Rating Calculation Questions

Yeah this is the rating system working as intended. It won't be as accurate for your friend because you always play together and a team of coordinated 900s will probably beat a team of uncoordinated 1100s (idk for sure, just a guess) which will cause some rating inflation, but eventually it'll settle down for him. There's not a snowball's chance in hell your friend hits 2k unless he can actually play at least close to a 2k level if you guys are mostly playing 4v4 tmm. You already see his rating starting to level off in the graph.

You even said it yourself, he has an over 50% w/l ratio for the entire time his account has existed, why wouldn't he get more points for win than losing? Eventually you guys are going to drop below a 50% w/l and then things will change. You're not going to magically have a 1500 rating average on your team because of this one dude and beat a team of actual 1500s.

Edit: Plus this isn't even what this thread is about, though it's an easy enough mistake to make here I think

@exselsior For every game we loose he looses 1 more point than us, but for every win he gains 4/5 more points than us. And this is after over 100 games. That way he is slowsly but surely overtaking our rating and continue to climb. Our ratings will be underrated and he is already overrated by about 350 points (he's probably around 900 rating skillwise).

What should be done is that after a game the guy with the highest rating gains the least amount of points, and the guy with the lowest gains the most points, that's how it works in 90% of other games using ELO system in team games.
This way while playing together we would settle around equal rating among us.
Currently it seems that the amount of rating gained/lost is tied to how many games you have on your account.

Current system causes unfair matchmaking if we decided to play alone. My rating is underrated, so I'd be smurfing a bit, and our friend will get paired with 1300 rated players while he is 900 in reality, and that affects both ranked 4v4 and custom games, bc his global rating is even more inflated.

For us 3 as long as we play together it doesn't matter, but when we play alone is where the problem with current rating system is apparent.

Then have him play with and against other people. The whole problem is you keep talking about how this dude meets some objective skill level while saying he only plays in a specific subset environment. Of course the rating will be inaccurate, any rating system will make his rating inaccurate. The system is rating your team as the sum of its ratings and gauging a probability of victory based on the enemy team sum rating and then divulging points based on individual player uncertainty. If you guys "only" play with each other then the system can and will never be able to tell who of you is actually the critical piece.

@deszcz_pyr Still not a problem with the rating system. It's impossible to account for the fact that you only play together in any robust rating system. If I was playing LoL for instance and was somehow playing on a team of all high diamond players and was a silver player myself, I doubt they'd have much issue carrying me to plat or whatever. Same goes with literally any other game ever that has team game rating systems. FAF doesn't even use ELO, it uses trueskill https://wiki.faforever.com/en/Rating-System

If you want him to quickly get to his true rating and stop gaining/losing more points per game than you guys then get him to play custom ranked 1v1s. What you're describing is not in any way an issue with the rating system, and it's not actually a good idea in practice for the lowest rated person to get more points than the highest rated person, and afaik no major team game has a rating system that does that.

There's no possible way to truly know who contributed the most to a win to attribute points in such a manner. It can only be based on if it's a win or a loss and what your current rating and deviation are against the likelihood of the outcome.

@ftxcommando said in League System Feedback Thread:

Then have him play with and against other people. The whole problem is you keep talking about how this dude meets some objective skill level while saying he only plays in a specific subset environment. Of course the rating will be inaccurate, any rating system will make his rating inaccurate. The system is rating your team as the sum of its ratings and gauging a probability of victory based on the enemy team sum rating and then divulging points based on individual player uncertainty. If you guys "only" play with each other then the system can and will never be able to tell who of you is actually the critical piece.

While it is true that our ratings will never be accurate, there is no reason for our teammate to gain more points than us even 100 games in. Same with me and my other teammate. This drifting away between our ratings is ridiculous, if we were gaining same points, or even better our ratings would bu pulled closer together, that'd be much better.

I really believe we will drift away indefinetly and after 5k games we would get absurd results.

If he gets 2k and you guys are ~1300 then your rating average would be almost 1500. If he's actually a 900, do you really think you're going to win vs another team who's average rating is higher than your best players actual skill level? How do you think he's going to keep climbing indefinitely, if me or someone better than me is vs your team due to inflated ratings we will crush. You're not going to keep winning to the point where his rating indefinitely increases, it just can't happen. It will get to the point where your team's w/l will drop and his rating increase will stop. It's already getting to that point as the graph shows. If not, then maybe he's actually better than 900 and is carrying you guys and deserves it.

Once your combo of 3 dudes are losing close to 50% of your games against some sum equivalent, your uncertainty levels will all decrease. If you're mad that a dude is overrated, stop playing with him and prove it to the system. That's it.

@exselsior He will climb because our ratings will decline as his is rising. And this is happening right now. We reach our true team rating, we drop some, and then we win again and he keeps running away with his rating. If we loose 4 games in a row and win 2, he looses 4 points more than we do for the lossess, and gains 9 more for the 2 wins, so even such bad streak he is pulling away. Eventually as his rating will be growing ours will decline, even if our true team performance stays the same. Matchmaking will treat 3x 1000 same as 2x 500 and 1x 2000.

@ftxcommando In our last 20 games we won 10 times and lost 10 times, we are around 50% for a long time already and changes nothing. Im not mad he is inflated, but don't come at me when he joins your game and underperforms hard compared to his rating number. It's not my fault, and it's not his fault.

That simply isn't how this works because his uncertainty is dropping, although maybe too slowly in your eyes. Yes, he will always have an inflated rating if he only plays with you guys. The only way to fix it is to have him play without you guys. It wouldn't take that many global 1v1s to fix this issue as long as they're reasonable. Meaning don't have a 2k global player who doesn't want to be 2k play him and let him win or some bs like that. This is not an issue with the rating system. The rating system simply isn't built to handle this situation, nor should it be.

FAF has been around since 2014. There are plenty of people who only play relatively low level premade team games. There's not a single instance in the past 8 years of some random dude hitting 3k because of this or whatever.

This whole thing is about like a 200 variance in rating, that isn't even a standard deviation with FAF parameters. Your friend has like 70% odds to win a game against you in 1v1, if it isn't true then go play 10 1v1s with him and it'll fix itself.

Hey @Deribus I think @Freedom_'s post actually is on topic for the thread it was moved from

There have been situations of some dude farming up to like 2100 rating or so just from having like a 100% win rate against his two friends that are horrible. Good old TuxedoSam.

1500 is the expected median, if you have some artificial environment where it consists of like 4 dudes, even if they are like 200, 300, 600, and 1000 in the general FAF ecosystem if they purely play each other in custom games they could be grossly different.

Yes but that's playing against and not with, sounds like they're always on the same team

@ftxcommando well it's 4v4 matchmaker so 1v1-ing him wont fix it. And playing 10 4v4 games to troll other people and drop his rating, is a lot of time time invested for something not fun for anyone. Also if he lost 10 games he would drop 81 points given he looses 9 points a loss.

@exselsior getting 900 rated player to 1400 in global by casually playing 4v4 matchmaker for 1,5 months seems broken enough for me.

I mean he's probably better than 900 by now anyway, with barely 100 games played I'd hope he's still improving at least a small amount day over day especially since with what you're saying he's constantly playing vs people better than 900s. You're still missing the general point here though, and that's that the rating system isn't and shouldn't be designed to handle this scenario. Get him to play 4v4 tmm by himself if this is such an issue for you, and if you guys only ever plan on playing 4v4 tmm together and nothing else, who cares? Your team's overall rating will be accurate and that's what matters at the end of the day.

It's intentional that your matchmaker rating transfers to global. It will stop once it reaches 1400.

@exselsior he wasn't 900 to begin with, he was a "0", learning the game as we went. I estimate his skill is at 900 now with how he performs. So 900 is his peak not starting point. All the 100 games in the 4v4 matchmaker with us, is all his games ever on FAF.

@exselsior said in Rating Calculation Questions:

Hey @Deribus I think @Freedom_'s post actually is on topic for the thread it was moved from

Whoopsie, moved it back, ty