The Genie Scout problem has thrown a spanner in the works at the moment that I need to work out first, but here's what I've been working on for the next version:
So I know that you can win the Premier League consistently with 1 CA players so long as they all have just about 20 pace/acc. But of course this isn't realistic to have.
On the other hand, what is realistic is that you can take a bunch of ~12-14 pace/acc players and get them to 16-18 pace/acc with ~4 years of meta training. 16-17 pace/acc for all outfield positions is the minimum you need to win the Premier league. So winning in the end, isn't really an all too difficult challenge anymore, and it gets easier if it's Championship or lower level.
Now if you take a look at starting top Premier league players in the game, you'll see quite a few with 14 pace/acc, which got me thinking. First, it means it's probably possible to have mentals/technicals be enough to make up for the low pace/acc, to the extent of 14 pace/acc. Second, that elevating mentals/technicals just enough to allow for 15 or lower pace/acc should be the name of the game now.
As it turns out, it's difficult to make even 15 pace/acc viable, but it is possible, even with a whole bunch of attributes left at '1' and others at a reasonable '13' it would win the league. But when I was initially testing it, I left the players with perfect 0 CA cost stats.
I realized I had to make it fully realistic, reducing all those 0 CA stats to ~13 as well, which reduced performance quite a lot. At the moment I'm getting results of ~7th, but the conclusion I've come to is that for the Premier League, that's good enough for a first season. The idea would be, tailor the Genie Scout ratings to get these 15 pace/acc 13 other guys for a 1st season mid-table finish (if newly promoted team) and then reach 16-18 pace/acc through training/purchases the following year for a guaranteed win.
Here's an example (DL/DR) of where I'm at so far (a team of players like this achieved 7th):
The plan is to keep reducing each attribute one by one until I whittle it down to the essentials. Then I can test some positional variety (maybe only DC/DL/DR need higher concentration) and then make some final adjustments for CA weight and pull things a little closer to HarvestGreen's findings on mentals/technicals just to be safe (some attributes I discard initially might have a small beneficial effect I didn't notice anyway). Training will also be taken into account, to the extent its still relevant. I am also testing in the Championship, as I feel it's important to be able to win that.
This will be more transparent, less arbitrary, and closely aligned to real results in a real and popular league.
LightningFlik said: 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?) Expand If I recall correctly, 18 position competency is same as 20. If it is 16, you do need to take that difference into account. Genie Scout does adjust rating using position proficiency, but it's probably somewhat inaccurate.
Your second question perplexed me for a while, but I think I get what's happening now. You are using the same start date/vanilla game data, but your manual calculation (WITHOUT using Genie Scout) does not match my Genie Scout results.
I haven't done the math on it myself, but this could indeed be a Genie Scout problem. If I clear the rating data and just make it acc 100 for ST, then I end up with a player with 16 acc equal to one with 17 acc (both are 20 position proficiency). I found 2 more examples, and it's clear that the extra footedness is making the extra contribution.
But that would only be a small part of the difference here.
It turns out Genie Scout isn't weighting all attributes equally. This has always been at the back of my mind, but I figured we've got what we've got and ended up neglecting this even as I got more seriously into it. Thankfully it turns out to only affect the personality attributes and ones with negative values.
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.
Negative attributes like injury proneness work strangely. '100' acc, '-100' injury makes 65% go to 22%, so a 65% reduction.
There is also an up to 6% discrepancy occurring. I thought maybe Genie Scout is assessing values non-linearly, but even when controlling with identical weighted attributes/proficiency/feet, I get 3 STs that are 74.00%, 73.00%, 72.00%. And it seems like all possible correlates are ruled out too, as far as I can tell, which suggests it might be just a bug. Overall it seems like it would add ~2% margin-of-error to the results, which is moderately significant.
This is all a genie scout problem, so I would think your own calculated player rankings are correct and you should use them.
tl;dr for anyone using the Genie Scout files, it turns out personality is just not working in a way I can understand in Genie Scout, and I don't have a solution right now. I'll probably take a few days to work this out.
keithb said: Stamina is definitely ahead of those. Its very important for midfield and full backs. I thought you had done some kind of values by position? Anticipation might also be ahead.
I just asked a question. You said your tests got other results? Expand No. I could win the premier league with a team of 1 CA players that have 1 stamina. And the reason stamina was left at 1 was because I found increasing it had no statistically significant effect, at least not any greater than 10 or so other attributes.
That does not mean that stamina does nothing, but it demonstrates that stamina is far from essential, even for DL/DR. HarvestGreen's data shows 6 > 18 stamina is +8.1% win rate, which is less than concentration (+8.6%), which was the last attribute I included on my toplist.
It's one thing to posit challenges based on your own intuition, nothing wrong with that. And even arrogance has preservation of dignity as is its virtue. But where is the dignity in insulting me only to piss up into your own face in front of everyone?
Kma said: @GeorgeFloydOverdosed Please do you have in-game filters for the vanilla game ? I don't use GS and other fm lab files Expand I did before. They're somewhere, but I recommend just filtering for some key attributes, say this (for adult player):
pace/acc 12 drib 8 concentration 8 (on DC/DL/DR only) work rate 6
at this point do you think this is the best one? Expand I definitely recommend the new ones over the old file. It's an improvement in spite of the flaws.
If i understand it right, with this new file the attributes of acceleration and speed are not as valued, also taking into account the issue of training and possible growth. However, I am using the fm match lab training file, will it have an impact on the development of speed and acceleration and therefore will this have an effect on this ratings?
I hope you can understand my question Expand Not that familiar with FM Match Labs, but I think they change everything up to balance things right? So it will make the ratings less accurate, but not by that much. The main things being considered are performance and CA cost, and the training bonuses/penalties I've treated as the strawberry on top.
Honestly, I'm thinking of doing another redo from scratch that will use a more grounded method that I can present more clearly and transparently.
keithb said: You dont think stamina is important for a full back? Definitely ahead of jumping, balance and strength. I would also include work rate.
Later on you go on to say you had to accept harvest green was correct about some things. Had you previously said he was wrong?
Sorry im obsessed again but your attributes for full back seem wrong. Expand Concentration 25 was the last on that list. Stamina is 17. I missed work rate, which is 31. Balance and Jumping Reach have 2 weight, stamina is 6 weight, so it's apples and oranges.
No I did not previously say HarvestGreen was wrong, I simply meant that in assessing the performance of balance & strength I trust his results because I know other attributes he gives values for are pretty precisely correct. Part of it is that his own results has changed, due to him using a new method, but mainly it is that I have taken 6 > 18 results as the basis instead of 1 > 18 results (since it is unlikely we would sign players with 1 strength, or otherwise be unable to train them to ~6).
keithb said: I will say again multiple times you have boldly declared other's work and findings to be wrong, only at a later date to retract and say you were wrong. Once is fine, maybe even twice. But you were prolific. Expand Instead of providing a single example demonstrating the above claim, you are instead trying to ask me now 'what about this one.. i-is that a time you said HarvestGreen was wrong??'
LightningFlik said: 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? Expand Yes, benefit tails off greatly or is statistically insignificant above that value.
rfsm said: Using the blended file for the GS ratings and can someone explain me why Martim is better than Baio?
Based on the physical side always assumed it would be the other way around. Expand Assuming you're assessing for DR position, then the main attributes for DR are pace, acc, drib, jump, bal, injury, str, concentration.
So at a glance they look fairly even, trading 3 points of pace/acc for ~13 points of ~30% valued other attributes. Not to mention Martim is doing it with 8% less CA cost, but this isn't part of the rating, just one of the goals.
I agree with you though that at a glance, it wasn't obvious that Martim was as good as Baio, but the math seems to check out.
Keep in mind that 'blended' devalues pace/acc a bit compared to other attributes, because it's imbued with the expectation that over ~4 years of training, pace/acc will grow but technicals/mentals will stagnate/decline. But even if you use the 'performance' file, they will probably still be near even given the math above.
I know I'm blathering too much, but I would also like to say that my old file values pace/acc more, but I simply had to accept what HarvestGreen's results are showing. Based on my 1 CA tests, I don't see evidence for balance and strength being worth ~30% weighting, but his results have always ended up being on the money so I decided to accept and include it.
Panneton0 said: I am currently trying a Youth Academy Challenge, which makes me unable to buy players, and only use my academy newgens.
With that in mind, I'm trying to put together the best way to approach the GS rating files you provided (I am on FM26 so I know they are not fully compatible, but I am wondering about the philosophy of it anyways, more than the actual scores). In this challenge, I only want to assess my own team a) probable best performers b) probable future best performers. But I don't need to find players who are "good even if low CA" because anyways I have a fixed roster and can't try to find a cheap hidden elite.
The "pure performance" file would provide me with an indication of probable best performers in my team. That part seems straightforward.
But is the "youth" file as relevant in that context? Wouldn't a player's potential, rated by the "pure performance" file, be a better prospect than a player with good "youth" rating? Do we know how GS rates "potential" of a player is computed? Is it simply considering a direct scaling of their current attributes as if linearly scaled up to their PA?
Sorry if these questions are unclear, I'm still trying to wrap my head around all this! Expand How GS computes potential is not known to me, and really I should have thought before to say bluntly that using GS 'potential' rating is probably too inaccurate with my files.
We've all used it for years, or at least I know I have, but just as with the rating values it's just not accurate enough anymore given what we know now. It would still give you a good indication of whether the player has room to grow or not, and that's a key thing, but we know that attribute distribution matters more than PA now. So I would look at current rating + PA + CA-to-PA gap + overall picture (i.e. injury 18 would rule a player out for me), and make a judgement based on that.
In your case I would still use the 'youth' file, as it optimizes for low CA (therefore can attain higher peak performance later) and takes into account the effect of training over ~4 years. If you're choosing youth to play first team games, then just switch to 'performance' for that temporarily to assess (or use 'blended' ).
If you just use the 'performance' file or similar, even if you intend to use this youth player in your first team straightaway then he will probably be subpar first team player at first (very few youth would have the pace/acc required immediately), and then a limited player later (high pace/acc, but low mentals/technicals that never grow).
tam1236 said: I have the same feeling from my long-time savescummings (actually just tests - I played whole, normal season after that). The general shape of your newgens is decided during a previous intake-day. It can bye modified by youth facilities etc, by random-generator in intake-day, but not so much (and maybe this is a mysterious hidden factor - for example you don't have facilities which you do have next year). And that's why first intake is much better - it is "half-blind-generated" in a different way maybe during generating profile when starting new game? I need two weeks to be 100% certain. Expand Yes, it's a good possibility that the game is simply giving it a running start for the first year.
Looking back at the original post of this thread, I realized this could all be a symptom perhaps of affiliate clubs. I had forgotten about this since it was over half a year ago, but I said that affiliate clubs reduced median PA for Man City by ~15-20, which is the same drop we see. And it would make sense that players from affiliates don't come through in the first year. But then again Aston Villa, Exeter and Forest Green shouldn't really suffer from this problem.
EDIT 2: It's not 'youth importance' of the club either. I've tested youth importance before and found it does nothing for PA, but I thought maybe there's a multi-year effect I somehow missed.
It's bugging me, so I've been looking into it more. This is about the median dropping in subsequent years, not game importance anymore.
Normal Man City test
2024: 144, 3.5 star, Aston Villa 134 2025: 126, 2 star, Aston Villa 144 2026: 102, 1 star, Arsenal 113, Aston Villa 133.5 2027: 94.5, 0.5 star, Arsenal 124.5, Aston Villa 124 2028: 119.5, 3 star, Arsenal 122.5, Aston Villa 121 2029: 112, 2 star, Arsenal 103, Aston Villa 126 2030: 103.5, 2 star, Arsenal 120, Aston Villa 125 2031: 120.5, 2.5 star, Arsenal 123, Aston Villa 134
Average: 115.25, Aston Villa 130.1875
Normal Aston Villa test
2024: 136.5, 5 star, Man City 132 2025: 129, 3 star, Man City 138 2026: 90, 0.5 star, Man City 121 2027: 99.5, 3 star, Man City 140 2028: 120, 3 star, Man City 120 2029: 110.5, 3 star, Man City 122 2030: 106.5, 1.5 star, Man City 123 2031: 113.5, 2.5 star, Man City 135
Average: 113.1875, Man City 128.875
Forest Green (league 2) non-player: 70, 91.5 Forest Green player: 52 (note: was 0.5 star)
Overall this is a relief, because it seems to be the case that the median of all non-player teams doesn't decline or go all over the place in subsequent years. That median variation of -/+ ~10 is normal if you're unfamiliar.
It's only the player team that is affected. And this seems to be consistent across different division levels/club quality.
It's not about existing youth players affecting things, I tested that, and the AI's median remaining unaffected which I found out later is further proof of this.
I thought maybe what's causing it is the manager itself, specifically either the reputation or the player & youngster knowledge of the human manager. But changing those attributes didn't seem to make a difference.
I think that the median being normal in the first year might be a clue as to what's causing it.
So does this mean I can set all the aforementioned attributes to 1 in this template? Or does it mean those attributes were lowered to 1 and other attributes were raised? Please let me know if there are any templates with a lower CA. Expand Yes, you'll be fine with players that have '1' in those attributes.
I recommend using my genie scout rating file, it will find you the most CA-efficient players.
keithb said: its been funny to see him say other people's work is wrong and then have to retract multiple times. Expand You keep saying this so I assume you must actually believe it to be true. I feel like I have to address it even though it's unsubstantiated, because people aren't going to be digging up my posts to check for themselves and they might just assume you're half correct.
I have never said the work of HarvestGreen or EBFM is wrong, except on some niche aspects. For Instance, I think EBFM is wrong specifically about youth facilities making a minor contribution to PA, probably because he used average instead of median as measurement. And I think he is wrong on his 'draft hypothesis' for youth recruitment (though even he wasn't 100% on this theory). But apart from these two things, I don't think I disagree about anything EBFM presented.
With HarvestGreen I think I've only ever disagreed on certain things open to interpretation. For instance I favor a different training schedule to him, but this is because he weights attributes differently in his thinking, i.e. just how negatively should 'decisions' growth be weighted.
I've pushed back against people on a few things. One was about player fitness. Another was about player personality attributes being random or not.
There are more I've forgotten, but the player personality one is one I ended up conceding on. It turned out player personality attributes are not just random, and that wasn't someone's 'work' it was just a claim made that I subsequently tested and changed my mind on. But I really can't think of anything more substantial than that I've had to retract.
White Europe said: Ok great 👍, thanks for answer, just wondered if I want to use other style than gegenpress, or just create my own tiki taka tactic is those ratings still helpful? Or they just to support meta gegenpress tactics? Expand Whatever tactic you use, the ratings will work well.
The values are necessarily quite a bit airy fairy anyway. The main thing it's doing is that it's prioritizing key attributes that are universally good such as pace, acc, dribbling, work rate, pressure, and so forth. But then when it comes to something like 'long shots' say, maybe that deserves a 1% weighting or maybe it's 11%.. but this is only going to move the overall needle a fraction of a percent anyway. But then if you add up all these little errors together, that might end up being an error of ~3%.
A simple change you could do to tailor it more to other tactics I guess is to simply reduce the pace/acc weighting by ~10-20%. I think that was the gist of the main difference I saw in HarvestGreen's findings of tactic differences.
keithb said: Again, what? Your replies are incoherent. Half a year? Obsessed? Still going on? You either waffle on with stuff that makes no sense, or in this instance throw a few words out at me.
I was merely questioning why it's taken you this long to determine several things we knew Long time ago. I have noticed multiple times you've declared other people's work and findings wrong, only to later retract and say you made a mistake. You're cosplaying being an elite tester for football manager, but you're sloppy at best.
I only came on to see if strikerless was still meta in 26. But the forum is so full of your posts I had a peek for a laugh. I see above you've mentioned certain attributes are more important for different positions. Is this true? Expand If you look at your posts page, your last post before the one in this thread was in February saying to me:
keithb said: What a load of shit😂. Do you think we're five years old?! Exposing the truths about football manager has nothing to do with your username.
Clearly you're desperate to be someone in the community, but all you're mainly doing is regurgitating other people's work. Well done. Bravo. You're a nobody. But at least you've got that username, really sticking it to SI!! Expand And then if you scroll down, there's a few more that are replies to me or about me. Two of them about being upset about my username are from October last year.
What you claim about me in relation to other people's work is simply wrong. Example at hand: Who else is attempting to update FM Genie Scout ratings values, in a way that merges HarvestGreen's findings with positional weighting of attributes?
White Europe said: Guys im new to gennie scout so i have a question: I have a guy for DM position with 70 rating in general for DM position and 69 for volante role which im using in my tactic and a guy with 68 general rating for DM position but 71 for Volante role: so which rating is more important general or role?? I hope this make sense Expand Always use the position rating, not the role rating. You can find evidence on this forum that where certain roles will say they don't need acc/pace, they still need it just as much as roles that do have them listed as requirements. Essentially, roles seem misleading and cosmetic.
I suppose it's possible that there are still variations in terms of tactical role. I.e. if you set DL to 'dribble more' maybe it benefits from better dribbling more. But I will say that in trying to adjust one of Knap's top tactics myself along these lines (to suit/fit better a certain set of attributes), I couldn't get better results, so I doubt it matters here either. HarvestGreen has found different attribute results for different tactics used, but the differences weren't that big.
I continued testing game importance at the 5 year mark and got these additional results:
England normal ('very important' ) samples:
119 120.5 113.5 106 118.5
England 'unimportant' samples:
140 143.5 140.5
2nd 'very important' samples:
105.5 111 101
At this point I stopped because there is something extra going on here that is more notable in itself. Still, I think we can pretty much wrap up the game importance thing, as ~140 PA median for Man City (or almost any club really) is about as high as it gets.
But now the matter of these highly variable medians. I have a feeling I already knew this years ago but I have forgotten about it, as I was recalling recently how I would savescum a whole year in advance to see the newgen results instead of reloading just before newgen day, in recent versions, and I couldn't remember why. Now I see it's no doubt because certain things are being set a few months earlier than intake day with the new 'Youth Preview' system.
Honestly I'm still hazy right now on it and I can't really be bothered digging deep into it, but I know this doesn't affect any data I've presented, except this multi-year attempt thing.
It's not to do with facilities, reputation, club coefficients, etc. changing - I double checked all those to make sure. What it does seem to correlate strongly with is your youth intake quality star rating. Yes, this star rating is relative to your existing squad, but it seems also nonetheless to be reflecting the differences in your median PA.
If median PA can vary for even a top club by as much as ~30% randomly, then does knowing the newgen factors even matter at this point? In a way this is a point even I don't want to admit, but it's one reason FM was losing its lustre for me even before FM26, they ruined the fun of newgens. On the other hand, the factors still work, and the year-to-year randomness doesn't mean inherent randomness - it's a dampener for sure, but still means you have a reasonably predictable and consistent system. To illustrate the difference:
Before:
Man City - 140, 143, 135, 146, 148 Arsenal - 136, 129, 138, 139, 139
Now (nation-based year-to-year randomness):
Man City - 140, 110, 115, 130, 145 Arsenal - 136, 107, 110, 119, 142
Inherent randomness (~30%):
Man City - 140, 110, 115, 130, 160 Arsenal - 158, 181, 132, 105, 116
That's if nation-based year-to-year randomness is what is actually going on here. As I said, I'm quite hazy on it right now. I'll probably try and get to the bottom of it sometime down the track. I know this all seems to put things in the wrong direction of progress, and people generally don't like that, but better to acknowledge the setbacks/problems than keep them on the down low and end up fooling myself/others about it.
keithb said: What are you on about?? You seem to be discovering a lot of things years after most people knew that already. Whats next? the sun is hot? Expand Oh I didn't realize you're the guy who is obsessed with me. It's been half a year and you're still going on like this. I wanted to communicate to others the points I made anyway.
BaZuKa said: Ok this new file is Giga Broken Season 3 almost won everything with a low-tier club from Portugal Expand The first line sunk me for a moment there. It's good to hear it's working well for you, I don't actually know how well in reality these files are going to go.
I keep forgetting to mention things. I used HarvestGreen's 6 > 18 attribute data mainly this time, as using 1-20 overvalues things like pressure and work rate. I figure in cases where pressure or work rate is very low, you can either just filter out those players or train/tutor them up a bit if you do buy them. And getting '20' is less likely as well as more CA-inefficient, so that's another reason I favor the 6>18 measure.
keithb said: Haaland is clear in FM 24. Surely it hasn't taken this much research to determine that? Mbappe is the best lw. Glad we've been finally able to clear up that FM genie scout default ratings are way off. Expand
Just piggybacking off this to say that I haven't used the player names/clubs to calibrate my values. There's a good reason for this, and that is that in-game player ratings don't correspond exactly to actual performance.
You can see evidence of this in this OmegaLuke video, where technicals give substantially higher player ratings than physicals, even though physicals actually won all the games. Orion's coefficients, which use in-game player ratings to deduce the best attributes, I've found are outdone by HarvestGreen's data which assesses according to goals scored or games won.
So this is why I don't try and align the values to fit player ratings or the best players, as that would achieve the opposite of what I want to achieve, which is show players who punch above their weight in a way the in-game AI doesn't recognize.
But I think there is some value in comparing the results afterwards, just to make sure one isn't completely off-track. Regardless of rating, we know Haaland gets goals, so if he's not up there then there's something amiss.
Additionally I think it's notable how the starting data closely aligns with physicals matter and technicals don't. If you plug in HarvestGreen's data, it just so happens that the top players start in the top clubs in the game. That sounds as straightforward 2+2=4, except realize that this means that SI knows exactly how the attributes are skewed and disingenuous. If they believed what they tell you about how the game works, then we should see these initial players at top clubs failing to perform. In fact, I think perhaps this in-game rating bias towards technicals is to try and stop AI managers simply buying up completely lopsided physical beasts as the game goes on.
I can think of a reasonable counter-argument or two to the above, but there is also some circumstantial evidence as I see it. If you use Genie Scout default ratings, 7 out of 9 top players (one of each position) are white. If you use HarvestGreen's data it changes to 7 out of 9 being black. Either SI still has a racism problem where black people are portrayed as mentally/technically poor physical beasts, or this is intentional. Or both.
LightningFlik said: 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.
(No rush, whenever you have time). Expand OpticFawn said: Yes please, I want to use on personal tool pls Expand Here are the values for FM24 blended.
I forgot to say that 'sweeper' position is intended to be my attempt at a tutor rating, and for Target Striker I've simply upped jumping reach to 100, as high pace/acc ST is simply better than a slow target man and I'm not sure if heading or whatnot affects the target man's performance (jumping reach certainly matters).
I haven't done FM26 versions yet and it'll be a while before I do, I recommend just using the FM26 ratings file I did before and replacing the GK ratings with the new values in FM24 Blended. You could use these new files in FM26, but it would be a bit off (still much better than FM Genie Scout default).
I'm increasingly unsatisfied with the result as I've ran into further issues, but I've decided to just put these out there otherwise it'll never get done and it does seem to work better despite the issues.
I've had a dilemma about merging in some of my own findings. One example is that in my 1 CA player testing I found that concentration matters on CB/DL/DR, not other positions. Another is that extra finishing on CBs made no difference to goals scored, which means it's low weight on CBs for a reason. These findings are a part of my previous ratings file, so I wanted to include them again here. But I also figured that my data isn't as strongly evidenced as HarvestGreen's, and some of my conclusions could just simply be wrong. On the other hand, there were a number of different examples which can't all be wrong and it makes sense that different positions benefit from certain attributes to different degrees, so I don't think it's best just to use HarvestGreen's data flatly amongst all players. In the end I decided to just apply a simple -/+25% to a few attributes that I was quite sure had pronounced positional differences. I.e. dribbling 32 > 40, vision 5 > 7. So nothing too drastic, but this leaves me feeling like it is too arbitrary a change and yet it also doesn't go far enough probably. The only reassurance I have is that the actual results seem to be better, which you can see a sample of below.
Another issue I've had is with the new GK data. Initially I just plugged in HarvestGreen's new figures. But looking at the results, I get the impression something isn't quite right with the new data. And if you look at HarvestGreen's old GK data, which only had conceded goals as a measure but was also measuring 10>20 instead of 1>20, you can see there are contradictions. From my own 1 CA player tests, I found that acc & pace seemed best value around ~5-8 rather than 20. And if you look at actual players in the database, a bunch of the best have acc of just 8-10. So I think this is mainly a case of missing the data that shows where ~6-14 is sufficient, which we know some attributes such as work rate are like. So I made some appropriate adjustments. Definitely an improvement on my old ratings file here.
Here are some example results. Two important differences that stand out to me are that the GKs are significantly more correct in the new version (Pickford shouldn't be behind Forster!) and there are no longer 2 japanese players from Celtic in the top 10 ST list (they do have high acc/pace, but this seems an appropriate change). And clearly with Kane as no. 1 ST, FM Genie Scout default ratings are way off.
New (blended):
Pickford 69.53% Pope 67.67% Ramsdale 67.43% Steele 65.41%
Old:
Pope 69.74% Forster 65.89% Pickford 65.13% Ramsdale 65.00%
HarvestGreen (new GK data):
Pope 75.44% Pickford 73.89% Ramsdale 72.98% Butland 72.85%
jimmysthebestcop said: Extremely interesting. I havent even bought or played the demo of Fm26 I am staying away from it as it is a flaming dumpster of poop and I cant reward SI with my money. Maybe I will jump back in fm27. Is it different in fm26? No idea.
I would be interested in your results over a 10-20 year span. At least in Fm24 and all previous versions if Game Importance wasnt set to very high in a decent Youth nation you couldnt produce NewGens even if your were #1 club in the world.
Singapore could never have good players because there youth rating is bad. Czech would be a good test as their youth rating is 90-100+ while game importance is not set to very important.
I dont think this is an issue for most players since most people play in a big nation, build a nation saves arent that popular in the community. So most people wont notice it.
I am honestly probably not explaining myself well. I just know in fm24 and prior when doing build a nation saves even if you changed youth rating to 100 the nation couldnt ever produce wonder kids even when getting league to top 5 and all the clubs in the top 50 if Game Importance wasnt set to very important.
I dont know if there is some kind of time factor and that is why you need to sim 10-20 seasons or what tbh. Expand I should have mentioned that I used FM24. I guess some will say then what's the point of doing an updated test if you're not going to even use the most recent edition, but for me FM26 just has to be foregone.
I actually completely forgot to do one thing I wanted to do this time, which was to test what it looks like 5-10 years into the future.
I wanted to do USA initially, to address your claim about the MLS exactly, but found newgen intake works a bit differently and I couldn't be bothered trying to work it out. So my plan after that was to choose a nation with low game importance but high youth rating, but there aren't really any that are loadable - and singapore was one of the best options that was left. But this is also why I did Man City, because I had in mind (and generally what people play with) is clubs with top facilities in top nations, and England is no.1 for that.
As I was writing this I was doing my testing, and I was about to say that I've been wrong about game importance all these years, as 5 years in, the Man City (unimportant England) median had slipped to 117.5 PA (155 peak) which is outside of the range one would expect for the 131.166 PA averaged median I had got for the initial year. But then I did the control test (normal England), and that was 122 PA median, and 115.5 PA on the second sample.
This is bringing back what I experienced regarding game importance back when I initially tested it in FM19. No one has so far pointed out that I said game importance has no effect, yet later claim it has a minor effect, but I want to clarify this. As you can see, game importance has a minor effect that is only sometimes apparent. When I started testing it, I used average PA as a measure and so it was invisible to me and I thought it had no effect. Once I cottoned on to median PA being more accurate, I observed that questionable minor effect. In multi-year tests I thought I had borne out its effect, only to be confounded again in the same way this time. So I had to conclude whether a handful of PA difference, some of the time, constituted something real or an illusion. I concluded it was illusory, although the bunching up around the median effect is obviously real. It's a bit difficult to explain this fully if you haven't tried this kind of testing yourself, but basically it's very easy to be hoodwinked by limited samples and the inherent randomness. If I test a club and get 139, 134, 140, 137 then it's very tempting to say this club is a ~137.5 PA club. But then I would get 153 and 145 if I took 2 more samples. So this is why when I saw game importance raising PA by at most ~10% and inconsistently, I put it down to randomness.
I will continue this testing a bit, to clarify results at the 5 year and 10 year mark, as it does require at least 3 (preferably 5) samples to give any reliable indication.
So my take on the valuation of free CA is roughly this:
We know it has always has some value, because high CA-PA gap = faster training = higher pace/acc faster.
Therefore a zero CA cost attribute such as 'pressure' that has a significant effect on win rate, is even more valuable than an equivalent attribute that costs CA, say dribbling. But by how much exactly? 2 points of pace and acceleration each perhaps? I prefer to go on the safe side and guesstimate 1-2 points of just pace say (I think about what gets 'sacrificed' to stay within PA limit at the end of ~4 years of training).
But then you have positions such as DM where you can easily max out pace/acc to 20 without hitting the PA cap, at least if you're in a good division (which I think most players are in, or plan to end up in).
So I figure the bonus for DM should be reduced, and conversely the bonus for a CA-tight position such as AML should be increased.
But this leads to the peculiar consequence that positions that benefit most from high pace/acc, such as AML, value them least. Not to mention the fact that lowering the immediate gains (from high pace/acc) for expected future gains that may never even eventuate. And then there are also inherent variables whose expected ranges exceed the capacity of this predictive method. For instance, even assuming everyone uses meta training, one player might go from 15>17 pace, another will go from 14>20.
So in the end what I decided is, I will have a set of values for youth/optimization, a set of values for age26+/team selection/pure performance (which will be very closely aligned with HarvestGreen's findings), and then a blend of the two - which will be the file I recommend to use as switching between files is tedious. Is a 50/50 blend the best? Probably not, but it's the best I can do so far. I have checked in genie scout the actual results of these new values, and it seems to be working as it should - the best players are those from Man City, Barcelona, Real Madrid, etc. as you'd expect. I've looked for outliers that have changed positions the most, and overall I'd say the margin of error could be something like -/+3% genie scout rating, which I think is satisfactory.
Here's an example to give you an idea about things:
Default Genie Scout - Kane 91.47%, Haaland 90.95%, Mbappe 88.30% Orion's Coefficients (not my file) - Haaland 88.01%, Mbappe 86.31%, Kane 81.63% My existing file - Mbappe 77.59%, Haaland 77.25%, Kane 67.35% New file (youth/optimization) - Mbappe 79.73%, Haaland 75.69%, Kane 65.64% New file (age26+/pure performance) - Haaland 95.34%, Mbappe 93.59%, Kane 82.21% New file (blended) - Mbappe 85.33%, Haaland 83.74%, Kane 72.46%
Ignore the numbers themselves, it's about how relative they are to each other
suwit13 said: Hello, I've been using your player rating coeff on Genie Scout and it's been working very well.
Recently, harvestgreen22 released a new test for GK attributes. Can you convert the data to your player rating coeff? @GeorgeFloydOverdosed Expand I've actually been working on this recently
Initially I was just going to redo the GK, but then I realized I want to take a new approach to the whole thing.
I've ended up not entirely satisfied with the new approach, basically I think it's definitely better but even more arbitrary in a way. It's mostly to do with balancing CA weighting vs. raw performance, and the problematic question is if an attribute gives a slight boost to win rate but on the other hand also only has very low CA weight (or zero CA weight, like personality attributes), do you give that some kind of bonus and if so to what degree? Basically, how much is free CA even worth exactly. So I've come up with a take on this, that also integrates training results with good accuracy (i.e. if meta training is expected to lower passing by 1 over 4 years, I adjust the passing value slightly accordingly - this is more pronounced for pace/acc which will increase by ~4 in youth over 4 years!).
It's largely done now, just have to finish off some positions I don't use like WBL/MC/AMC and also create a FM26 adjusted version before I post it. Will be soon.
jimmysthebestcop said: I wish I was mistaken all of the Build a Nation players know about Game Importance superseding NewGen settings. Go watch a ton of of SecondYellowCard Build a Nation setup videos and read his discords where his followers test all of the NewGen stuff for the lowest nations.
I can tell you my last 3 build a nation saves were hungary before I knew about game importance, Andorra and Faroe Islands. Hungary even after 60 years where I gave every club billions and billions to max out their facilities could never produce a NewGen that was as good as IRL national club member. The way you give clubs billions is by buying their horrendous youth players for 10 million each as the board never turns down 10 million offers. So you are buying 30+ players per club eventually. Giving clubs 300 million each season. Expand Coincidentally it was a recent video by SecondYellowCard where he tested the DoF role that got me thinking that youth recruitment might work similarly. Turns out it likely doesn't, but that's what led me to investigate the 'place of birth' thing.
I haven't been able to find this discord you mention yet, but I've done some testing of game importance anyway, not only to draw a conclusion here but because it's good to retest things once every few years to see if the mechanic has changed and also I'd like further clarity/precision on the distribution effect I mentioned. I retested recently the hidden nation factor thing I assert is the case, and I can confirm it still exists.
England very important (default) Man City average (3 samples):
median 142.166 average 135.375 range 68.333-172.333
england-wide stats:
range 30-179 10th best player = 157.333 PA
England unimportant Man City average (3 samples):
median 131.666 average 132.687 range 78.666-162
england-wide stats:
range 34-170.33 10th best player = 153.666 PA
Conclusions
Game importance has no strong effect on PA. Previously I had said it bunches up the PA around the median and ends up affecting the PA to the tune of ~10%, and this is what we see. But as I tried to communicate in qualifying that 'it's not directly comparable to the other PA factors so it's hard to pin down a precise figure', it's not a simple flat 10% - in Singapore's case, the effect was 0% or even negative. In England's case, it was an 8% difference, for both Man City and (to a similar degree) the nation as a whole.
The main takeaway from game importance as I see it is that because it bunches up PA around the median, it does actually dampen your wonderkid chances just enough to make one not want to dismiss it entirely, but it's not that significant or consistently applicable as other factors such as junior coaching are. 160 PA instead of 170 PA, in some cases.
I haven't done a deep analysis of it, but just my general impression is that its not that Game Importance is shifting up or down the general quality of the newgens either, and this is somewhat indicated by the low difference in the average as well (2% difference for the Man City samples). I think it's simply bunching up the distribution closer to the median.
I hope this also serves as an illustration of why I use the median instead of the average. The median is pretty stable and therefore predictable unlike the average. While the median can be predicted with strong likelihood of being within a range of -/+ 5 PA or thereabouts with enough samples, the average will end up with an uncertainty range of ~10 PA or more even if you collect many samples.
I included height in the singapore samples as I thought it was interesting to observe; I wondered if its distribution has some correlation, or something in common, with the PA distribution.
Here's my rough mockup of what the 'place of birth' selection actually looks like.
This whole thing appears to be purely cosmetic, so don't try and work out how to get better newgens with it. There is no actual competing for newgens, or even newgen generation in these cities, going on.
There are three clubs: Ljubljana, Kamnik, and Maribor.
The rectangle is area the club can draw newgens from.
It is not that newgens are generated at each city and a few of them get picked up, it is that the club generates precisely 16 newgens. There 'place of birth' is probabilistic based on distance from club and 'inhabitants range' of the city. The circles are just to convey the idea of decreasing probability the further you go out from the club location.
Notice how Ljubljana and Maribor each draw ~4 players from their own city, but Kamnik only draws 2 and takes many more from nearby the nearby capital of Ljubljana. This is because Kamnik has low inhabitants range, and Ljubljana is high probability because of proximity + high inhabitants.
I do not think city 'attraction' affects it, I have tested it but the results are not 100% clear but clear enough to rule it out I think. From memory, 'inhabitants range' is also relative to 'inhabitants range' of other cities. That is to say, if Ljubjlana only had 10,000 people, it would still be top dog if all other cities are 1000< pop, but also it would be less commonly the place of birth than if it had 20mil people and others 1000< pop. There are some further nuances that reveal themselves when you try to break it with extremes like this, but since the mechanic is cosmetic, I won't go further into that.
Lastly there is the matter of exceptions. As you can see, sometimes there can be instances outside the rectangle. I don't know what exactly is going on here, I suspect it has something to do with youth recruitment perhaps. Maybe it's even just a randomness factor they've put in to try and better reflect reality.
So I know that you can win the Premier League consistently with 1 CA players so long as they all have just about 20 pace/acc. But of course this isn't realistic to have.
On the other hand, what is realistic is that you can take a bunch of ~12-14 pace/acc players and get them to 16-18 pace/acc with ~4 years of meta training. 16-17 pace/acc for all outfield positions is the minimum you need to win the Premier league. So winning in the end, isn't really an all too difficult challenge anymore, and it gets easier if it's Championship or lower level.
Now if you take a look at starting top Premier league players in the game, you'll see quite a few with 14 pace/acc, which got me thinking. First, it means it's probably possible to have mentals/technicals be enough to make up for the low pace/acc, to the extent of 14 pace/acc. Second, that elevating mentals/technicals just enough to allow for 15 or lower pace/acc should be the name of the game now.
As it turns out, it's difficult to make even 15 pace/acc viable, but it is possible, even with a whole bunch of attributes left at '1' and others at a reasonable '13' it would win the league. But when I was initially testing it, I left the players with perfect 0 CA cost stats.
I realized I had to make it fully realistic, reducing all those 0 CA stats to ~13 as well, which reduced performance quite a lot. At the moment I'm getting results of ~7th, but the conclusion I've come to is that for the Premier League, that's good enough for a first season. The idea would be, tailor the Genie Scout ratings to get these 15 pace/acc 13 other guys for a 1st season mid-table finish (if newly promoted team) and then reach 16-18 pace/acc through training/purchases the following year for a guaranteed win.
Here's an example (DL/DR) of where I'm at so far (a team of players like this achieved 7th):
The plan is to keep reducing each attribute one by one until I whittle it down to the essentials. Then I can test some positional variety (maybe only DC/DL/DR need higher concentration) and then make some final adjustments for CA weight and pull things a little closer to HarvestGreen's findings on mentals/technicals just to be safe (some attributes I discard initially might have a small beneficial effect I didn't notice anyway). Training will also be taken into account, to the extent its still relevant. I am also testing in the Championship, as I feel it's important to be able to win that.
This will be more transparent, less arbitrary, and closely aligned to real results in a real and popular league.
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?)
If I recall correctly, 18 position competency is same as 20. If it is 16, you do need to take that difference into account. Genie Scout does adjust rating using position proficiency, but it's probably somewhat inaccurate.
Your second question perplexed me for a while, but I think I get what's happening now. You are using the same start date/vanilla game data, but your manual calculation (WITHOUT using Genie Scout) does not match my Genie Scout results.
I haven't done the math on it myself, but this could indeed be a Genie Scout problem. If I clear the rating data and just make it acc 100 for ST, then I end up with a player with 16 acc equal to one with 17 acc (both are 20 position proficiency). I found 2 more examples, and it's clear that the extra footedness is making the extra contribution.
But that would only be a small part of the difference here.
It turns out Genie Scout isn't weighting all attributes equally. This has always been at the back of my mind, but I figured we've got what we've got and ended up neglecting this even as I got more seriously into it. Thankfully it turns out to only affect the personality attributes and ones with negative values.
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.
Negative attributes like injury proneness work strangely. '100' acc, '-100' injury makes 65% go to 22%, so a 65% reduction.
There is also an up to 6% discrepancy occurring. I thought maybe Genie Scout is assessing values non-linearly, but even when controlling with identical weighted attributes/proficiency/feet, I get 3 STs that are 74.00%, 73.00%, 72.00%. And it seems like all possible correlates are ruled out too, as far as I can tell, which suggests it might be just a bug. Overall it seems like it would add ~2% margin-of-error to the results, which is moderately significant.
This is all a genie scout problem, so I would think your own calculated player rankings are correct and you should use them.
tl;dr for anyone using the Genie Scout files, it turns out personality is just not working in a way I can understand in Genie Scout, and I don't have a solution right now. I'll probably take a few days to work this out.
Yes
keithb said: Stamina is definitely ahead of those. Its very important for midfield and full backs. I thought you had done some kind of values by position? Anticipation might also be ahead.
I just asked a question. You said your tests got other results?
No. I could win the premier league with a team of 1 CA players that have 1 stamina. And the reason stamina was left at 1 was because I found increasing it had no statistically significant effect, at least not any greater than 10 or so other attributes.
That does not mean that stamina does nothing, but it demonstrates that stamina is far from essential, even for DL/DR. HarvestGreen's data shows 6 > 18 stamina is +8.1% win rate, which is less than concentration (+8.6%), which was the last attribute I included on my toplist.
It's one thing to posit challenges based on your own intuition, nothing wrong with that. And even arrogance has preservation of dignity as is its virtue. But where is the dignity in insulting me only to piss up into your own face in front of everyone?
Kma said: @GeorgeFloydOverdosed Please do you have in-game filters for the vanilla game ? I don't use GS and other fm lab files
I did before. They're somewhere, but I recommend just filtering for some key attributes, say this (for adult player):
pace/acc 12
drib 8
concentration 8 (on DC/DL/DR only)
work rate 6
That would filter out most of the complete duds.
at this point do you think this is the best one?
I definitely recommend the new ones over the old file. It's an improvement in spite of the flaws.
If i understand it right, with this new file the attributes of acceleration and speed are not as valued, also taking into account the issue of training and possible growth. However, I am using the fm match lab training file, will it have an impact on the development of speed and acceleration and therefore will this have an effect on this ratings?
I hope you can understand my question
Not that familiar with FM Match Labs, but I think they change everything up to balance things right? So it will make the ratings less accurate, but not by that much. The main things being considered are performance and CA cost, and the training bonuses/penalties I've treated as the strawberry on top.
Honestly, I'm thinking of doing another redo from scratch that will use a more grounded method that I can present more clearly and transparently.
keithb said: You dont think stamina is important for a full back? Definitely ahead of jumping, balance and strength. I would also include work rate.
Later on you go on to say you had to accept harvest green was correct about some things. Had you previously said he was wrong?
Sorry im obsessed again but your attributes for full back seem wrong.
Concentration 25 was the last on that list. Stamina is 17. I missed work rate, which is 31. Balance and Jumping Reach have 2 weight, stamina is 6 weight, so it's apples and oranges.
No I did not previously say HarvestGreen was wrong, I simply meant that in assessing the performance of balance & strength I trust his results because I know other attributes he gives values for are pretty precisely correct. Part of it is that his own results has changed, due to him using a new method, but mainly it is that I have taken 6 > 18 results as the basis instead of 1 > 18 results (since it is unlikely we would sign players with 1 strength, or otherwise be unable to train them to ~6).
keithb said: I will say again multiple times you have boldly declared other's work and findings to be wrong, only at a later date to retract and say you were wrong. Once is fine, maybe even twice. But you were prolific.
Instead of providing a single example demonstrating the above claim, you are instead trying to ask me now 'what about this one.. i-is that a time you said HarvestGreen was wrong??'
LightningFlik said: 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?
Yes, benefit tails off greatly or is statistically insignificant above that value.
Based on the physical side always assumed it would be the other way around.
Assuming you're assessing for DR position, then the main attributes for DR are pace, acc, drib, jump, bal, injury, str, concentration.
Baio vs Martim:
16 acc > 14
15 pace > 14
11 drib < 14
10 jump < 12
13 bal < 15
11 inj < 7
11 str | 11
12 con < 13
So at a glance they look fairly even, trading 3 points of pace/acc for ~13 points of ~30% valued other attributes. Not to mention Martim is doing it with 8% less CA cost, but this isn't part of the rating, just one of the goals.
I agree with you though that at a glance, it wasn't obvious that Martim was as good as Baio, but the math seems to check out.
Keep in mind that 'blended' devalues pace/acc a bit compared to other attributes, because it's imbued with the expectation that over ~4 years of training, pace/acc will grow but technicals/mentals will stagnate/decline. But even if you use the 'performance' file, they will probably still be near even given the math above.
I know I'm blathering too much, but I would also like to say that my old file values pace/acc more, but I simply had to accept what HarvestGreen's results are showing. Based on my 1 CA tests, I don't see evidence for balance and strength being worth ~30% weighting, but his results have always ended up being on the money so I decided to accept and include it.
Panneton0 said: I am currently trying a Youth Academy Challenge, which makes me unable to buy players, and only use my academy newgens.
With that in mind, I'm trying to put together the best way to approach the GS rating files you provided (I am on FM26 so I know they are not fully compatible, but I am wondering about the philosophy of it anyways, more than the actual scores). In this challenge, I only want to assess my own team a) probable best performers b) probable future best performers. But I don't need to find players who are "good even if low CA" because anyways I have a fixed roster and can't try to find a cheap hidden elite.
The "pure performance" file would provide me with an indication of probable best performers in my team. That part seems straightforward.
But is the "youth" file as relevant in that context? Wouldn't a player's potential, rated by the "pure performance" file, be a better prospect than a player with good "youth" rating? Do we know how GS rates "potential" of a player is computed? Is it simply considering a direct scaling of their current attributes as if linearly scaled up to their PA?
Sorry if these questions are unclear, I'm still trying to wrap my head around all this!
How GS computes potential is not known to me, and really I should have thought before to say bluntly that using GS 'potential' rating is probably too inaccurate with my files.
We've all used it for years, or at least I know I have, but just as with the rating values it's just not accurate enough anymore given what we know now. It would still give you a good indication of whether the player has room to grow or not, and that's a key thing, but we know that attribute distribution matters more than PA now. So I would look at current rating + PA + CA-to-PA gap + overall picture (i.e. injury 18 would rule a player out for me), and make a judgement based on that.
In your case I would still use the 'youth' file, as it optimizes for low CA (therefore can attain higher peak performance later) and takes into account the effect of training over ~4 years. If you're choosing youth to play first team games, then just switch to 'performance' for that temporarily to assess (or use 'blended' ).
If you just use the 'performance' file or similar, even if you intend to use this youth player in your first team straightaway then he will probably be subpar first team player at first (very few youth would have the pace/acc required immediately), and then a limited player later (high pace/acc, but low mentals/technicals that never grow).
And that's why first intake is much better - it is "half-blind-generated" in a different way maybe during generating profile when starting new game? I need two weeks to be 100% certain.
Yes, it's a good possibility that the game is simply giving it a running start for the first year.
Looking back at the original post of this thread, I realized this could all be a symptom perhaps of affiliate clubs. I had forgotten about this since it was over half a year ago, but I said that affiliate clubs reduced median PA for Man City by ~15-20, which is the same drop we see. And it would make sense that players from affiliates don't come through in the first year. But then again Aston Villa, Exeter and Forest Green shouldn't really suffer from this problem.
EDIT: It's not the affiliates
Man City affiliates removed:
2024 - 142.5
2025 - 119.5
2026 - 123
2027 - 101
2028 - 113
2029 - 107.5
2030 - 114.5
Average: 117.28
EDIT 2: It's not 'youth importance' of the club either. I've tested youth importance before and found it does nothing for PA, but I thought maybe there's a multi-year effect I somehow missed.
Man City affiliates removed 20 youth importance:
2024 - 141.5
2025 - 126
2026 - 121.5
2027 - 107.5
Normal Man City test
2024: 144, 3.5 star, Aston Villa 134
2025: 126, 2 star, Aston Villa 144
2026: 102, 1 star, Arsenal 113, Aston Villa 133.5
2027: 94.5, 0.5 star, Arsenal 124.5, Aston Villa 124
2028: 119.5, 3 star, Arsenal 122.5, Aston Villa 121
2029: 112, 2 star, Arsenal 103, Aston Villa 126
2030: 103.5, 2 star, Arsenal 120, Aston Villa 125
2031: 120.5, 2.5 star, Arsenal 123, Aston Villa 134
Average: 115.25, Aston Villa 130.1875
Normal Aston Villa test
2024: 136.5, 5 star, Man City 132
2025: 129, 3 star, Man City 138
2026: 90, 0.5 star, Man City 121
2027: 99.5, 3 star, Man City 140
2028: 120, 3 star, Man City 120
2029: 110.5, 3 star, Man City 122
2030: 106.5, 1.5 star, Man City 123
2031: 113.5, 2.5 star, Man City 135
Average: 113.1875, Man City 128.875
Forest Green (league 2) non-player: 70, 91.5
Forest Green player: 52 (note: was 0.5 star)
Exeter (league 2) non-player: 107, 108
Exeter player: 80, 79
Overall this is a relief, because it seems to be the case that the median of all non-player teams doesn't decline or go all over the place in subsequent years. That median variation of -/+ ~10 is normal if you're unfamiliar.
It's only the player team that is affected. And this seems to be consistent across different division levels/club quality.
It's not about existing youth players affecting things, I tested that, and the AI's median remaining unaffected which I found out later is further proof of this.
I thought maybe what's causing it is the manager itself, specifically either the reputation or the player & youngster knowledge of the human manager. But changing those attributes didn't seem to make a difference.
I think that the median being normal in the first year might be a clue as to what's causing it.
(templates:115 CA average)
So does this mean I can set all the aforementioned attributes to 1 in this template?
Or does it mean those attributes were lowered to 1 and other attributes were raised?
Please let me know if there are any templates with a lower CA.
Yes, you'll be fine with players that have '1' in those attributes.
I recommend using my genie scout rating file, it will find you the most CA-efficient players.
You keep saying this so I assume you must actually believe it to be true. I feel like I have to address it even though it's unsubstantiated, because people aren't going to be digging up my posts to check for themselves and they might just assume you're half correct.
I have never said the work of HarvestGreen or EBFM is wrong, except on some niche aspects. For Instance, I think EBFM is wrong specifically about youth facilities making a minor contribution to PA, probably because he used average instead of median as measurement. And I think he is wrong on his 'draft hypothesis' for youth recruitment (though even he wasn't 100% on this theory). But apart from these two things, I don't think I disagree about anything EBFM presented.
With HarvestGreen I think I've only ever disagreed on certain things open to interpretation. For instance I favor a different training schedule to him, but this is because he weights attributes differently in his thinking, i.e. just how negatively should 'decisions' growth be weighted.
I've pushed back against people on a few things. One was about player fitness. Another was about player personality attributes being random or not.
There are more I've forgotten, but the player personality one is one I ended up conceding on. It turned out player personality attributes are not just random, and that wasn't someone's 'work' it was just a claim made that I subsequently tested and changed my mind on. But I really can't think of anything more substantial than that I've had to retract.
White Europe said: Ok great 👍, thanks for answer, just wondered if I want to use other style than gegenpress, or just create my own tiki taka tactic is those ratings still helpful? Or they just to support meta gegenpress tactics?
Whatever tactic you use, the ratings will work well.
The values are necessarily quite a bit airy fairy anyway. The main thing it's doing is that it's prioritizing key attributes that are universally good such as pace, acc, dribbling, work rate, pressure, and so forth. But then when it comes to something like 'long shots' say, maybe that deserves a 1% weighting or maybe it's 11%.. but this is only going to move the overall needle a fraction of a percent anyway. But then if you add up all these little errors together, that might end up being an error of ~3%.
A simple change you could do to tailor it more to other tactics I guess is to simply reduce the pace/acc weighting by ~10-20%. I think that was the gist of the main difference I saw in HarvestGreen's findings of tactic differences.
I was merely questioning why it's taken you this long to determine several things we knew Long time ago. I have noticed multiple times you've declared other people's work and findings wrong, only to later retract and say you made a mistake. You're cosplaying being an elite tester for football manager, but you're sloppy at best.
I only came on to see if strikerless was still meta in 26. But the forum is so full of your posts I had a peek for a laugh. I see above you've mentioned certain attributes are more important for different positions. Is this true?
If you look at your posts page, your last post before the one in this thread was in February saying to me:
keithb said: What a load of shit😂. Do you think we're five years old?! Exposing the truths about football manager has nothing to do with your username.
Clearly you're desperate to be someone in the community, but all you're mainly doing is regurgitating other people's work. Well done. Bravo. You're a nobody. But at least you've got that username, really sticking it to SI!!
And then if you scroll down, there's a few more that are replies to me or about me. Two of them about being upset about my username are from October last year.
What you claim about me in relation to other people's work is simply wrong. Example at hand: Who else is attempting to update FM Genie Scout ratings values, in a way that merges HarvestGreen's findings with positional weighting of attributes?
White Europe said: Guys im new to gennie scout so i have a question:
I have a guy for DM position with 70 rating in general for DM position and 69 for volante role which im using in my tactic and a guy with 68 general rating for DM position but 71 for Volante role: so which rating is more important general or role?? I hope this make sense
Always use the position rating, not the role rating. You can find evidence on this forum that where certain roles will say they don't need acc/pace, they still need it just as much as roles that do have them listed as requirements. Essentially, roles seem misleading and cosmetic.
I suppose it's possible that there are still variations in terms of tactical role. I.e. if you set DL to 'dribble more' maybe it benefits from better dribbling more. But I will say that in trying to adjust one of Knap's top tactics myself along these lines (to suit/fit better a certain set of attributes), I couldn't get better results, so I doubt it matters here either. HarvestGreen has found different attribute results for different tactics used, but the differences weren't that big.
England normal ('very important' ) samples:
119
120.5
113.5
106
118.5
England 'unimportant' samples:
140
143.5
140.5
2nd 'very important' samples:
105.5
111
101
At this point I stopped because there is something extra going on here that is more notable in itself. Still, I think we can pretty much wrap up the game importance thing, as ~140 PA median for Man City (or almost any club really) is about as high as it gets.
But now the matter of these highly variable medians. I have a feeling I already knew this years ago but I have forgotten about it, as I was recalling recently how I would savescum a whole year in advance to see the newgen results instead of reloading just before newgen day, in recent versions, and I couldn't remember why. Now I see it's no doubt because certain things are being set a few months earlier than intake day with the new 'Youth Preview' system.
Honestly I'm still hazy right now on it and I can't really be bothered digging deep into it, but I know this doesn't affect any data I've presented, except this multi-year attempt thing.
It's not to do with facilities, reputation, club coefficients, etc. changing - I double checked all those to make sure. What it does seem to correlate strongly with is your youth intake quality star rating. Yes, this star rating is relative to your existing squad, but it seems also nonetheless to be reflecting the differences in your median PA.
If median PA can vary for even a top club by as much as ~30% randomly, then does knowing the newgen factors even matter at this point? In a way this is a point even I don't want to admit, but it's one reason FM was losing its lustre for me even before FM26, they ruined the fun of newgens. On the other hand, the factors still work, and the year-to-year randomness doesn't mean inherent randomness - it's a dampener for sure, but still means you have a reasonably predictable and consistent system. To illustrate the difference:
Before:
Man City - 140, 143, 135, 146, 148
Arsenal - 136, 129, 138, 139, 139
Now (nation-based year-to-year randomness):
Man City - 140, 110, 115, 130, 145
Arsenal - 136, 107, 110, 119, 142
Inherent randomness (~30%):
Man City - 140, 110, 115, 130, 160
Arsenal - 158, 181, 132, 105, 116
That's if nation-based year-to-year randomness is what is actually going on here. As I said, I'm quite hazy on it right now. I'll probably try and get to the bottom of it sometime down the track. I know this all seems to put things in the wrong direction of progress, and people generally don't like that, but better to acknowledge the setbacks/problems than keep them on the down low and end up fooling myself/others about it.
Oh I didn't realize you're the guy who is obsessed with me. It's been half a year and you're still going on like this. I wanted to communicate to others the points I made anyway.
BaZuKa said: Ok this new file is Giga Broken
Season 3 almost won everything with a low-tier club from Portugal
The first line sunk me for a moment there. It's good to hear it's working well for you, I don't actually know how well in reality these files are going to go.
I keep forgetting to mention things. I used HarvestGreen's 6 > 18 attribute data mainly this time, as using 1-20 overvalues things like pressure and work rate. I figure in cases where pressure or work rate is very low, you can either just filter out those players or train/tutor them up a bit if you do buy them. And getting '20' is less likely as well as more CA-inefficient, so that's another reason I favor the 6>18 measure.
Just piggybacking off this to say that I haven't used the player names/clubs to calibrate my values. There's a good reason for this, and that is that in-game player ratings don't correspond exactly to actual performance.
You can see evidence of this in this OmegaLuke video, where technicals give substantially higher player ratings than physicals, even though physicals actually won all the games. Orion's coefficients, which use in-game player ratings to deduce the best attributes, I've found are outdone by HarvestGreen's data which assesses according to goals scored or games won.
So this is why I don't try and align the values to fit player ratings or the best players, as that would achieve the opposite of what I want to achieve, which is show players who punch above their weight in a way the in-game AI doesn't recognize.
But I think there is some value in comparing the results afterwards, just to make sure one isn't completely off-track. Regardless of rating, we know Haaland gets goals, so if he's not up there then there's something amiss.
Additionally I think it's notable how the starting data closely aligns with physicals matter and technicals don't. If you plug in HarvestGreen's data, it just so happens that the top players start in the top clubs in the game. That sounds as straightforward 2+2=4, except realize that this means that SI knows exactly how the attributes are skewed and disingenuous. If they believed what they tell you about how the game works, then we should see these initial players at top clubs failing to perform. In fact, I think perhaps this in-game rating bias towards technicals is to try and stop AI managers simply buying up completely lopsided physical beasts as the game goes on.
I can think of a reasonable counter-argument or two to the above, but there is also some circumstantial evidence as I see it. If you use Genie Scout default ratings, 7 out of 9 top players (one of each position) are white. If you use HarvestGreen's data it changes to 7 out of 9 being black. Either SI still has a racism problem where black people are portrayed as mentally/technically poor physical beasts, or this is intentional. Or both.
LightningFlik said: 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.
(No rush, whenever you have time).
OpticFawn said: Yes please, I want to use on personal tool pls
Here are the values for FM24 blended.
I forgot to say that 'sweeper' position is intended to be my attempt at a tutor rating, and for Target Striker I've simply upped jumping reach to 100, as high pace/acc ST is simply better than a slow target man and I'm not sure if heading or whatnot affects the target man's performance (jumping reach certainly matters).
https://files.catbox.moe/owbw3u.grf
FM24 Age 26+/Pure performance:
https://files.catbox.moe/8wskjk.grf
FM24 Blended (recommended):
https://files.catbox.moe/f8hmj7.grf
I haven't done FM26 versions yet and it'll be a while before I do, I recommend just using the FM26 ratings file I did before and replacing the GK ratings with the new values in FM24 Blended. You could use these new files in FM26, but it would be a bit off (still much better than FM Genie Scout default).
I'm increasingly unsatisfied with the result as I've ran into further issues, but I've decided to just put these out there otherwise it'll never get done and it does seem to work better despite the issues.
I've had a dilemma about merging in some of my own findings. One example is that in my 1 CA player testing I found that concentration matters on CB/DL/DR, not other positions. Another is that extra finishing on CBs made no difference to goals scored, which means it's low weight on CBs for a reason. These findings are a part of my previous ratings file, so I wanted to include them again here. But I also figured that my data isn't as strongly evidenced as HarvestGreen's, and some of my conclusions could just simply be wrong. On the other hand, there were a number of different examples which can't all be wrong and it makes sense that different positions benefit from certain attributes to different degrees, so I don't think it's best just to use HarvestGreen's data flatly amongst all players. In the end I decided to just apply a simple -/+25% to a few attributes that I was quite sure had pronounced positional differences. I.e. dribbling 32 > 40, vision 5 > 7. So nothing too drastic, but this leaves me feeling like it is too arbitrary a change and yet it also doesn't go far enough probably. The only reassurance I have is that the actual results seem to be better, which you can see a sample of below.
Another issue I've had is with the new GK data. Initially I just plugged in HarvestGreen's new figures. But looking at the results, I get the impression something isn't quite right with the new data. And if you look at HarvestGreen's old GK data, which only had conceded goals as a measure but was also measuring 10>20 instead of 1>20, you can see there are contradictions. From my own 1 CA player tests, I found that acc & pace seemed best value around ~5-8 rather than 20. And if you look at actual players in the database, a bunch of the best have acc of just 8-10. So I think this is mainly a case of missing the data that shows where ~6-14 is sufficient, which we know some attributes such as work rate are like. So I made some appropriate adjustments. Definitely an improvement on my old ratings file here.
Here are some example results. Two important differences that stand out to me are that the GKs are significantly more correct in the new version (Pickford shouldn't be behind Forster!) and there are no longer 2 japanese players from Celtic in the top 10 ST list (they do have high acc/pace, but this seems an appropriate change). And clearly with Kane as no. 1 ST, FM Genie Scout default ratings are way off.
New (blended):
Pickford 69.53%
Pope 67.67%
Ramsdale 67.43%
Steele 65.41%
Old:
Pope 69.74%
Forster 65.89%
Pickford 65.13%
Ramsdale 65.00%
HarvestGreen (new GK data):
Pope 75.44%
Pickford 73.89%
Ramsdale 72.98%
Butland 72.85%
Actual England selected:
Pickford
Pope
Ramsdale
HarvestGreen data:
Haaland 95.34% (Man City)
Mbappe 93.59% (PSG)
Osimhen 93.39% (Napoli)
Vinicius 92.70% (R.Madrid)
Thuram 90.54% (Inter)
Nunez 89.32% (Liverpool)
Isak 88.71% (Newcastle)
Moffi 88.48% (Nice)
Martinez 88.32% (Inter)
Jesus 88.28% (Arsenal)
New (performance):
Haaland 95.25% (Man City)
Mbappe 93.68% (PSG)
Osimhen 93.65% (Napoli)
Vinicius 92.47% (R.Madrid)
Thuram 90.43% (Inter)
Nunez 89.29% (Liverpool)
Moffi 88.57% (Nice)
Isak 88.52% (Newcastle)
Lukaku 87.99% (Roma)
Maritnez 87.51% (Inter)
New (blended):
Mbappe 84.94% (PSG)
Vinicius 83.95% (R.Madrid)
Haaland 82.94% (Man City)
Osimhen 82.80% (Napoli)
Thuram 80.15% (Inter)
Nunez 80.14% (Liverpool)
Jesus 79.15% (Arsenal)
Isak 79.09% (Newcastle)
Messi 78.83% (Inter Miami)
Martinez 78.65% (Inter)
Old:
Mbappe 77.59% (PSG)
Haaland 77.25% (Man City)
Salah 76.89% (Liverpool)
Vinicius 76.86% (R.Madrid)
Martinez 75.23% (Inter)
Son 73.80% (Tottenham)
Furuhashi 73.10% (Celtic)
Nunez 73.08% (Liverpool)
Maeda 72.92% (Celtic)
Osimhen 72.65% (Napoli)
Default Genie Scout:
Kane 91.47%
Haaland 90.95%
Messi 90.03%
Lewandowski 88.43%
Mbappe 88.30%
Benzema 87.21%
Salah 86.97%
Son 86.51%
Martinez 85.91%
Vinicius 85.71%
I would be interested in your results over a 10-20 year span. At least in Fm24 and all previous versions if Game Importance wasnt set to very high in a decent Youth nation you couldnt produce NewGens even if your were #1 club in the world.
Singapore could never have good players because there youth rating is bad. Czech would be a good test as their youth rating is 90-100+ while game importance is not set to very important.
sortitoutsi even has always listed game importance for finding newgens in their charts.
https://sortitoutsi.net/football-manager-2026-youth-ratings
I dont think this is an issue for most players since most people play in a big nation, build a nation saves arent that popular in the community. So most people wont notice it.
I am honestly probably not explaining myself well. I just know in fm24 and prior when doing build a nation saves even if you changed youth rating to 100 the nation couldnt ever produce wonder kids even when getting league to top 5 and all the clubs in the top 50 if Game Importance wasnt set to very important.
I dont know if there is some kind of time factor and that is why you need to sim 10-20 seasons or what tbh.
I should have mentioned that I used FM24. I guess some will say then what's the point of doing an updated test if you're not going to even use the most recent edition, but for me FM26 just has to be foregone.
I actually completely forgot to do one thing I wanted to do this time, which was to test what it looks like 5-10 years into the future.
I wanted to do USA initially, to address your claim about the MLS exactly, but found newgen intake works a bit differently and I couldn't be bothered trying to work it out. So my plan after that was to choose a nation with low game importance but high youth rating, but there aren't really any that are loadable - and singapore was one of the best options that was left. But this is also why I did Man City, because I had in mind (and generally what people play with) is clubs with top facilities in top nations, and England is no.1 for that.
As I was writing this I was doing my testing, and I was about to say that I've been wrong about game importance all these years, as 5 years in, the Man City (unimportant England) median had slipped to 117.5 PA (155 peak) which is outside of the range one would expect for the 131.166 PA averaged median I had got for the initial year. But then I did the control test (normal England), and that was 122 PA median, and 115.5 PA on the second sample.
This is bringing back what I experienced regarding game importance back when I initially tested it in FM19. No one has so far pointed out that I said game importance has no effect, yet later claim it has a minor effect, but I want to clarify this. As you can see, game importance has a minor effect that is only sometimes apparent. When I started testing it, I used average PA as a measure and so it was invisible to me and I thought it had no effect. Once I cottoned on to median PA being more accurate, I observed that questionable minor effect. In multi-year tests I thought I had borne out its effect, only to be confounded again in the same way this time. So I had to conclude whether a handful of PA difference, some of the time, constituted something real or an illusion. I concluded it was illusory, although the bunching up around the median effect is obviously real. It's a bit difficult to explain this fully if you haven't tried this kind of testing yourself, but basically it's very easy to be hoodwinked by limited samples and the inherent randomness. If I test a club and get 139, 134, 140, 137 then it's very tempting to say this club is a ~137.5 PA club. But then I would get 153 and 145 if I took 2 more samples. So this is why when I saw game importance raising PA by at most ~10% and inconsistently, I put it down to randomness.
I will continue this testing a bit, to clarify results at the 5 year and 10 year mark, as it does require at least 3 (preferably 5) samples to give any reliable indication.
We know it has always has some value, because high CA-PA gap = faster training = higher pace/acc faster.
Therefore a zero CA cost attribute such as 'pressure' that has a significant effect on win rate, is even more valuable than an equivalent attribute that costs CA, say dribbling. But by how much exactly? 2 points of pace and acceleration each perhaps? I prefer to go on the safe side and guesstimate 1-2 points of just pace say (I think about what gets 'sacrificed' to stay within PA limit at the end of ~4 years of training).
But then you have positions such as DM where you can easily max out pace/acc to 20 without hitting the PA cap, at least if you're in a good division (which I think most players are in, or plan to end up in).
So I figure the bonus for DM should be reduced, and conversely the bonus for a CA-tight position such as AML should be increased.
But this leads to the peculiar consequence that positions that benefit most from high pace/acc, such as AML, value them least. Not to mention the fact that lowering the immediate gains (from high pace/acc) for expected future gains that may never even eventuate. And then there are also inherent variables whose expected ranges exceed the capacity of this predictive method. For instance, even assuming everyone uses meta training, one player might go from 15>17 pace, another will go from 14>20.
So in the end what I decided is, I will have a set of values for youth/optimization, a set of values for age26+/team selection/pure performance (which will be very closely aligned with HarvestGreen's findings), and then a blend of the two - which will be the file I recommend to use as switching between files is tedious. Is a 50/50 blend the best? Probably not, but it's the best I can do so far. I have checked in genie scout the actual results of these new values, and it seems to be working as it should - the best players are those from Man City, Barcelona, Real Madrid, etc. as you'd expect. I've looked for outliers that have changed positions the most, and overall I'd say the margin of error could be something like -/+3% genie scout rating, which I think is satisfactory.
Here's an example to give you an idea about things:
Default Genie Scout - Kane 91.47%, Haaland 90.95%, Mbappe 88.30%
Orion's Coefficients (not my file) - Haaland 88.01%, Mbappe 86.31%, Kane 81.63%
My existing file - Mbappe 77.59%, Haaland 77.25%, Kane 67.35%
New file (youth/optimization) - Mbappe 79.73%, Haaland 75.69%, Kane 65.64%
New file (age26+/pure performance) - Haaland 95.34%, Mbappe 93.59%, Kane 82.21%
New file (blended) - Mbappe 85.33%, Haaland 83.74%, Kane 72.46%
Ignore the numbers themselves, it's about how relative they are to each other
Recently, harvestgreen22 released a new test for GK attributes. Can you convert the data to your player rating coeff? @GeorgeFloydOverdosed
I've actually been working on this recently
Initially I was just going to redo the GK, but then I realized I want to take a new approach to the whole thing.
I've ended up not entirely satisfied with the new approach, basically I think it's definitely better but even more arbitrary in a way. It's mostly to do with balancing CA weighting vs. raw performance, and the problematic question is if an attribute gives a slight boost to win rate but on the other hand also only has very low CA weight (or zero CA weight, like personality attributes), do you give that some kind of bonus and if so to what degree? Basically, how much is free CA even worth exactly. So I've come up with a take on this, that also integrates training results with good accuracy (i.e. if meta training is expected to lower passing by 1 over 4 years, I adjust the passing value slightly accordingly - this is more pronounced for pace/acc which will increase by ~4 in youth over 4 years!).
It's largely done now, just have to finish off some positions I don't use like WBL/MC/AMC and also create a FM26 adjusted version before I post it. Will be soon.
I can tell you my last 3 build a nation saves were hungary before I knew about game importance, Andorra and Faroe Islands. Hungary even after 60 years where I gave every club billions and billions to max out their facilities could never produce a NewGen that was as good as IRL national club member. The way you give clubs billions is by buying their horrendous youth players for 10 million each as the board never turns down 10 million offers. So you are buying 30+ players per club eventually. Giving clubs 300 million each season.
Coincidentally it was a recent video by SecondYellowCard where he tested the DoF role that got me thinking that youth recruitment might work similarly. Turns out it likely doesn't, but that's what led me to investigate the 'place of birth' thing.
I haven't been able to find this discord you mention yet, but I've done some testing of game importance anyway, not only to draw a conclusion here but because it's good to retest things once every few years to see if the mechanic has changed and also I'd like further clarity/precision on the distribution effect I mentioned. I retested recently the hidden nation factor thing I assert is the case, and I can confirm it still exists.
'Game Importance' Test results
Singapore unimportant (default) LC sailors:
Sample 1
75,68,57,51,49,49,48,42,42,39,37,36,32,31,31,20
median 42
average 44.1875
range 20-75
height 1.59-1.90, 1.73 median
singapore-wide stats:
range 17-80
10 best player = 73 PA
Sample 2
74,73,71,68,60,54,53,52,51,41,37,35,34,34,30,21
median 51.5
average 49.25
range 21-74
height 1.63-1.87, 1.715 median
singapore-wide stats:
range 15-86
10th best player = 74 PA
Sample 3
89,67,64,64,62,61,59,50,50,48,46,39,34,32,29,20
median 50
average 50.875
range 20-89
height 1.63-1.92, 1.725 median
singapore-wide stats:
range 13-89
10th best player = 69 PA
Singapore very important LC sailors:
Sample 1
59,53,50,48,47,46,45,44,43,39,38,36,32,31,21,21
median 43.5
average 40.8125
range 21-59
singapore-wide stats:
range 12-80
10th best player = 74 PA
Sample 2
78,70,66,66,61,60,60,53,48,46,41,39,38,37,20,17
median 51.5
average 50
range 17-78
height 1.64-1.92, 1.725 median
singapore-wide stats:
range 10-80
10th best player = 70 PA
Sample 3
53,52,48,48,48,47,44,44,44,44,43,31,31,22,20,20
median 44
average 39.9375
range 20-53
height 1.63-1.93, 1.785 median
singapore-wide stats:
range 8-115
10th best player = 71 PA
Singapore unimportant (default) LC sailors average (3 samples):
median 47.833
average 48.104
range 20.333-79.333
singapore-wide stats:
range 15-85
10th best player = 72 PA
Singapore very important LC sailors average (3 samples):
median 46.333
average 43.583
range 19.333-63.333
singapore-wide stats:
range 10-91.666
10th best player = 71.666 PA
England very important (default) Man City:
Sample 1
167,162,160,152,146,144,143,142,137,136,133,128,117,94,83,83
median 139.5
average 139.1875
range 83-167
england-wide stats:
range 30-179
10th best player = 157 PA
Sample 2
172,167,157,155,148,147,146,142,141,134,115,98,97,96,77,64
median 141.5
average 134.75
range 64-172
england-wide stats:
range 30-178
10th best player = 157 PA
Sample 3
178,167,165,160,157,155,148,147,144,142,141,104,89,85,75,58
median 145.5
average 132.1875
range 58-178
england-wide stats:
range 30-180
10th best player = 158 PA
England unimportant Man City:
Sample 1
167,157,149,148,146,145,140,137,117,111,104,98,97,90,85,81
median 127
average 135.75
range 81-167
england-wide stats:
range 34-175
10th best player = 154 PA
Sample 2
160,157,148,148,148,138,136,133,123,122,101,101,92,88,87,75
median 128
average 122.3125
range 75-160
england-wide stats:
range 34-168
10th best player = 154 PA
Sample 3
159,156,153,151,150,150,141,140,140,140,138,116,110,95,93,80
median 140
average 132
range 80-159
england-wide stats:
range 34-168
10th best player = 153 PA
England very important (default) Man City average (3 samples):
median 142.166
average 135.375
range 68.333-172.333
england-wide stats:
range 30-179
10th best player = 157.333 PA
England unimportant Man City average (3 samples):
median 131.666
average 132.687
range 78.666-162
england-wide stats:
range 34-170.33
10th best player = 153.666 PA
Conclusions
Game importance has no strong effect on PA. Previously I had said it bunches up the PA around the median and ends up affecting the PA to the tune of ~10%, and this is what we see. But as I tried to communicate in qualifying that 'it's not directly comparable to the other PA factors so it's hard to pin down a precise figure', it's not a simple flat 10% - in Singapore's case, the effect was 0% or even negative. In England's case, it was an 8% difference, for both Man City and (to a similar degree) the nation as a whole.
The main takeaway from game importance as I see it is that because it bunches up PA around the median, it does actually dampen your wonderkid chances just enough to make one not want to dismiss it entirely, but it's not that significant or consistently applicable as other factors such as junior coaching are. 160 PA instead of 170 PA, in some cases.
I haven't done a deep analysis of it, but just my general impression is that its not that Game Importance is shifting up or down the general quality of the newgens either, and this is somewhat indicated by the low difference in the average as well (2% difference for the Man City samples). I think it's simply bunching up the distribution closer to the median.
I hope this also serves as an illustration of why I use the median instead of the average. The median is pretty stable and therefore predictable unlike the average. While the median can be predicted with strong likelihood of being within a range of -/+ 5 PA or thereabouts with enough samples, the average will end up with an uncertainty range of ~10 PA or more even if you collect many samples.
I included height in the singapore samples as I thought it was interesting to observe; I wondered if its distribution has some correlation, or something in common, with the PA distribution.
This whole thing appears to be purely cosmetic, so don't try and work out how to get better newgens with it. There is no actual competing for newgens, or even newgen generation in these cities, going on.
There are three clubs: Ljubljana, Kamnik, and Maribor.
The rectangle is area the club can draw newgens from.
It is not that newgens are generated at each city and a few of them get picked up, it is that the club generates precisely 16 newgens. There 'place of birth' is probabilistic based on distance from club and 'inhabitants range' of the city. The circles are just to convey the idea of decreasing probability the further you go out from the club location.
Notice how Ljubljana and Maribor each draw ~4 players from their own city, but Kamnik only draws 2 and takes many more from nearby the nearby capital of Ljubljana. This is because Kamnik has low inhabitants range, and Ljubljana is high probability because of proximity + high inhabitants.
I do not think city 'attraction' affects it, I have tested it but the results are not 100% clear but clear enough to rule it out I think. From memory, 'inhabitants range' is also relative to 'inhabitants range' of other cities. That is to say, if Ljubjlana only had 10,000 people, it would still be top dog if all other cities are 1000< pop, but also it would be less commonly the place of birth than if it had 20mil people and others 1000< pop. There are some further nuances that reveal themselves when you try to break it with extremes like this, but since the mechanic is cosmetic, I won't go further into that.
Lastly there is the matter of exceptions. As you can see, sometimes there can be instances outside the rectangle. I don't know what exactly is going on here, I suspect it has something to do with youth recruitment perhaps. Maybe it's even just a randomness factor they've put in to try and better reflect reality.