of an edge Oblak has, I have also tested the highest rated GK (Marc-Andre ter Stegen, 175 CA, 75.17%, Barcelona) to compare:
ter Stegen - Sacked November (17th) 6.60, Sacked April (11th) 6.64, 5th 6.88, 4th 6.80, 6th 7.05 = Average 8.6th 6.79
Comparing ter Stegen to the template, the two areas he is deficient in are Aerial Reach and Pace. In fact, a close look at Oblak's stats shows he fits the template better. The reason ter Stegen is rated higher is because of some very high values for key mentals and generally higher physicals. As mentioned before, this is the inherent downside of weightings, it can't tell the difference between 1 pace 20 acc and 10 pace 10 acc.
Comparing ter Stegen and Muric, the only advantages Muric has are Aerial Reach (16 > 13) and Jumping Reach (18 > 16). ter Stegen absolutely dominates him in other areas. Yet we see that Muric performs as good or slightly better. For Oblak vs. ter Stegen it is largely the same story; ter Stegen mogs Oblak in everything except Aerial Reach and Handling.
The conclusion I draw is that Aerial Reach has been slightly underweighted.
It is worth noting that if we assessed by passing, first touch, and whatnot, then ter Stegen would be rated even higher above Oblak and Muric.
I know from testing that Aerial Reach is the least forgiving GK attribute, and you can see evidence of me believing this here where I give Aerial Reach the highest rating of 80 in a 'weighted weights' version of my file. You can see that by contrast I give Reflexes only 28 and Jumping Reach 32. So ter Stegen going under the 14 minimum is pretty clearly to me the cause of his poor performance. The problem with the weighted weights is that they simply didn't perform better in comparison testing - it's not the end of the story, I just haven't progressed further with it yet.
Now hopefully that settles the facts of the matter.
I would now like to address the theoretical part.
1) I see ongoing conflation of two separate things. Criticizing basing weights on what gets ~5th-7th instead of 1st is one thing, criticizing that certain attributes are weighted at zero is another. They are two separate things. In regards to weighting useless attributes at zero, I stand firm on this - there is no '1%' gain.. there is 0% gain, or negative gain. Now I hear you, you say 'just suppose it is 1%, what's the harm in giving it a little weight?' - the harm would be that if start weighting all these worthless attributes even a little, you would end up with 36 year olds being recommended as 1st-choice strikers.
2) I did not wake up and decide to aim for 5th place in my making my latest weightings file. As you'd expect, I aimed for 1st. But what I quickly came to realize is that coming 1st is only feasible if all your players have 18+ pace/acc, or otherwise simply spend a billion dollars on standard players that will win without using the speed exploit. Now you could say, well that's how it has to be. But the conclusion I arrived at is that GS ratings are kind of pointless if there aren't any players you can buy or even exist. That 18 pace/acc requirement I mentioned.. it's not just that, you need to pair it with say high drib and jumping reach and a few other things; high pace/acc alone is not enough. So I thought, ok what can we realistically get this pace/acc requirement down to.. and that turned out to be ~14, which so happened to manage ~5th. 13 = relegated, and 17 = no win. You could argue that I should pick 15 for 4th, or 16 for 3rd, that's a fair point of debate. But you are not aware of the nuance of it if you believe that calibrating weightings towards ensuring 5th means they ensure 5th more than they ensure 1st. Here are the parts of the puzzle you are missing:
- Generally speaking, '20' pace is worth no more than '18' pace in the Premier League. If you have 18, you win all games. If you have 20, you win all games.
- I have tested every such threshold. As you can see with the GKs above, '18' concentration shows negligible benefit compared to '13'. With some attributes, such as dribbling, it's generally the higher the better (linear). With Aerial Reach, it has a concrete minimum ('14' ) but quickly diminishing returns above that minimum.
- If I weight attribute X at '15', that does not mean it is weighted to get 5th. It is weighted to get 1st so long as the cost is not too great. In the end, only a few attributes are weighted below 1st. These are: pace, acc, jump, drib and perhaps another I have forgotten.
- If you say 'well let's boost pace, acc, jump, drib to 100 weighting!' then you would end up with key technical and mentals falling out of favor, and the whole optimized structure falling apart. It is more important to get 12 concentration instead of 11 concentration, than 17 pace instead of 16 pace. Obviously this gets too complicated to fully think out, but I have a hunch it's why my 'weighted weightings' did worse than the standard weightings in my comparison test, and hence why it's given me current pause for thought on the matter.
- If you say 'to hell with the 'structure'! give me copious amounts of pace, acc, jump and drib at 100 with some passing/tackling slop on top and be done with it.. keithb, give me your ratings file!', then you end up with bloated players on top of the list that fall short in some key regard.
but people don't use the ratings for that, they use to find the best players, so when you put in your ratings ZERO for long shot, technique, first touch, etc. It will fall apart. It will not show the best players, but the more cost effective maybe.
Take a look at your picture here from GenieScout, there is a GK from KV Mechelen 112Ca better than Mamardashville 148ca..near as good as Diogo Costa and Verbruggen.
But he is not that good, am i right?
Your rating show that, well forGk with 110-120ca he is the best cost effective in the game but not the best player to play for my team now.
There is no way in hell Courtouis it's not the best Gk in the game, there is no way a Gk from Indepediente Del Valle and Burnley better than Oblak, just because the Aerial Reach and Pace (???) is a little bit better than Oblak (discredit for all other atributes).
So you misguiding people when we talk about find the best players to play for my team now.
But you help a lot, and helped me a lot when we talk about cost effective for training.
For many attributes, including all the ones you mention, there is simply no benefit to having them. They are in fact often a net negative because they take up CA, reducing available CA for useful attributes such as pace/acc and slowing development (CA-PA gap strongly influences growth).
That may sound hyperbolic, but consider the following example:
A player has 5 extra passing.
+5 passing = Approx +0.96 win rate according to HarvestGreen's data
I took a random 140 CA player and gave him +5 passing. His RCA went to 142 CA. So +2 CA for +5 passing.
If I give him +1 acceleration instead, his RCA is 143. So +3 CA for +1 acceleration.
+1 acceleration = Approx +4.78% win rate according to HarvestGreen's data.
If we adjust the CA to match, it's +0.96 (passing) vs +3.186 (acceleration).
So extra passing is just losing.
That's HarvestGreen's data.
For my own reference, I use my own data. I change an attribute from 1 to 20 for all players in a team, and if the team has an identical or worse position after a full season, I say that attribute is worthless, end of story, and mark it '1'. If going from 1 to 20 changes the position by 1 place, I put it in my weights - even if it costs a lot of CA. An example of this is strength.
A zero weighting does not mean they are penalized for having the attribute, it just doesn't give them an advantage for it.
Now as for saying Messi is obviously better than Joe Blow from Slough.. I have resigned myself to not making presumptions. First off, I'm not an expert on every great player in the footballing world, and my knowledge is probably a bit outdated. Second, it would obviously be a form of confirmation bias to judge results by name recognition.
The gold standard for me is results. I have compared a few top strikers, testing them in the same league as a control.
A good second option I believe is to see who the top clubs actually start with or buy. You say Courtois is not the best GK in the game. Well, he's the 1st choice GK for Real Madrid, and in a save I checked he averaged 7.21 rating over 5 years with 2 runner-up World Goalkeeper of the Year Awards and 3 third-place awards, which doesn't sound too shabby to me. I guess my question to you is.. on what basis do you know he is not one of the best GKs?
In regards to best players vs CA cost-effective, the current weightings I present do not take CA cost-effectiveness into account. I previously have, if you are familiar with the 'Blended' vs. 'Pure Performance' files. The only way it is currently coming into consideration is as follows:
Suppose I test X attribute
'20' finishes 5th. '12' finishes 5th. '11' finishes 6th. '10' finishes 7th. '9' finishes 10th.
My thinking would be that I should choose somewhere between '10' and '12'. If attribute X is very CA costly, or more importantly difficult to find/train, then I will likely choose '11' or possibly '10'. Otherwise I will choose '12'.
How you should actually present your argument therefore is that I should target 1st, not 7th. But I think your typical player is taking Slough to the Champions League, not assembling the World's best XI in Season 1.
I mean honestly, what do you even suppose to propose? I can only imagine you would suggest weighting passing as 5 instead of 0 for example. But then you would have 36 year olds giving Div 2 strikers a run for their money, because their 16s in various dud technicals & mentals would make up for the difference in speed.
Initially I explained things in more extensive and precise detail, but realized it would be too burdensome to read through.
It's still fairly long and could do with another touch up soon, but it has new sections and new info.
For instance I have previously characterized newgen PA as essentially:
40% Junior Coaching
25% Youth Recruitment
25% Nation Youth Rating
~10% Game Importance
0% Youth Facilities
and also have tried to emphasize that Unique Nation ID matters a lot - as important as junior coaching or more.
But I've never discussed the randomness factor, let alone put a precise figure on it. And many people just handwave discussion of newgens away with 'it's all dominated by randomness anyway'.
So I think it is time to start determining and stating just how random it is exactly. After a brief thinkle I've put it at 50% and junior coaching at 13% for a comparison, but this is just a starting point for a more precise estimation.
Passing, first touch, long shots, and technique have negligible benefit or even possibly negative correlation for outfield players.
If you don't want to take my word for it, refer to HarvestGreen's data on this:
You're also not attempting a fair example. No players are going to have 20 passing, first touch, long shots, and technique. Nor is a player going to have 20 dribbling and 5 in those other stats.
the problem is people trust this ratings blindly i made tests with this rating with gsscout and it show goalkeepers with 100 or 110 pa better than great goalkeepers just because they are tall and have enough aerial reach and jumping reach to surpass lack of reflexes, 1-1, etc.
If you doubt it, you can edit an English Premier League's team in the editor to match these attributes and see if they come ~7th in combination with Knap tactic as I claim they do. For hiddens, convert the values here (simply divide by 4).
It requires a bit of work, but the difficult legwork is already done. It would take about 30 minutes, and then perhaps another 30 minutes to test.
I have tried to state clearly that while the attribute templates are very reliable (plug those figures in and you *will* get 7th on average and with not too much variation), weightings have the downside of being unable to distinguish between 1 acc/20 pace and 10 acc/10 pace.
You mention 'prem 1.0 ratings that show pace is more important for a GK than reflexes' and 'aerial reach and jumping reach to surpass lack of reflexes, 1vs1', which are examples of this inherent lack of discrimination. I have attempted a weighted weights version in attempt to try and compensate for this problem, however my comparison testing found that it did performed worse than the standard Premier League 1.0 weights under normal playing conditions (8.333 position result average vs 7.333 for standard Premier League 1.0 weights).
The comparison tests included a control of Orion's weights, which averaged 13.166 position, so my weightings are definitely worth using above any random set of values, and I daresay anyone else's weights presented so far.
You say that my weightings 'show goalkeepers with 100 or 110 pa better than great goalkeepers'. This irks me because this is just straight up disinfo. Now maybe what you mean to say is that sometimes a 100 CA goalkeeper can supposedly perform at the highest level, but cmon man, anyone can load up my ratings with the starting database in 3 minutes and see this:
I genuinely have this feeling that a goalkeeper with poor Reflexes is basically a walking sieve, haha! But at the same time, what you said makes perfect sense, because a goalkeeper isn't supposed to save everything on his own.
I'll post a screenshot so you can tell me what you think. This was my goalkeeper: he conceded 58 goals over the season. I replaced him the following year, but according to the ratings he's still considered the best goalkeeper in my save... despite having only 9 Reflexes. That just feels strange to me.
Anyway, thanks again. I also noticed you mentioned FMSuperScout. I've never used it before. Is it as accurate as Genie Scout when it comes to calculations and visible attributes? I see that FMSuperScout also takes hidden attributes into account, so it's definitely worth testing. I'll have a closer look at it. If you've already done some testing with it, I'd be really interested to hear your thoughts, especially regarding the reliability of its ratings and percentages.
One last thing: would it be possible to talk to you via DM instead of posting publicly here? Do you accept private messages? If not, no worries at all!
Thanks again, and speak soon!
The only issue with that player is the 9 reflexes. I would expect that player to do better than 6.7 rating in England Div 3. Could it be that the player has poor hiddens?
It's an inherent downside of weightings that it can't distinguish between (1 pace/20 acc) and (10 pace/10 acc). I think it can be mitigated in various ways though, similar to how there are perceptual tunings available in video encoding to improve the subjective quality.
I have attempted this by weighting the weightings, which you can find here. Here are the weighted weightings for GK, which I haven't posted before:
It's a trade off where determination for GK is hardly taken into account anymore, but on the other hand aerial reach will always get the priority it deserves. It should be an improvement, because it's worse to be deficient in aerial reach than in determination.
But in a controlled test of it, it finished position 8.5 on average, which was worse than 7.333 for the standard Premier League 1.0 file. I will still be going down this route further, but that is where things are at right now. You can try using these weighted weights if you like.
I respond to DMs but obviously I only have the bandwidth for a few people from time to time
Pace: 6, Acceleration: 6, JumpingReach: 5.2, Dribbling: 5.2, Balance: 4.8, Concentration: 5.2, Anticipation: 5.2, Determination: 5.6, Agility: 4, Stamina: 5.2, Composure: 4.4, WorkRate: 4.4, Aggression: 4.4, ImportantMatches: 5.2, NaturalFitness: 4.8, Consistency: 5.2, Pressure: 5.2, Professionalism: 5.2,
};
And why is the highest value 6 if the scale goes up to 20?
Does what need to be changed? Seems correct
My presumption is that the weights are relative to each other rather than absolute. The results appear similar to FM26 Scoring System, so I haven't bothered verifying. I've simply divided by my weightings by 10.
You can find my instructions on how to use my weights with the program here.
Regarding the Premier League 1.0 ratings, I feel like the goalkeeper ratings are a bit off. Quite often, average or even poor goalkeepers end up with very high percentages. In fact, the highest-rated player in my save is a goalkeeper with only 9 Reflexes who conceded 58 goals during the season. Do you think there's any way to improve the goalkeeper ratings in the file?
You also mentioned the Genie Scout bug regarding hidden attributes. In your original FM26 ratings file, hidden attributes are taken into account. I'm completely willing to modify the Premier League 1.0 ratings file to include only the hidden attributes, but do you know what weights or values they should be given? Would it simply make sense to import the hidden attribute weights from the original FM26 ratings file into the Premier League 1.0 file?
Thanks again in advance for your reply. What you're doing is genuinely impressive, and I really appreciate the amount of work you've put into it. I'll definitely continue supporting your work in the future.
For reference, these are the optimal minimum set of attributes for GK derived from testing:
The proper version has 8s instead of 1s, but this makes it clearer what attributes matter and which don't.
I've had a double check and it seems to be working as intended. The top rated GKs are the ones you'd expect to be (1st GKs from the very top clubs), and if I filter for current rep 5000< or 3000< then the top players even if low in value or have weird looking distribution of attributes nonetheless are getting high ratings and are getting picked up by higher clubs which are strong indications it's on the right track.
Some examples:
Keep in mind that a GK alone is not going to stop goals. The results depend largely on the rest of your team. HarvestGreen concluded that a GK is worth 20% of an outfield player. My impression is that it's quite a bit more than that, but a GK alone certainly won't save a bad team. What I assess by is the improvement or drop in league position.
In regards to Hidden Attributes in Genie Scout, the problem is not that I can't work out what weighting to give them - you can find those weightings here. The problem is that GS does not calculate using those weightings correctly. It uses some complex formula that makes setting them accurately difficult if not impossible. Think of it as two curves that don't match up and therefore can't be integrated together even if you work out what those curves exactly are, this is the challenge I'm facing. The band-aid solution right now is to combine the weights with a basic filter for hiddens (i.e. min 8 pro, pressure, etc.).
Now if you use FM26 you may be in luck because there is now a program that can do what we're looking for. See my post below this.
For anyone interested in using my weightings with it, what you need to do is the following:
1) Go to install folder
2) Go inside 'app' folder
3) Open 'app.js' with notepad
4) Find 'meta-score (FM-Arena attribute testing)'
5) replace the weightings with the following and save:
Pace: 6, Acceleration: 6, JumpingReach: 5.2, Dribbling: 5.2, Balance: 4.8, Concentration: 5.2, Anticipation: 5.2, Determination: 5.6, Agility: 4, Stamina: 5.2, Composure: 4.4, WorkRate: 4.4, Aggression: 4.4, ImportantMatches: 5.2, NaturalFitness: 4.8, Consistency: 5.2, Pressure: 5.2, Professionalism: 5.2,
This fixes the inability to weight hidden attributes that Genie Scout & FM26 Scoring System suffer from. I assume it calculates properly..
bf3metro said: Hi @GeorgeFloydOverdosed,
I noticed that some people have been criticizing your ratings file. I've been using your FM26 Genie Scout Ratings file since almost the beginning of FM26. Then I stopped playing around February and only came back in June. Since then, I've still been using the FM26 Genie Scout Ratings file, but I recently saw that you now recommend using the Premier League 1.0 Genie Scout Ratings file instead.
First of all, congratulations on all the work you've put into this. It's rare to see this level of dedication and effort.
I have three questions:
Are you still actively improving and refining your ratings file?
If you had to estimate it, how reliable and accurate do you think the Premier League 1.0 ratings are, as a percentage?
Finally, which file should we actually be using: the FM26 Genie Scout Ratings file or the Premier League 1.0 Genie Scout Ratings file?
That's all! I hope you'll have the time to reply. Thanks again for all your hard work, and have a great day!
I recommend the Premier League 1.0 file, while using filtering for hiddens. It is designed to 'save' you in your first season, but I think this is the right balance and does not impair min-maxxing.
It would be less accurate for FM26 most likely, but I suspect the difference would be small.
This question gets asked a lot, so I'll briefly explain why. Take long shots - it's ~3x importance in FM26 supposedly, ok that's 3x0 = 0 for me.. maybe that ends up being a 10 minimum for the attribute, which is a minor difference since most players will have ~6-8+ by default anyway. Going by HarvestGreen's data, there don't seem to be any paradigm-shifting changes, and the thing is, if you suppose balance now has say a '16' requirement, that's highly unlikely to be the case because very few players would have 16 balance in the game. If I had to make a rough guess, the differences between FM24 and FM26 might be a player going from rank 57th > 68th say. Not nothing, but nothing huge either.
The ratings file will continue to be improved, but there is no set timetable.
The main goals I would like to achieve are:
1) Re-integrate hidden stats again. Either with a program that comes along that can do so, or by essentially backporting the rankings FM26 scoring system gives and fiddling with the numbers until it largely fits the mold.
2) Create a FM26 version. Waiting until FM26 is free on Epic, or otherwise settle for working with HarvestGreen's FM26 data. It's not unfeasible, it's just a lot of effort for little reward whichever way I do it.
3) Further improve attribute value accuracy and weightings accuracy. Ideally I want to be able to say this is player is 5th, that player is 6th, etc. with high confidence.
If I had to put my impression of the accuracy of my GS rating files as a % based purely on intuition, I would put it like this:
88% - Premier League 1.0 (in FM24)
83% - Premier League 1.0 (in FM26)
82% - FM24 Pure Performance
80% - FM26 Pure Performance
75% - If you just plugged in pace/acc at 100 weight and added a few others as keithb suggests (basically every simple interpretation of HarvestGreen's data)
70% - File based on Orion's data
56% - Default Genie Scout
I can't say I even trust my own intuition on this though. I know the attribute figures are very close to accurate, translation to GS weights is another matter. Here is what the first control testing showed (number is average finishing position in Premier League):
6.5 - FM24 Pure Performance
7.333 - Premier League 1.0
13.166 - Orion's weights
So I think there's pretty clear evidence it works well, but my intuition about Premier League 1.0 being an improvement over the previous version may be wrong. I still think Premier League 1.0 is better, because it's based on solid data (my extensive attribute testing). It could just be that beyond a certain point, further precision is superfluous (i.e. if one file gives 73.54% and the other 72.88%, you're going to end up picking the same player anyway). I anticipated this and measured overlap of players. From memory the overlap was only around 50% for all three, so there is still substantial variation going on, which was actually quite surprising to me.
BrushlessPlaymaker said: Still on this topic:
Like I said, I prefer may saves with generated players rather than real ones and, since you mentioned it, I'm noticing a lot of young and potentially good player with very low jumping. Is it worth it trying to train that up? I've switched a couple of my own players with decent pace/Acc from quickness to Strength Individual training (which supposedly improves strength and jumping reach) but I'm yet to notice improvements.
I assessed alternative training focuses and found it to be a bad idea. But if a player is particularly deficient in a certain key attribute, it is probably worthwhile training them up to ~8 in that attribute.
With jumping reach, would it be worth say training up further to 12 instead of pace 16 > 18? It's hard to say. Personally I would just get the high jump reach player to begin with, because it will take time anyway, and jumping reach is particularly unforgiving if you lack it. You can get away with 14 pace; you can't get away with DCs that have 14 jump.
This set of guidelines for Italian FM volunteer researchers for FM20 is something of a goldmine when you plug it into Google Translate. Most of the FM info dished out by SI staff is pure BS, but here we have some tidbits of proper info about how it works under the hood.
There are two interesting things in it on height specifically:
1) There are conversation tables for height to jumping reach (one for GK, one for outfield).
2) Pace - 'Very short players cannot have a very high Speed ​​value (at most they will have a high acceleration value)'
This means you can use minimum height to help filter players in the starting database, as the researchers have to follow these strict guidelines in the player data they're putting in.
We can deduce that a player under 1.95m won't ever reach 20 jump even with training. 1.87m would be the minimum to eventually reach 17 jump (ideal minimum for DC). 1.82m is the bare minimum for a DC. Under 1.72m would be too deficient in jumping reach in any position, and should probably be avoided.
We are told that short players are pace-constrained to boot as well.
Now that is the starting database, but if you think about it, they've probably programmed the game to generate new players according to the same guidelines.
I wouldn't know as I haven't used those mods, because I like playing the same game mostly everyone else is playing.
I suspect that problems may arise if you make meaningful changes in isolation. For instance if you somehow reduced the importance of physicals to winning, that would probably mess up the transfers side of the game.
So if I'm using the file to figure out the "best XI" of my team for a given tactic at a given date, your earlier "FM26 GS file" would make more sense, but for long-game recruiting purposes / understanding best promising youngster from my academy, your new "Premier League 1.0" file is prbly better?
I had to go back over my posts and have a few puffs of the pipe on this one.
I misremembered the reasoning for pace/acc being low. I've been away from it all for about a month.
The goal I had started with in my 3rd iteration was this: Get the pace/acc requirement as low as possible, because getting a full team of 18 pace/acc players your first season in the Premier League isn't realistic. It turned out that ~14-15 was feasible in combination with certain levels of other attributes - it wouldn't win the Premier League, but it would guarantee avoiding relegation and even potentially achieve a continental qualification finish.
Now it's true that I also adjusted the values for training later, however pace/acc remained at 15. And yes, the idea is to train them to ~18, but this is beyond what is strictly necessary to place highly. The idea is that growing pace/acc to 18+ will ensure finishing 1st in the 2nd season onwards.
The history of this can get very confusing, but I think this timeline makes it pretty simple:
Dec 15th - First GS ratings file released.
May 7th - 'FM24 Pure Performance' is released. This is the "best XI" ratings file you have in mind. It is mostly based on HarvestGreen's data.
May 14th - LightningFlik posts the discrepancy that leads me to finding Genie Scout is bugged.
June 12th - 'Premier League 1.0' is released, which is mostly based on my own testing.
June 21st - 'Premier League 1.0' tested in comparison to 'FM24 Pure Performance', with the latter edited to avoid causing the GS bugs. Position results were 7.333 and 6.5 respectively, with Orion's weights as a control being 13.166. I had also created another version of 'Premier League 1.0' which refined it according to the essentiality of the attributes and its result was 8.333. I think I planned for it to become 'Premier League 1.1' but either the result dissuaded me or I forgot about it and took a break.
So I recommend just using the 'Premier League 1.0' file.
I agree that speed and acceleration might be underweighted in the premier league GS file. Jumping reach having high weight is an interesting take, but can you remind me why it has actually higher weight that pace/acc? I feel like nothing should be higher, according to actual testing. Not saying that it's a bad idea, but I wonder the rational behind that choice.
Also, any pointers on how (according to your file, or of general understanding of positional proficiency) we should interpret GS ratings of sub-20 position for a player as per my messages before the interruption?
The reason why pace/acc is now seemingly under-weighted is because it is expected that you will train these up using the quickness focus in training. I have based the adjustment on real training results, not just the optimal possible. They still remain the highest weighted attributes.
Jumping Reach is only weighted higher in 2 positions, DC and ST target man, but generally it is up there with pace and acc. I can't remember all the nuances around it, but here are the key ones I can recall:
1) Jumping Reach is critical on DC and has a high minimum requirement. Something like the difference between 17 and 13 is 5th > Relegated say (you can probably find these results somewhere earlier in the thread).
2) It's not that you just need one or two tall players to do the job, and that having your MC be a bit taller makes no difference. Increases even at mid levels across the team result in significantly better performance.
3) Jumping Reach is harder to train than pace/acc, at least when we assume quickness focus is being used.
I read your post on position proficiency, but don't have anything to contribute to it really. I do make sure to search for players with only green position proficiency in Genie Scout. I've dabbled with position proficiency in various ways, but I don't think to find performance level at 1 proficiency say, and I'm too hazy on it now anyway.
What I will say is that I think certain dual positions that have no CA penalty (i.e. AML/AMR) are beneficial in a practical sense.
I vaguely recall experimenting with playing optimal players out of position to try and cut the CA cost and that it failed miserably. So it does have a big impact.
As I was thinking about this, it was lurking at the back of my mind that I'm sure someone had tested proficiency and I had seen the results once. It took a while, but I think I've found what I was recalling. HarvestGreen tested it:
He also did FM26 OOP proficiency and found the win rate difference between 4 OOP proficiency and 20 OOP proficiency to be 40.9% > 46.4%. Link to thread.