Maps and MMR question

Is it me or does your MMR change for specific maps.
e.g. if you win alot on a certain Map it’ll view you like someone of a higher rank on say Oasis and if you lose all your matches on Kings Row it adjusts your MMR downwards essentially SR being the average of all those map-specific MMRs (still entirely W/L based minus PBSR).

It’s just you. Competitive MMR is completely described by a value and an uncertainty for each role. See How Competitive Matchmaking and Rating Works (Season 18).

We still have little or no evidence that competitive ladder is goverend by an elo/trueskill derived ratings system.

There are infinitely (quite literally) many-to-few scoring functions that map N parameters down into just two (sr, mmr). There could be some hidden map bias, streak magic, etc, along with a plethora of other factors, and we wouldn’t really know.

A dev update/comment on the matter would be nice.

Every developer quote is consistent with a TrueSkill type system with tweaks.

Unfortunately, Scott has a bad habit of speaking in terms of SR, even though MMR is the real rating.

Here Jeff says that MMR is the real rating: Seagull & Jeff Kaplan: NEW Hero Confirmed! - YouTube

No. I mean theres stuff like performance-based sr where in-game metrics are used to adjust the ranking parameters. The OP is asking if maps are factored in for matchmaking (are we putting together a team of players that flop on 2cp vs. a team that does well, etc.) Beyond maps, there could be all kinds of stuff they track to bias the 2-3 inputs into sr/mmr calculations.

We just don’t know what their scoring functions are. Its not a simple matter of sr/mmr. They have said many times they track many things and that they ‘need data’ to know how you play, what your intent is, and classifying where you belong. Thats why people wonder if they try and figure out your ‘intended’ chars to play, whether you’re throwing or not, etc.

So we have like 1 post to go from. 1-2 posts from dev sources vs. the 1000s of annectotes from players experience. Something something dunning-kruger something something elotrueskill. You’ve been lied to or are naive if you think they implement an agnostic elo/trueskill when pbsr is right there next to them.

There is far more to their matchmaking than a win% chance with numbers (sr, mmr).

I dont think its map specific, its probably just stats over time in general. Doesnt matter on which map you had your stats.

Post from developers have much more value than post from random forum users. People on these forum like to lie a lot or not saying whole truth. + its just stories usualy, without proper evidence.

1 Like

Unknown at this point, but they did say that they calculated your PBSR on if you are defending, attacking, and what map your on.

they probably do, but we will never know.

Yet why wouldn’t they apply their TrueSkill variation per map?
It should, in theory, improve your accuracy.

The reason I was asking is this btw.:

This isn’t the case for most matches on console right now (that’s a fact), it could be the result of having groups in all of the games (not visible) or the result of limited player pool in two of three roles forcing yet in a lot of matches both teams ranks per role let alone per team don’t match.

Its based on the assumption that your skill varies greatly due to the map you are playing on. Map specific hero selection (some maps benefit your go to hero more etc.) and in general how people usually have no clue on maps they played rarly on yet with each additional match might tend to perform more steady.

Ive also noticed how most of the maps Ive played rarely on or are new tend to be placement matches. This could either be due to a low degree of confidence in skill or flawed perception on my end.

PS: Ignore any attempt of topic hijacking

How Competitive Matchmaking and Rating Works (Season 18) has 37 references, many of them from developers. Player anecdotes are of limited value. Even if for some silly reason you consider them trustworthy, player anecdotes contradict each other all the time.

I have no idea what you mean by agnostic here. Note that TrueSkill 2 has performance metric bonuses.

It’s a matter of getting enough statistics to be ranked accurately. By my estimation it takes ~150 games per role to have a high degree of certainty on a role MMR. To break that down by map would take 150 * 20 = 3000 matches to be accurately ranked. I’m exaggerating (someone won’t be a GM on Hanamura but gold on El Dorado), but you should get the general idea.

What Scott describes is the ideal. In practice, it is often not possible because of groups or low populations in a day/time/rank/platform. But you indicate that you know this, so I’m not sure what the confusion is.

I’m not really sure which you’re referring to, but if you’d like me to not reply to certain posts, I can do that. You’ll have to tell me which ones though. Of course “topic hijacking” does bump your thread, so it is not the worst possible thing.