ATTN: 343 Ind - Help stop ranked quitting by providing more in-game information on how Trueskill 2 rank impacts players

I have been curious about how the halo ranking system calculated things for awhile, but I had no real insight as to how it worked as a player. In hindsight, this led me to make decisions in game that were not in the interest of ranking up.

During the beta, on more than one occasion, I would drop a connection in a ranked game. When I would return to look at my rank from the connection loss, I would not see an immediate chance in my rank. This led me to falsely believe that I could protect my rank by quitting a match if I ever ended up in a 3v4 game or worse. Or in general that ‘quitting =/ massive penalty, except for ban duration’.

Now maybe I am in the minority in thinking this way, but I do think that how things are calculated it pretty opaque. I recently was pointed to the Microsoft publication describing how Trueskill 2 works, and I was led to believe you use that or some derivative of it. After reading the publication, I now (think I) understand that quitting or connection loss is the worst possible outcome for your rank, even if it isn’t immediately reflected in your rank upon quitting.

I think the community would benefit from a more in-depth explanation in-game under the ‘ranked help’ toggle button. It can be simple, but it should outline a few fundamental ideas so that people know they should not quit a game that is going poorly.

E.g The help would say something like:
1} Quitting or losing connections to ranked games gives you the worst possible rank outcome possible.
2) The game heavily factors in team size when calculating the win chance E.g a team of 3 is not expected to win against a team of 4. Even if you still get a penalty for losing a 3v4 game, it is a dramatically better outcome than quitting the game is.
3) While Win vs Lose is a heavily weighted factor in the equation, your ‘raw score’ (kills vs deaths etc…) is also an important factor.
4) Playing in squads (fireteam party members) makes the system expect more from you to get the same rank up. E.g the system expects a squad to always outperform random people (all else being equal).

For those interested, here is a link to the Microsoft Trueskill publication: https://www.microsoft.com/en-us/research/uploads/prod/2018/03/trueskill2.pdf

Now I don’t know for sure all of my statements are 100% factual. I am basing it off my limited understanding of that Truskill 2 publication. But even if my statements need adjusting, this information should be readily accessible in-game. I think if people actually understood that quitting is the worst rank outcome you can have, and that the system is sympathetic to disadvantaged team sizes, then people would probably quit less.

I frequently find myself in matches where one person drops and then because of that, everyone quits. If they knew that was the worst thing they could do then they might reconsider it.

Another option you should potentially introduce is an option to ‘peacefully surrender’ by vote. This would only be available/apply to ranked games where there is a plyer number disadvantage. The disadvantaged team should be able to prompt a team vote to discontinue the game without penalty.

Thanks for listening.

2 Likes

I concur. They’ve been rather silent about their MMR system as far as how it works, and them offering an understanding of it would probably resolve a lot of frustrations.

Not to say that quitting isn’t a problem, but it seems to me most of the time its PC players games crashing. A lot of times when I see a player quit, I check their profile and they are almost always a PC player.

3 Likes

That does add up, especially considering how unstable the game is on PC as of late. If I recall correctly, memory leaks are one of the game’s major issues on computers.

1 Like

So tired of the quitting or crashing or whatever in ranked. Tired of losing CSR for 5 games worth of wins for losing a 3 vs 4 match when we get slammed in KD ratio because we lost map control due to missing a player. Its terrible and something needs to change. I shouldnt have to play 4 hours and get riddled with quitters, uneven teams, lag, etc., and rank up, just to lose it all again from these factors. Fix it, or your ranked players will leave at some point in the near future.

2 Likes

Nobody cares about their rank because quitting won’t likely cause them to drop back a full rank.

A 1-50 ranking system would fix this because you would actually see a drop let’s say from Rank 42 to 40 for quitting, wheras people don’t perceive the rank drop impact as much where there is more flexibility within the structure of Silver/Gold/Diamond etc

the game assumes you cant improve so it keeps you in a rank forever after a certain time. It gives impossible games to win like 4 high onyx vs your team of mid to high diamonds, and it will derank you if you do bad but its just keeping you were you are, and even if you do win some games it will derank you back to the spot you were in from the first game after hours of playing better than you used to. Best thing you can do is create a new account and start over as long as you really are playing better. I would see this back in halo 5 a pro player played with a high diamond player, they would win games but only the pro would rank up and the diamond never made it to onyx LOL also the devs can change up the ranking system somehow, in the last season of h5 many diamond players who were diamond previously started to stay in platinum except for a select few who always made to onyx and their winning percentage would be in the high 60s, meaning they had more easy winnable games and ranked up while others stayed in mid 40 to 50%.

1 Like

The system is so broken. Its infuriating. Im at my breaking point now where Im about to walk away until its fixed.

I don’t want to burst anyone’s bubble, but I’d be super surprised if we got much info about it. In all of H5’s life cycle I think 343 only have given us many details about H5 one time, and it was like 2 paragraphs in a monthly blog update

1 Like

There is a fair bit of info out there about ELO systems - including TrueSkill2.

But I agree. They need more on the spot info. A post game breakdown of what just happened and why.

TrueSkill2 actually does assume that the player is going to improve. It was one of the improvements over the original TrueSkill which assumed it was equally likely in either direction.

After a while your rank does tend to stabilise. It will resist going up (and down) unless you can definitely prove you are good enough. This involves beating the better teams you are put up against.

Different accounts will have different trajectories depending on the shape of their skill curve.

Remember that the system never actually “knows” you MMR - it just has a curve with a mean (your probable MMR) and a width (the confidence it has in knowing your skill).

Your new “pro” would have had a wider curve. The system is still volatile for them and would have been happier ranking them up faster because it had less confidence in their actual rank. And the pro would probably have triggered a couple of other weightings (eg. kill rate) that suggested to the system they needed to be higher.

The player who tended to slow down towards high diamond probably had a narrower curve. Their game history suggested that they were around this level and the system started to resist wild fluctuations. Just as you don’t want to lose rank with a couple of bad games - you can’t also rocket up on the back of a couple of good ones. The important part is that the system is using more than just the last couple of games to make these decisions. The player can still rank up - they just need to prove it.

A low population with a disproportionate skill distribution is always going to act weird. The players left would not have fitted a normal distribution curve - but the system would still have assumed they did - hence the weird behaviour.

Actually 343 was super involved. Josh Menke was active on the forums. There were lots of detailed posts and explanations. And he would even look up individual’s records and justify their rank.

Going by the paper they are fitting the statistical model used inside TrueSkill 2 based on prior matches. This means the system looks at a collection of matches and determines which metrics (K/D/A, time the objective was held, etc.) contributed the most to winning a match. In all likelihood they used beta flight data to do this modelling. During the flights people “just played the game” without consideration for “rank”. It seems that this data shows that K/D is basically the most important metric in that regard (which arguably makes sense - the player holding the Oddball for example is quite limited in what he can contribute to the defense - the others need to defend him).

However people found out which metrics are seen as favorable by the system and have adapted their playstyle to that. So basically people started going on killing sprees instead of defending the carrier. Due to that the assumptions the model contains are no longer valid in the real world (while a kill previously meant the carrier was save for a time - it now might only bring marginal advantage to achieving the objective).

This is a common issue in for example traffic management in cities. You optimize the system (traffic lights, etc.) for optimal flow overall. However this shifts the traffic around (people don’t like to wait at traffic lights, even if that is best for overall flow in the city) and you get unforeseen issues elsewhere.

The system should work better if they retrain their predictive model (perhaps they will do that for the mid-season reset). That is until people optimize their playstyle to the new system again. Especially for objective modes they might need to manually incentivize the system to reward playing the objective, as it might not be a good predictor of win/loss but it’s a necessary pre-condition for winning at all.

I might be wrong of course. What are your thoughts on this?

Game assumes you have an equal chance of improving or getting worse over time.

Give your account to a more experienced player and they will rank up extremely fast.

The more info people have, the more it will be abused and played in an undesirable fashion. Best we don’t know too much.

Lots of people will always complain they should be higher. This has been true of every ranking system in every game since the dawn of time.

Well, you’re right that people think we’re better than we are, but I think it should be a little more transparent and that people have a clear idea of ​​why they rank up or down.

Personally, I’ve been playing Halo and other shooters for many years and it’s not as important for me to rank up as it is for games to be balanced. If you are Gold, then that all the members of the teams are gold or very close. What you can’t do is match up with people with very different ranks or team up with 1 platinum, 1 silver, and 2 unranked. And that’s what’s happening right now in Halo.

I also don’t mind losing a game if the opposing team is better, but what can’t happen is lag, desync, server issues, etc. do not allow me to enjoy and compete on equal terms. And that’s the case with Halo’s ill-fated multiplayer. and the same thing happened in Halo 5.

Should help when players are split more appropriately, too many people in Plat/Diamond. Think they are reworking to ensure people are split more so some will be lower and some may be higher.

Hmm interesting. I remember there being a small period of time where there was a lot of discussion about it on the forums, but I didn’t remember getting that much info. I was on and off the waypoint forums during H5 though, so maybe the big discussions were while I just wasn’t spending much time on the forums