I inputted my player template that finished ~5th in the Premier League in FM24, into FM17, and got 14th with the 'GOODBYE 442' tactic. But that meant the good results were within reach, so I increased pace & acc from 15 to 16, and got 8th:
So it seems that most attributes not mattering is true in FM17 as it is in FM24. Perhaps not exactly the same, but not far off in any case.
I suspect the difference between 14th/8th for FM17 and 5th for FM24 is actually the quality of the tactic. And perhaps it is the case that the introduction of gegenpress, and/or just the incompetence of the new match engine team (Nic Madden who joined SI for FM16 replaced Paul Collyer as head of match engine/match ai from FM18 onwards according to ChatGPT?), permitted for more overpowered tactics.
So based on this, it seems like player attributes in FM have always been misleading BS, but perhaps at least in some earlier versions, it was adequately obscured by the constraints of the tactic, which is overall more consequential than attributes themselves (though it won't save a dud team, and the ratio seems something like 70/30), and were either more balanced before or it simply wasn't yet exploited by people like Knap fully. Personally I reckon it's both things together.
This may actually be quite useful to know, because if the underlying attribute mechanics have always been the same, and they only have some degree of control over the tactic impact, it seems to me to mean that FM26, FM27, FM28, etc. are either going to have strong exploit tactics, in which case the attribute meta works the same and we can predict that just by looking at the exploit tactic degree of success, or tactics have been flattened down, in which case the attribute meta still applies but now you need more of those lesser attributes in the meta (i.e. anticipation, composure, dribbling, etc.) in order to actually win the season. And they can't go too far down the 2nd route, because then players would complain about not being able to win anything. That's my theory anyway.
A4 said: I only heard that traits don’t really matter much in FM, but I’ve always noticed how good players with “tries tricks” are. I even started teaching it to everyone in my youth academy. Of course, it’s anecdotal evidence, but when youngsters learn it—or even just try to do it—their impact and match ratings seem to be so much higher.
I’ve also noticed the same thing with the Light-Hearted personality. Somehow, good players with that personality seem to perform so much better for me than players with other personalities.
Is this something anyone has ever tested before? Expand
So I've tested now Osimhen having both 'shoots with power' and 'plays one-twos' as ST, as these are the two preferred moves that supposedly had the best benefit in Stybb's testing.
Victor Osimhen default - 3rd, 3rd, 7th, 3rd, 3rd, 4th, 3rd, 4th = 3.833 position Osimhen w/ shoots with power, plays one-twos - 5th, 6th, 6th, 3rd, 5th = 5 position
I don't think there's enough samples to conclude the preferred moves made him worse. I think it's unlikely, but it is possible since it could be overriding his role in the Knap tactic.
What we can be sure of is that it definitely hasn't made him better.
I picked Osimhen because I already have test data on him in a fairly controlled team, he's one of more consistent players of the players I have data on, he's a good ST but not overly good (we want some headroom for any improvement to shine through), and he's a well rounded player that excels in pace & acc & finishing.
bf3metro said: any FMSS update? please, thanks mate! Expand Busy in life for a few days right now, but I'll get round to it probably after that
OpticFawn said: what values would you put for Injury Proneness, natural fitness, dirtiness and aggression as simple values on other platforms? Expand You could input the values I give for Genie Scout
The relative importance of those attributes are:
Dirtiness = Very important Injury proneness = Moderate importance Natural fitness & Aggression = Minor importance
LightningFlik said: Why is the left foot worth more than the right? Expand It's just something I added in to toy around with later.
To compare the relative impacts:
left20/right6 vs right20/left6 = 20 difference 1 loyalty vs 20 loyalty = 19 difference
That's ~0.2 pace.
Thinking about it now, I should probably remove the left foot preference for the next version. 0.2 pace is small, but not exactly nothing.
I tried thinking of a valid reason to keep it:
Even if left vs. right makes no performance difference, when it does in real life, there are fewer left-footed players in the database, and we need some left-footed players for set pieces so we should skew slightly towards them.. but then again, set piece attributes are proven to be useless.
I guess the only real testing I can think of when it comes to preferred foot performance is HarvestGreen's test, and while his results show that foot side matters depending on position, I don't think we have any real information on left vs right in a central position such as ST or DC.
I'm not quite sure what to make of Yarema's above 2 posts to be honest. Seemed like a jab to me. I figured he is passive aggressively saying, if this is the unrefined method with which you handle this aspect of your data, then how can we trust your handling of the supposedly 'objective' position data? In other words, conflating the assessment of error of the weightings, with the assessment of error of the position data. He seems contented in having muddied the waters.
Maybe I'm over-interpreting his vaguery, but nonetheless I'm going to clarify it if he won't.
For the position data, ChatGPT gives me the following 95% confidence intervals:
Things get hairy for the handful below Dybala.
It surmizes it all overall as:
Minimum ±0.38 1st quartile (Q1) ±0.82 Median ±1.38 Mean ±1.52 3rd quartile (Q3) ±1.69 90th percentile (approx.) ±2.50
Now the color coded system I used for the % figures, which represents the best fit of the attribute weightings to the position data, is just to show which players evidently require further attribute weight adjustments in order to be reasonably consistent with the position data. There will likely always be outliers, so I also wanted to show just large this group is - it's currently ~12-15%, which I think isn't too bad.
As for the confidence interval of the % results, it took a while in ChatGPT and I'm not entirely sure I've done it correctly, but it should be this:
95% prediction interval is ±9–12 percentage points 80% prediction interval is ±6.3-7.7 percentage points
This should take into account both the position data uncertainty and the best fit uncertainty of the % figures.
If the % was fitted to the average position data perfectly, it would have a 95% confidence interval of ±3.6 percentage points. And that itself could be improved with more samples taken.
Yarema said: I was planning on staying out of this thread, but ... you're actually calculating error rate by "I checked 32 players and 4 didn't fit my model"? Expand Statistical accuracy of my player rating system
I compared my 60–90% player ratings against the objective average positions from 32 players.
Results:
Pearson correlation: −0.80 — strong relationship between rating and objective position R²: 63.8% — the rating explains approximately 64% of the variation in position Unexplained variation: 36.2% Regression slope: −0.241 positions per percentage point — higher ratings predict better positions MAE: 0.73 positions — the rating misses the objective position by ~0.73 places on average RMSE: ~1.05 positions — larger errors are penalised more heavily Normalised MAE: ~15.5% — average error relative to the mean objective position
In short: the rating system has a strong relationship with the objective rankings, with an average error of around 0.73 positions.
For your forum, I'd describe this as:
Using a ±5 percentage-point tolerance, 5 of 32 players (15.6%) fall outside the expected range, meaning 84.4% of the ratings fall within ±5% of the regression-derived expected value.
Thanks ChatGPT!
2017 called, they want their 'Do you even have a PhD in Statistics?' line back.
mytreds said: Would any of the advice in the OP work in older FM titles? Expand For attributes, too hard to say, but I think it's more likely things have stayed the same than that they have changed. I suspect it would largely hold true going back to ~FM18 at least, but that's just my guess.
For everything else, it should be true going back a long way.
Ndour17 said: Reading back through this thread's evolution, a few points seem worth flagging clearly — not to dismiss the overall contribution (the physical-attributes-dominate thesis holds up well against Zippo's isolated A/B test and skawkclsrn's large-scale randomized study), but because several specific claims here don't hold up as well as they're presented:
The "position proficiency caps at 18" claim (repeated with confidence early on) turned out to be wrong once tested — worth flagging since it may still be floating around in older posts/quotes. The Genie Scout personality-weighting bug wasn't caught internally — it took LightningFlik's independent tool to demonstrate the Osimhen-over-Haaland inconsistency. Worth being more explicit that GS output shouldn't be trusted at face value for hidden attributes even now. The "Blended" file and the "15 pace/acc realistic template" contradict each other on your own numbers (the Newcastle-vs-synthetic-squad comparison) — this was acknowledged but never really resolved, just reframed as "the file is a rough guesstimate." That's a fair characterization, but it should probably be stated up front rather than discovered by a reader cross-checking the math. The "attribute thresholds are relative to competition level" theory, used to explain away discrepancies with ykykyk05251/Zippo/Orion/HarvestGreen22, has never actually been tested — and skawkclsrn's much larger randomized dataset found no thresholds anywhere, which cuts against it rather than for it. A lot of the decimal-precision weights (Pressure=52, Determination=56, etc.) come from single small-sample runs, some explicitly labeled "pure speculation" or "I don't remember why I set this" in your own posts. That's fine as a starting point, but the precision of the numbers implies a confidence level the underlying data doesn't really support.
None of this undercuts the value of the overall synthesis — it's clearly the most complete one on the forum. But the granular numbers probably deserve a "confidence level" caveat next to them, given how often they've had to be walked back or reframed. Expand Yes, these are good points, and most of them I still keep in mind, but it can be difficult to remember everything. I can see you've done a close reading, because I know some things here that I think everybody would have skipped over.
I will clarify a few things:
1) The hiddens bug with Genie Scout indeed remains unresolved, so I highly recommend to use FMST26 or FMSS. I choose to put out a GS file anyway because there are still useless features it has that the other two lack, plus its the established tool. I halved the personality weightings in my GS file to try and temper whatever inaccuracies arise, as personality turns out to be relatively unimportant for performance, but they remain included because I have to keep injury proneness and dirtiness in there anyway - there's no way around it. I'm no longer doing the ranking-to-weights match up work in GS, but I was initially, so this should bypass the bug issue to a degree, though now I do the weight work in FMST26 so it will probably drift more and more away from being accurate in GS. Long story short though, I think it's viable to continue using GS.
2) Pressure, determination, I think perhaps aggression, and no doubt a couple of others that are mainly 0 CA attributes were never properly refined in my template tests. So I tested determination at 14, but unlike many/most other attributes, I didn't get round to trying it at 13, 12, and so on. So in theory an attribute like aggression could be anywhere between '1' and '11' as a requirement. It's set at 11 because I believe it matters, unlike say passing, based on either other people's tests, or my own previous tests (say the 1CA tests). I figured at the time that since these are 0 CA cost attributes, or necessary for player growth anyway (i.e. determination), then it's not a priority to pin them down like I pinned down agility to 10. I ended up never getting around to completing it thoroughly.
3) I am pretty dead certain there are attribute thresholds relative to competition level. HarvestGreen22's data shows this clearly. It's what I have observed, even now with the real player rank data albeit its more convincing than definitive - Haaland comes no.1, but he has work rate of 13. Luis Suarez has godly technicals & mentals, but very low acc/pace/sta, and we see he is one of the worst performers. Those are the obvious examples, but I've seen many more subtle ones that are nonetheless identifiable. All that said, work rate does appear to be of importance even beyond ~11-13, and we see similar things for stamina and anticipation and whatnot, so while I'm sure there are attribute thresholds, the understanding of the curvature could do with refinement. Part of it could be that certain things kick in at certain levels. My observation with GKs is that at a mid level, reflexes matter less, but reflexes seem to take precedence at a high level (~15+) while aerial reach is essential at mid level but is significantly less essential at a high level (~15+).
4) I don't recall the Blended file vs template problem. And I'm not sure which came first, so I can't assume it's that the template invalidated the Blended file. All I can say is that when I did my best attempt at an objective comparative test of all the files, the Blended one did poorly. From memory, the Blended file was mainly based on HarvestGreen22's data + consideration of CA weights + consideration of player growth. I was also still using positional differences at that time, which now turns out to be bunk. As a point of history now, we can go back and ask 'Well, was the Blended file actually a step backwards from even the first file, and therefore do you regret putting it out there?'. At least that's the question I ask myself. And for the other files, I can say, 'they were better than the first file, but perhaps only marginally in hindsight'. Actually was striking to me was how although the improvement was marginal, the improvements of my files and those of others appear remarkably gradual and consistent. I would have expected things to be more over the place. And I could be wrong, but I think this latest method and the weightings derived from it are a big step forward.
5) I think it needs to be restated the current precision of the attribute weights does not reflect the actual real influence/value of the attribute. I'm merely moving numbers up and down to get a best fit of the data. It's like wedging a piece of wood inbetween two metal joints to make the leaning tower stable. It's not the right material, but if it works, it works, and we can replace it with the nuts and bolts later.
Recall that 87% for ST is 2.571 and 73% for ST is 6.111, and that positions other than ST appear to be consistently ~1 position lower.
Every position for the Knap 424 formation has been sampled now, and I'm satisfied that the same set of required attributes apply to all outfield positions in the same way.
I've done some readjustments to the weightings.
I've dropped Sterling the equivalent of ~1 position, but this means he should we should expect no worse than ~4 position, yet his average is 5.375 if we give +1 for not being ST. This is the best I've been able to do.
Non-ST positions seem more comparable to ST now (less unexplained position difference), and stamina is now a more reasonable 40 instead of 94 weight - stamina seemed to be the partial cause of Sterling's overrating. Quite a few players have pleasant coincidental changes, such as Maeda going higher and Gyokeres lower, which is more line with the real rankings.
However some changes had to go with it, such as Vlahovic dropping a little below Messi. I figure it's more likely that Sterling should drop 1 position, than Vlahovic should not drop 0.5 position. And given most players seem more to be more accurate now, I'm not too concerned about Vlahovic.
Next step is translating to Genie Scout weights, then I can upload the updated file. Main reason for updated file will be adding the new GK weights of course.
Rain said: What am I missing here why do you think Mbappe and Sterling should be closer in terms of results? I would assume the +5 Pace, +3 Acceleration, +3 Dribbling, and +5 Determination alone would speak for itself. Expand The method I use is that I compare two players in conjunction with a few others, to rule out most of the attributes.
So for instance, let's say Mbappe is better because he has 20 pace/acc, then why does Haaland actually perform better? Then we find someone with same pace/acc as Haaland, but performs significantly worse. This player has -5 ant, -3 det, -3 work, -7 tack. We can rule out tackling off the bat, but how do we know how the remaining 3 attributes contribute? We compare two other completely different players to each other and one has +4 ant -2 work and the other has -4 ant +2 work, and find the +4 ant -2 work one performs much better. So now we know it's likely ant is the key missing factor, or at least contributes that precise position difference. We then verify that this would line up the rankings more correctly instead of no change or worse.
So that process gets us to the final weights, which we know are close to valid because almost all the players line up well without contradictions, and according to the weights Sterling should be closer to Mbappe.
If you compare Sterling to Mbappe, it's no problem, but what if you compare Sterling (6.6) to Saka (2.75):
There are a few differences, but nothing major. We know all the technicals are useless, aside from dribbling. That leaves a handful of minor to moderate differences in mentals, balance, and pressure (hidden), as the possible causes.
And the thing is, he's pretty well rounded in every attribute, so there's not obvious non-linear threshold being crossed for a certain attribute. I thought perhaps composure, but my template says 11 is the minimum, and he has 12. Thinking perhaps pressure right now.
A4 said: There are some weird results in this video, I don't know the guy and how trusted his test is but it's way different than pretty much all the actual results I saw
Expand Coincidentally I just made a comment on this video before coming here and seeing this post.
I was actually going to bring him up in my response concerning traits to you, and I'm guessing you asked it perhaps because you saw his video on traits too.
The jist of my post was that either this guy has something seriously wrong with him, or I'm missing something. Hence, another reason to give traits another test.
I first came across him ages ago, about 3 years ago, in which he claimed to do a proper test that showed scout JPA has a strong effect on newgens, which I knew from my own testing to be absolutely false. I left a comment about this, but he didn't respond.
Fast forward to now, and these past few months he's supposedly running and pumping out all these in-depth controlled tests with a discord community to back him, but if you go to it, it's a ghost town. Strange. And the tests themselves, obviously fly in the face of the results derived by others doing similar testing.
But I realized today it could be that it's because he's assessing by goals conceded and goals scored, which is not an adequate measure. I keep forgetting myself that final season position is the only reliable measure, at least that I know of. Still.. the scout JPA result can't be explained by this, so I simply don't trust the guy on anything. Maybe when I do the trait test, I'll also check to see if it affects goals difference as he claims it does.
I decided to leave all this out of my post before in the end, because obviously it sounds obsessive and veering off from what you asked, but its what came to mind. Occasionally youtubers can be a source of fresh good info.
75% | Kurt Zouma (DC) - 7th, 5th, 6th, 3rd = 5.6 position 73% | Sean Goldberg (DC) - 5th, 16th (sacked), 8th (sacked), 5th = 8.5 position
Notes:
We see with Mbappe that the weightings give the same result, whether Mbappe is playing ST or AML (note: in each test he has been edited to have 20 proficiency in only 1 position).
Kurt Zouma is a player keithb's weightings has at rank 7th in world, even though he's playing at West Ham. So I thought it would be a good candidate to test, especially since Zouma has 20 strength, 18 jump, 17 heading, 15 tackling, 16 determination, and so on - everything that lines up with what has conventionally been thought to constitute a great central defender. The result says otherwise. He is equivalent to Malen or Glatzel at ST (4.666-6.111 position; 73-74%).
Goldberg I elected to test because he has 12 acc/pace/jump yet has a decent rating. We see here it is viable, though the position seems to have slipped ~2 places. Though this could also be just because the early sacking 16th position result.
Raheem Sterling's result is surprisingly low. We see from Mbappe that it is not a disparity caused by the position itself. And keep in mind that he is conventionally found not far from the top of GS ratings (keithb has him 11th for AMR). When assessing this kind of situation, I've come to find that it's about what specific attribute the player lacks, rather than what attribute or attributes are over-weighted. Comparing with Mbappe narrows it down to comp or det it would seem to me, but I wonder if it could also be that Sterling is playing on the right side with a right foot. I recall HarvestGreen found some stuff about this. So I will probably test to see if side matters.
lasko911 said: I'm curious to see the DC testing results because I've had some interesting anecdotal evidence across several of my saves.
First I was using DCs with as high Jumping (17+) as possible, but only decent Acc/Pace (13-14) and results were mixed. Then I switched to DCs with high Acc/Pace (16+), but completely ignored Jumping (sometimes it was as low as 5).
The second type of defenders performed much better, their speed seemingly made up for their complete lack of aerial presence. Expand That could actually be because pace/acc is as crucial, or not far off it, as it is for STs.
When I reduced DC jump to 13, it was fine so long as an ST had 17 jump. So I would have thought previously that these two factors together were obscuring the legitimate causes for you.
Now a test of a real player is showing that a team can do with just 13 jump max. But here's another thing - Kounde has 13 jump, but also only has 15 acc/14 pace/15 stamina. If you look at Kounde, it's most likely what's making up for his deficient jump is one or more of the following: anticipation, determination, work rate, agility, balance, stamina.
I guess an interesting next one to test will be 'Sean Goldberg', who has a fairly high rating but only 12 acc/pace/jump.
A4 said: I only heard that traits don’t really matter much in FM, but I’ve always noticed how good players with “tries tricks” are. I even started teaching it to everyone in my youth academy. Of course, it’s anecdotal evidence, but when youngsters learn it—or even just try to do it—their impact and match ratings seem to be so much higher.
I’ve also noticed the same thing with the Light-Hearted personality. Somehow, good players with that personality seem to perform so much better for me than players with other personalities.
Is this something anyone has ever tested before? Expand I tested traits briefly before and found that they had zero effect, though I've been wanting to give it a retest soon to be sure.
ZaZ said: You can block sacking with the in game editor. Expand I don't have the in-game editor either, because that also costs money.
I know testers seem to typically use it a lot, and I would buy it if I needed it, but I've never felt the need to freeze things and whatnot.
johnconnerson said: After your recent findings, do you think these individual training focuses are still the best ones to use? Expand The optimal training will indeed have to be reassessed at some point, but it shouldn't change that much. There's a chance it wouldn't change at all. The quickness focus certainly wouldn't change.
BaZuKa said: I am doing a Moneyball save with Brentford. I added your CB meta attributes to my scouting app and got these results. Are these fine, or do I need to tweak anything else? Expand
The templates will do well, but the latest weights based on the real player testing will change things a bit. For instance, jump reach 17 is no longer necessary. It's unlikely this is because 17 jump reach wasn't necessary for the template, but that its now being compensated in some other way (i.e. perhaps higher anticipation reduces the need for jump reach, which is just an example I'm making up to illustrate what I mean).
To clarify things, and I think people more broadly will benefit from being aware of this, it seems to have turned out that indeed only the attributes that were selected in the template matter. It's just that the exact distribution for those attributes will change somewhat.
So if I was highlighting it in the same way roles are highlighted, this is how to assess Maeda (or any player, in any position, except GK):
I've started testing DC. I've started with Jules Kounde, as he has the highest rating with 13 jumping reach and the first thing I want to know is high jump reach on DCs still a valid requirement. Then I did one of the highest rated natural proficiency DCs who did have pretty high jumping reach of 16, Araujo.
82% | Kim Min Jae - 3rd, 3rd, 6th, 6th, 3rd, 4th, 5th = 4.286 position 83% | Ronald Araujo - 4th, 8th, 3rd, 1st, 7th = 4.6 position 79% | Jules Kounde - 5th, 4th, 7th 5th, 3rd = 4.8 position
So it's looking like high jumping reach isn't strictly necessary, and that the same set of weights works for every position, but still need to do a few more players to adequately demonstrate the latter.
Does one position contribute more than another? Compared to the STs, it seems as though DC is roughly ~1 position off, and the same was true of DR. And it's even clearer that GKs contribute less than STs, to getting higher position at least. The most likely explanation to my mind is that defenders do contribute less to winning than STs, because their focus is on not losing (defending). It'll be interesting to see if wingers reflect this once I test them.
Purity said: Why don't you make yourself unsackable with fmrte to avoid this ? Expand I don't have paid FMRTE. I don't know if the trial version can do that.
I could redo my database edit to use Luton instead of Man City, but it's more work than it's worth. I'm not particularly concerned about the sackings and lower GK results at the moment. If I did rule out sackings as the cause, I would be left with the problem of elevating GKs with middling stats, which I can't think of an obvious solution to at the moment. Which means I'd have to spend a lot of time coming up with sophisticated formulas. So I don't want to go down that path right now.
I have limited time and energy, and of what remains to be done, I think it's more important to get jump on DC assessed next, and then verifying the hypothesis that all the positions share the same attribute requirements - and if not, finding those differences.
bf3metro said: FMST more accurate or just simpler for you? Expand FMSS someone has said you can add code to make it do the same calculations, so FMST is not less accurate, but FMST allows me to add some complexity more easily right now. I can some utilize some complex operations in FMST without research - with FMSS, I'd have to investigate how to program it.
90% | Alisson - 6th, 3rd, 4th, 4th, 3rd = 4 position 85% | Thibaut Courtois - 3rd, 7th, 2nd, 5th, 8th = 5 position 83% | Jan Oblak - 5th, 8th, 6th, 4th, 3rd = 5.2 position 83% | Walter Benitez - 4th, 9th, 4th, 6th, 5th = 5.6 position 83% | Ugurcan Cakir - 7th, 9th, 2nd = 6 position 80% | Ederson - 6th, 6th, 4th, 8th, 7th (sacked), 7th (sacked) = 6.333 position 82% | Marc-Andre ter Stegen - 10th, 11th, 4th, 4th, 3rd = 6.4 position 78% | Anatolii Trubin - 15th (sacked), 6th, 4th (sacked), 5th = 7.5 position 70% | Davy Roef - 13th (sacked), 4th, 14th (sacked), 6th (sacked), 1st = 7.6 position 61% | Jack Stevens - 13th (sacked), 4th, 9th (sacked), 8th (sacked), 8th (sacked) = 8.4 position 76% | Sinan Bolat - 4th, 2nd, 19th (sacked), 5th, 14th (sacked), 9th (sacked), 12th (sacked), 5th = 8.75 position 68% | Ben Winterbottom - 10th (sacked), 5th (sacked), 12th (sacked), 18th (sacked) = 11.25 position
Adjustments were made to accommodate Ederson, who has gone from 76% > 80%. Ederson is an interesting case, because he's at a top team (Man City), yet predicted to be relatively low, and is quite low in what has been considered the key 3 attributes (aer, ref, agil).
I wanted to know if you’ve had the chance to form a strong opinion on FMST vs FMSS. Which one seems more accurate to you, and generally speaking, which one do you think is the better tool to use? Expand I don't have an opinion of which is better, but I'm using FMST right now because I can put in more sophisticated weighting formulas easier
I inputted my player template that finished ~5th in the Premier League in FM24, into FM17, and got 14th with the 'GOODBYE 442' tactic. But that meant the good results were within reach, so I increased pace & acc from 15 to 16, and got 8th:
So it seems that most attributes not mattering is true in FM17 as it is in FM24. Perhaps not exactly the same, but not far off in any case.
I suspect the difference between 14th/8th for FM17 and 5th for FM24 is actually the quality of the tactic. And perhaps it is the case that the introduction of gegenpress, and/or just the incompetence of the new match engine team (Nic Madden who joined SI for FM16 replaced Paul Collyer as head of match engine/match ai from FM18 onwards according to ChatGPT?), permitted for more overpowered tactics.
So based on this, it seems like player attributes in FM have always been misleading BS, but perhaps at least in some earlier versions, it was adequately obscured by the constraints of the tactic, which is overall more consequential than attributes themselves (though it won't save a dud team, and the ratio seems something like 70/30), and were either more balanced before or it simply wasn't yet exploited by people like Knap fully. Personally I reckon it's both things together.
This may actually be quite useful to know, because if the underlying attribute mechanics have always been the same, and they only have some degree of control over the tactic impact, it seems to me to mean that FM26, FM27, FM28, etc. are either going to have strong exploit tactics, in which case the attribute meta works the same and we can predict that just by looking at the exploit tactic degree of success, or tactics have been flattened down, in which case the attribute meta still applies but now you need more of those lesser attributes in the meta (i.e. anticipation, composure, dribbling, etc.) in order to actually win the season. And they can't go too far down the 2nd route, because then players would complain about not being able to win anything. That's my theory anyway.
I’ve also noticed the same thing with the Light-Hearted personality. Somehow, good players with that personality seem to perform so much better for me than players with other personalities.
Is this something anyone has ever tested before?
So I've tested now Osimhen having both 'shoots with power' and 'plays one-twos' as ST, as these are the two preferred moves that supposedly had the best benefit in Stybb's testing.
Victor Osimhen default - 3rd, 3rd, 7th, 3rd, 3rd, 4th, 3rd, 4th = 3.833 position
Osimhen w/ shoots with power, plays one-twos - 5th, 6th, 6th, 3rd, 5th = 5 position
I don't think there's enough samples to conclude the preferred moves made him worse. I think it's unlikely, but it is possible since it could be overriding his role in the Knap tactic.
What we can be sure of is that it definitely hasn't made him better.
I picked Osimhen because I already have test data on him in a fairly controlled team, he's one of more consistent players of the players I have data on, he's a good ST but not overly good (we want some headroom for any improvement to shine through), and he's a well rounded player that excels in pace & acc & finishing.
Busy in life for a few days right now, but I'll get round to it probably after that
OpticFawn said: what values would you put for Injury Proneness, natural fitness, dirtiness and aggression as simple values on other platforms?
You could input the values I give for Genie Scout
The relative importance of those attributes are:
Dirtiness = Very important
Injury proneness = Moderate importance
Natural fitness & Aggression = Minor importance
It's just something I added in to toy around with later.
To compare the relative impacts:
left20/right6 vs right20/left6 = 20 difference
1 loyalty vs 20 loyalty = 19 difference
That's ~0.2 pace.
Thinking about it now, I should probably remove the left foot preference for the next version. 0.2 pace is small, but not exactly nothing.
I tried thinking of a valid reason to keep it:
Even if left vs. right makes no performance difference, when it does in real life, there are fewer left-footed players in the database, and we need some left-footed players for set pieces so we should skew slightly towards them.. but then again, set piece attributes are proven to be useless.
I guess the only real testing I can think of when it comes to preferred foot performance is HarvestGreen's test, and while his results show that foot side matters depending on position, I don't think we have any real information on left vs right in a central position such as ST or DC.
Maybe I'm over-interpreting his vaguery, but nonetheless I'm going to clarify it if he won't.
For the position data, ChatGPT gives me the following 95% confidence intervals:
Things get hairy for the handful below Dybala.
It surmizes it all overall as:
Minimum ±0.38
1st quartile (Q1) ±0.82
Median ±1.38
Mean ±1.52
3rd quartile (Q3) ±1.69
90th percentile (approx.) ±2.50
Now the color coded system I used for the % figures, which represents the best fit of the attribute weightings to the position data, is just to show which players evidently require further attribute weight adjustments in order to be reasonably consistent with the position data. There will likely always be outliers, so I also wanted to show just large this group is - it's currently ~12-15%, which I think isn't too bad.
As for the confidence interval of the % results, it took a while in ChatGPT and I'm not entirely sure I've done it correctly, but it should be this:
95% prediction interval is ±9–12 percentage points
80% prediction interval is ±6.3-7.7 percentage points
This should take into account both the position data uncertainty and the best fit uncertainty of the % figures.
If the % was fitted to the average position data perfectly, it would have a 95% confidence interval of ±3.6 percentage points. And that itself could be improved with more samples taken.
Statistical accuracy of my player rating system
I compared my 60–90% player ratings against the objective average positions from 32 players.
Results:
Pearson correlation: −0.80 — strong relationship between rating and objective position
R²: 63.8% — the rating explains approximately 64% of the variation in position
Unexplained variation: 36.2%
Regression slope: −0.241 positions per percentage point — higher ratings predict better positions
MAE: 0.73 positions — the rating misses the objective position by ~0.73 places on average
RMSE: ~1.05 positions — larger errors are penalised more heavily
Normalised MAE: ~15.5% — average error relative to the mean objective position
In short: the rating system has a strong relationship with the objective rankings, with an average error of around 0.73 positions.
For your forum, I'd describe this as:
Using a ±5 percentage-point tolerance, 5 of 32 players (15.6%) fall outside the expected range, meaning 84.4% of the ratings fall within ±5% of the regression-derived expected value.
Thanks ChatGPT!
2017 called, they want their 'Do you even have a PhD in Statistics?' line back.
For attributes, too hard to say, but I think it's more likely things have stayed the same than that they have changed. I suspect it would largely hold true going back to ~FM18 at least, but that's just my guess.
For everything else, it should be true going back a long way.
FMST26 weights (copy paste this into statistics > custom metrics (advanced)):
Outfield:
(acceleration * 90.5 + pace * 94 + jumping_reach * 43 + pressure * 17 + dribbling * 14.3 + work_rate * 66.5 + agility * 16.0 + anticipation * 40 + composure * 26 + stamina * 50 + consistency * 14.05 + determination * 43.5 + balance * 14.25 + strength * 5.0 + concentration * 45 + finishing * 2.0 + important_matches * 10.1 + natural_fitness * 6.45 + professionalism * 15.0 + ambition * 2.0 + loyalty * 1.0 + aggression * 10 + vision * 0.5 + left_foot * 6 + right_foot * 5 - (injury_proneness * (48 / (natural_fitness * 0.75))) - (dirtiness * 32 * (aggression * 0.1))) / 113.2
GK:
((aerial_reach * 28.68 + command_of_area * 15.69 + first_touch * 16.99 + passing * 10.46 + reflexes * 86 + concentration * 82.35 + determination * 88.89 + work_rate * 38.68 + acceleration * 15.03 + balance * 10.46 + agility * 32.68 + jumping_reach * 14.30 + natural_fitness * 10.46 + pace * 28.30 + stamina * 32.68 + strength * 10.46 + technique * 15.69 + pressure * 15.03 + professionalism * 9.80 + flair * 49.02 - injury_proneness * 49.02 + vision * 25 - dirtiness * 16.34 - ca * 9.80)) / 84.5
Genie Scout file: https://files.catbox.moe/9n18q0.grf
I translated it to GS as best I could.
Outfield:
90% | Erling Haaland - 1st, 1st, 4th, 1st, 2nd, 1st = 1.666 position
88% | Kylian Mbappe - 2nd, 3rd, 3rd, 3rd, 5th, 2nd, 1st, 1st = 2.5 position
87% | Mohamed Salah* - 1st, 3rd, 5th, 3rd, 3rd, 2nd, 1st = 2.571 position
88% | Kylian Mbappe (AML) - 1st, 2nd, 3rd, 5th = 2.75 position
82% | Bukayo Saka (AMR) - 2nd, 3rd, 3rd, 3rd = 2.75 position
82% | Heung-Min Son - 3rd, 2nd, 2nd, 2nd, 2nd, 6th, 4th = 3 position
83% | Daizen Maeda - 2nd, 3rd, 4th, 4th, 2nd, 3rd, 3rd = 3 position
80% | Kyogo Furuhashi - 4th, 2nd, 3rd, 5th, 3rd, 2nd, 2nd = 3 position
82% | Vinicius Junior - 5th, 4th, 1st, 3rd = 3.25 position
83% | Eduardo Camavinga (DM) = 3rd, 5th, 2nd, 3rd, 4th = 3.4 position
82% | Harry Kane - 5th, 2nd, 3rd, 3rd, 1st, 7th = 3.5 position
83% | Robert Lewandowski - 2nd, 2nd, 2nd, 3rd, 5th, 2nd, 8th, 5th, 3rd = 3.555 position
77% | Dusan Vlahovic - 1st, 4th, 4th, 5th, 1st, 6th, 4th, 4th = 3.625
82% | Lautaro Martinez - 7th, 3rd, 1st, 5th, 2nd, 5th, 2nd, 5th = 3.75 position
80% | Victor Osimhen - 3rd, 3rd, 7th, 3rd, 3rd, 4th, 3rd, 4th = 3.833 position
78% | Lionel Messi - 4th, 2nd, 1st, 5th, 9th, 3rd = 4 position
79% | Viktor Gyokeres - 6th, 4th, 3rd, 4th, 5th, 5th, 2nd, 5th, 2nd = 4 position
81% | Giovanni Di Lorenzo (DR) - 3rd, 3rd, 2nd, 9th, 2nd, 5th, 5th = 4.143 position
82% | Kim Min Jae (DC) - 3rd, 3rd, 6th, 6th, 3rd, 4th, 5th = 4.286 position
82% | Ronald Araujo (DR) - 4th, 8th, 3rd, 1st, 7th = 4.6 position
73% | Donyell Malen - 2nd, 2nd, 7th, 5th, 5th, 7th = 4.666 position
78% | Jules Kounde (DR) - 5th, 4th, 7th 5th, 3rd = 4.8 position
75% | Kurt Zouma (DC) - 7th, 5th, 6th, 3rd = 5.6 position
72% | Robert Glatzel - 5th, 7th, 6th, 3rd, 12th, 6th, 2nd, 6th, 8th = 6.111 position
81% | Raheem Sterling (AMR) - 6th, 7th, 7th, 7th, 6th, 6th, 6th, 6th = 6.375 position
69% | Paulo Dybala - 5th, 5th, 6th, 10th, 3rd, 12th, 6th, 5th, 6th, 6th = 6.4 position
72% | Andrian Kraev (DM) = 9th, 8th, 5th, 5th = 6.75 position
62% | Adam Le Fondre - 8th, 11th (sacked), 5th, 7th (sacked), 6th (sacked), 8th (sacked) = 7.5 position
66% | Kingsley Schindler (DR) - 4th, 11th, 6th, 11th = 8 position
63% | Luis Suarez - 7th, 9th, 10th, 8th, 8th = 8.4 position
71% | Sean Goldberg (DC) - 5th, 16th (sacked), 8th (sacked), 5th = 8.5 position
77% | Sacha Boey (DR) - 7th (sacked), 12th (sacked), 12th (sacked) = 10.333 position
Error rate = 4/32 = 12.5%
Vlahovic = 6.5% off
Malen = 6.8% off
Sterling = 12.3% off
Boey = 22.1% off
GK:
90% | Alisson - 6th, 3rd, 4th, 4th, 3rd = 4 position
84% | Thibaut Courtois - 3rd, 7th, 2nd, 5th, 8th = 5 position
82% | Jan Oblak - 5th, 8th, 6th, 4th, 3rd = 5.2 position
82% | Walter Benitez - 4th, 9th, 4th, 6th, 5th = 5.6 position
82% | Ugurcan Cakir - 7th, 9th, 2nd = 6 position
79% | Ederson - 6th, 6th, 4th, 8th, 7th, 7th = 6.333 position
81% | Marc-Andre ter Stegen - 10th, 11th, 4th, 4th, 3rd = 6.4 position
84% | Gregor Kobel - 4th, 12th, 2nd, 10th (sacked), 3rd, 10th (sacked), 5th = 6.571 position
78% | Anatolii Trubin - 15th (sacked), 6th, 4th (sacked), 5th = 7.5 position
70% | Davy Roef - 13th (sacked), 4th, 14th (sacked), 6th (sacked), 1st = 7.6 position
61% | Jack Stevens - 13th (sacked), 4th, 9th (sacked), 8th (sacked), 8th (sacked) = 8.4 position
75% | Sinan Bolat - 4th, 2nd, 19th (sacked), 5th, 14th (sacked), 9th (sacked), 12th (sacked), 5th = 8.75 position
67% | Ben Winterbottom - 10th (sacked), 5th (sacked), 12th (sacked), 18th (sacked) = 11.25 position
Harder to tell who is inaccurate and who is error for GK..
Error rate = 3/13 = 23.1%
Kobel = 7.1% off
Roef = 7.1%(?) off
Stevens = 18.0%(?) off
The "position proficiency caps at 18" claim (repeated with confidence early on) turned out to be wrong once tested — worth flagging since it may still be floating around in older posts/quotes.
The Genie Scout personality-weighting bug wasn't caught internally — it took LightningFlik's independent tool to demonstrate the Osimhen-over-Haaland inconsistency. Worth being more explicit that GS output shouldn't be trusted at face value for hidden attributes even now.
The "Blended" file and the "15 pace/acc realistic template" contradict each other on your own numbers (the Newcastle-vs-synthetic-squad comparison) — this was acknowledged but never really resolved, just reframed as "the file is a rough guesstimate." That's a fair characterization, but it should probably be stated up front rather than discovered by a reader cross-checking the math.
The "attribute thresholds are relative to competition level" theory, used to explain away discrepancies with ykykyk05251/Zippo/Orion/HarvestGreen22, has never actually been tested — and skawkclsrn's much larger randomized dataset found no thresholds anywhere, which cuts against it rather than for it.
A lot of the decimal-precision weights (Pressure=52, Determination=56, etc.) come from single small-sample runs, some explicitly labeled "pure speculation" or "I don't remember why I set this" in your own posts. That's fine as a starting point, but the precision of the numbers implies a confidence level the underlying data doesn't really support.
None of this undercuts the value of the overall synthesis — it's clearly the most complete one on the forum. But the granular numbers probably deserve a "confidence level" caveat next to them, given how often they've had to be walked back or reframed.
Yes, these are good points, and most of them I still keep in mind, but it can be difficult to remember everything. I can see you've done a close reading, because I know some things here that I think everybody would have skipped over.
I will clarify a few things:
1) The hiddens bug with Genie Scout indeed remains unresolved, so I highly recommend to use FMST26 or FMSS. I choose to put out a GS file anyway because there are still useless features it has that the other two lack, plus its the established tool. I halved the personality weightings in my GS file to try and temper whatever inaccuracies arise, as personality turns out to be relatively unimportant for performance, but they remain included because I have to keep injury proneness and dirtiness in there anyway - there's no way around it. I'm no longer doing the ranking-to-weights match up work in GS, but I was initially, so this should bypass the bug issue to a degree, though now I do the weight work in FMST26 so it will probably drift more and more away from being accurate in GS. Long story short though, I think it's viable to continue using GS.
2) Pressure, determination, I think perhaps aggression, and no doubt a couple of others that are mainly 0 CA attributes were never properly refined in my template tests. So I tested determination at 14, but unlike many/most other attributes, I didn't get round to trying it at 13, 12, and so on. So in theory an attribute like aggression could be anywhere between '1' and '11' as a requirement. It's set at 11 because I believe it matters, unlike say passing, based on either other people's tests, or my own previous tests (say the 1CA tests). I figured at the time that since these are 0 CA cost attributes, or necessary for player growth anyway (i.e. determination), then it's not a priority to pin them down like I pinned down agility to 10. I ended up never getting around to completing it thoroughly.
3) I am pretty dead certain there are attribute thresholds relative to competition level. HarvestGreen22's data shows this clearly. It's what I have observed, even now with the real player rank data albeit its more convincing than definitive - Haaland comes no.1, but he has work rate of 13. Luis Suarez has godly technicals & mentals, but very low acc/pace/sta, and we see he is one of the worst performers. Those are the obvious examples, but I've seen many more subtle ones that are nonetheless identifiable. All that said, work rate does appear to be of importance even beyond ~11-13, and we see similar things for stamina and anticipation and whatnot, so while I'm sure there are attribute thresholds, the understanding of the curvature could do with refinement. Part of it could be that certain things kick in at certain levels. My observation with GKs is that at a mid level, reflexes matter less, but reflexes seem to take precedence at a high level (~15+) while aerial reach is essential at mid level but is significantly less essential at a high level (~15+).
4) I don't recall the Blended file vs template problem. And I'm not sure which came first, so I can't assume it's that the template invalidated the Blended file. All I can say is that when I did my best attempt at an objective comparative test of all the files, the Blended one did poorly. From memory, the Blended file was mainly based on HarvestGreen22's data + consideration of CA weights + consideration of player growth. I was also still using positional differences at that time, which now turns out to be bunk. As a point of history now, we can go back and ask 'Well, was the Blended file actually a step backwards from even the first file, and therefore do you regret putting it out there?'. At least that's the question I ask myself. And for the other files, I can say, 'they were better than the first file, but perhaps only marginally in hindsight'. Actually was striking to me was how although the improvement was marginal, the improvements of my files and those of others appear remarkably gradual and consistent. I would have expected things to be more over the place. And I could be wrong, but I think this latest method and the weightings derived from it are a big step forward.
5) I think it needs to be restated the current precision of the attribute weights does not reflect the actual real influence/value of the attribute. I'm merely moving numbers up and down to get a best fit of the data. It's like wedging a piece of wood inbetween two metal joints to make the leaning tower stable. It's not the right material, but if it works, it works, and we can replace it with the nuts and bolts later.
73% | Andrian Kraev (DM) = 9th, 8th, 5th, 5th = 6.75 position
Recall that 87% for ST is 2.571 and 73% for ST is 6.111, and that positions other than ST appear to be consistently ~1 position lower.
Every position for the Knap 424 formation has been sampled now, and I'm satisfied that the same set of required attributes apply to all outfield positions in the same way.
I've done some readjustments to the weightings.
I've dropped Sterling the equivalent of ~1 position, but this means he should we should expect no worse than ~4 position, yet his average is 5.375 if we give +1 for not being ST. This is the best I've been able to do.
Non-ST positions seem more comparable to ST now (less unexplained position difference), and stamina is now a more reasonable 40 instead of 94 weight - stamina seemed to be the partial cause of Sterling's overrating. Quite a few players have pleasant coincidental changes, such as Maeda going higher and Gyokeres lower, which is more line with the real rankings.
However some changes had to go with it, such as Vlahovic dropping a little below Messi. I figure it's more likely that Sterling should drop 1 position, than Vlahovic should not drop 0.5 position. And given most players seem more to be more accurate now, I'm not too concerned about Vlahovic.
Next step is translating to Genie Scout weights, then I can upload the updated file. Main reason for updated file will be adding the new GK weights of course.
The method I use is that I compare two players in conjunction with a few others, to rule out most of the attributes.
So for instance, let's say Mbappe is better because he has 20 pace/acc, then why does Haaland actually perform better? Then we find someone with same pace/acc as Haaland, but performs significantly worse. This player has -5 ant, -3 det, -3 work, -7 tack. We can rule out tackling off the bat, but how do we know how the remaining 3 attributes contribute? We compare two other completely different players to each other and one has +4 ant -2 work and the other has -4 ant +2 work, and find the +4 ant -2 work one performs much better. So now we know it's likely ant is the key missing factor, or at least contributes that precise position difference. We then verify that this would line up the rankings more correctly instead of no change or worse.
So that process gets us to the final weights, which we know are close to valid because almost all the players line up well without contradictions, and according to the weights Sterling should be closer to Mbappe.
If you compare Sterling to Mbappe, it's no problem, but what if you compare Sterling (6.6) to Saka (2.75):
There are a few differences, but nothing major. We know all the technicals are useless, aside from dribbling. That leaves a handful of minor to moderate differences in mentals, balance, and pressure (hidden), as the possible causes.
And the thing is, he's pretty well rounded in every attribute, so there's not obvious non-linear threshold being crossed for a certain attribute. I thought perhaps composure, but my template says 11 is the minimum, and he has 12. Thinking perhaps pressure right now.
Coincidentally I just made a comment on this video before coming here and seeing this post.
I was actually going to bring him up in my response concerning traits to you, and I'm guessing you asked it perhaps because you saw his video on traits too.
The jist of my post was that either this guy has something seriously wrong with him, or I'm missing something. Hence, another reason to give traits another test.
I first came across him ages ago, about 3 years ago, in which he claimed to do a proper test that showed scout JPA has a strong effect on newgens, which I knew from my own testing to be absolutely false. I left a comment about this, but he didn't respond.
Fast forward to now, and these past few months he's supposedly running and pumping out all these in-depth controlled tests with a discord community to back him, but if you go to it, it's a ghost town. Strange. And the tests themselves, obviously fly in the face of the results derived by others doing similar testing.
But I realized today it could be that it's because he's assessing by goals conceded and goals scored, which is not an adequate measure. I keep forgetting myself that final season position is the only reliable measure, at least that I know of. Still.. the scout JPA result can't be explained by this, so I simply don't trust the guy on anything. Maybe when I do the trait test, I'll also check to see if it affects goals difference as he claims it does.
I decided to leave all this out of my post before in the end, because obviously it sounds obsessive and veering off from what you asked, but its what came to mind. Occasionally youtubers can be a source of fresh good info.
So it's not the footedness
83% | Bukayo Saka (AMR) - 2nd, 3rd, 3rd, 3rd = 2.75 position
So it does seem something specific to Sterling
84% | Raheem Sterling (AMR) - 6th, 7th, 7th, 7th, 6th = 6.6 position
75% | Kurt Zouma (DC) - 7th, 5th, 6th, 3rd = 5.6 position
73% | Sean Goldberg (DC) - 5th, 16th (sacked), 8th (sacked), 5th = 8.5 position
Notes:
We see with Mbappe that the weightings give the same result, whether Mbappe is playing ST or AML (note: in each test he has been edited to have 20 proficiency in only 1 position).
Kurt Zouma is a player keithb's weightings has at rank 7th in world, even though he's playing at West Ham. So I thought it would be a good candidate to test, especially since Zouma has 20 strength, 18 jump, 17 heading, 15 tackling, 16 determination, and so on - everything that lines up with what has conventionally been thought to constitute a great central defender. The result says otherwise. He is equivalent to Malen or Glatzel at ST (4.666-6.111 position; 73-74%).
Goldberg I elected to test because he has 12 acc/pace/jump yet has a decent rating. We see here it is viable, though the position seems to have slipped ~2 places. Though this could also be just because the early sacking 16th position result.
Raheem Sterling's result is surprisingly low. We see from Mbappe that it is not a disparity caused by the position itself. And keep in mind that he is conventionally found not far from the top of GS ratings (keithb has him 11th for AMR). When assessing this kind of situation, I've come to find that it's about what specific attribute the player lacks, rather than what attribute or attributes are over-weighted. Comparing with Mbappe narrows it down to comp or det it would seem to me, but I wonder if it could also be that Sterling is playing on the right side with a right foot. I recall HarvestGreen found some stuff about this. So I will probably test to see if side matters.
First I was using DCs with as high Jumping (17+) as possible, but only decent Acc/Pace (13-14) and results were mixed.
Then I switched to DCs with high Acc/Pace (16+), but completely ignored Jumping (sometimes it was as low as 5).
The second type of defenders performed much better, their speed seemingly made up for their complete lack of aerial presence.
That could actually be because pace/acc is as crucial, or not far off it, as it is for STs.
When I reduced DC jump to 13, it was fine so long as an ST had 17 jump. So I would have thought previously that these two factors together were obscuring the legitimate causes for you.
Now a test of a real player is showing that a team can do with just 13 jump max. But here's another thing - Kounde has 13 jump, but also only has 15 acc/14 pace/15 stamina. If you look at Kounde, it's most likely what's making up for his deficient jump is one or more of the following: anticipation, determination, work rate, agility, balance, stamina.
I guess an interesting next one to test will be 'Sean Goldberg', who has a fairly high rating but only 12 acc/pace/jump.
A4 said: I only heard that traits don’t really matter much in FM, but I’ve always noticed how good players with “tries tricks” are. I even started teaching it to everyone in my youth academy. Of course, it’s anecdotal evidence, but when youngsters learn it—or even just try to do it—their impact and match ratings seem to be so much higher.
I’ve also noticed the same thing with the Light-Hearted personality. Somehow, good players with that personality seem to perform so much better for me than players with other personalities.
Is this something anyone has ever tested before?
I tested traits briefly before and found that they had zero effect, though I've been wanting to give it a retest soon to be sure.
I don't have the in-game editor either, because that also costs money.
I know testers seem to typically use it a lot, and I would buy it if I needed it, but I've never felt the need to freeze things and whatnot.
johnconnerson said: After your recent findings, do you think these individual training focuses are still the best ones to use?
The optimal training will indeed have to be reassessed at some point, but it shouldn't change that much. There's a chance it wouldn't change at all. The quickness focus certainly wouldn't change.
BaZuKa said: I am doing a Moneyball save with Brentford. I added your CB meta attributes to my scouting app and got these results. Are these fine, or do I need to tweak anything else?
The templates will do well, but the latest weights based on the real player testing will change things a bit. For instance, jump reach 17 is no longer necessary. It's unlikely this is because 17 jump reach wasn't necessary for the template, but that its now being compensated in some other way (i.e. perhaps higher anticipation reduces the need for jump reach, which is just an example I'm making up to illustrate what I mean).
To clarify things, and I think people more broadly will benefit from being aware of this, it seems to have turned out that indeed only the attributes that were selected in the template matter. It's just that the exact distribution for those attributes will change somewhat.
So if I was highlighting it in the same way roles are highlighted, this is how to assess Maeda (or any player, in any position, except GK):
82% | Kim Min Jae - 3rd, 3rd, 6th, 6th, 3rd, 4th, 5th = 4.286 position
83% | Ronald Araujo - 4th, 8th, 3rd, 1st, 7th = 4.6 position
79% | Jules Kounde - 5th, 4th, 7th 5th, 3rd = 4.8 position
So it's looking like high jumping reach isn't strictly necessary, and that the same set of weights works for every position, but still need to do a few more players to adequately demonstrate the latter.
Does one position contribute more than another? Compared to the STs, it seems as though DC is roughly ~1 position off, and the same was true of DR. And it's even clearer that GKs contribute less than STs, to getting higher position at least. The most likely explanation to my mind is that defenders do contribute less to winning than STs, because their focus is on not losing (defending). It'll be interesting to see if wingers reflect this once I test them.
I don't have paid FMRTE. I don't know if the trial version can do that.
I could redo my database edit to use Luton instead of Man City, but it's more work than it's worth. I'm not particularly concerned about the sackings and lower GK results at the moment. If I did rule out sackings as the cause, I would be left with the problem of elevating GKs with middling stats, which I can't think of an obvious solution to at the moment. Which means I'd have to spend a lot of time coming up with sophisticated formulas. So I don't want to go down that path right now.
I have limited time and energy, and of what remains to be done, I think it's more important to get jump on DC assessed next, and then verifying the hypothesis that all the positions share the same attribute requirements - and if not, finding those differences.
bf3metro said: FMST more accurate or just simpler for you?
FMSS someone has said you can add code to make it do the same calculations, so FMST is not less accurate, but FMST allows me to add some complexity more easily right now. I can some utilize some complex operations in FMST without research - with FMSS, I'd have to investigate how to program it.
85% | Thibaut Courtois - 3rd, 7th, 2nd, 5th, 8th = 5 position
83% | Jan Oblak - 5th, 8th, 6th, 4th, 3rd = 5.2 position
83% | Walter Benitez - 4th, 9th, 4th, 6th, 5th = 5.6 position
83% | Ugurcan Cakir - 7th, 9th, 2nd = 6 position
80% | Ederson - 6th, 6th, 4th, 8th, 7th (sacked), 7th (sacked) = 6.333 position
82% | Marc-Andre ter Stegen - 10th, 11th, 4th, 4th, 3rd = 6.4 position
78% | Anatolii Trubin - 15th (sacked), 6th, 4th (sacked), 5th = 7.5 position
70% | Davy Roef - 13th (sacked), 4th, 14th (sacked), 6th (sacked), 1st = 7.6 position
61% | Jack Stevens - 13th (sacked), 4th, 9th (sacked), 8th (sacked), 8th (sacked) = 8.4 position
76% | Sinan Bolat - 4th, 2nd, 19th (sacked), 5th, 14th (sacked), 9th (sacked), 12th (sacked), 5th = 8.75 position
68% | Ben Winterbottom - 10th (sacked), 5th (sacked), 12th (sacked), 18th (sacked) = 11.25 position
Adjustments were made to accommodate Ederson, who has gone from 76% > 80%. Ederson is an interesting case, because he's at a top team (Man City), yet predicted to be relatively low, and is quite low in what has been considered the key 3 attributes (aer, ref, agil).
I wanted to know if you’ve had the chance to form a strong opinion on FMST vs FMSS. Which one seems more accurate to you, and generally speaking, which one do you think is the better tool to use?
I don't have an opinion of which is better, but I'm using FMST right now because I can put in more sophisticated weighting formulas easier