(Possibly) Incorrect and gamebreaking MMR gains and losses

[Hericho]

Hello,

We may have discovered a bug(?) that completely nullifies the concept of MMR, by playing as a Party.
Hear me out: Our party of 3-4 literraly lost dozen or even hundreds of Rank # places in the leaderboard really fast.
It would be fine if we did actually lose a lot of games given the magnitude of it but... We did win 15 games for 4 losses in the last two days.

Sure we also factored that there was a few games which were somewhat unbalanced and would potentially win us very low amount of MMR (If the current ranking system is alike dota2)
However in the past, those type of unbalanced games also happent in SoloQ and the MMR gained/lost felt fine.

Right now, we are thinking of not playing together & play solo to actualy try to regain our ranks, I feel like this is really bad for the future of the game, along the fact that this is possibly a tool that'd be used by smurfs.
I am not sure if this is reproducible or is even happening to other parties/solo players right now.

Here is some more information if you could investigate it with your internal data:

Match won id: 66436786 (Party: Yamato/Mo&Krill/Drifter/Wraith)
Match lost id: 66655977 (Party: Pocket/Warden/Yamato/Drifter)
Full list of the 19 games: 66710419 66699276 66688784 66676918 66666091 66655977 66644638 66635118 66626793 66617284 66436786 66428047 66413074 66402964 66394118 66376827 66360602 66352116 66341975

For myself I went from roughly about top 53 in Asia leaderboard to top 99 after these 19 party games ~ which is close to going from low E6 to high E4 for reference. (This would usualy take me about 15-30 losses in a row for it to happen)

Please contact me if you want further informations.

 

[ivb))))]

This is the reason why games are so unbalanced. Lots of parties versus solo players with a 75-80% win rate.

 

[The Great Goostby]

This is the reason why games are so unbalanced. Lots of parties versus solo players with a 75-80% win rate.

Yup parties ruining the game just like dota.

250 solo queue games, 250 of them with 3-5 kids on MY team going -10 KDA or worse (while I still managed a fucking 20 KDA on my mains, come the fuck on)

IN 250 solo queue games, the enemy has had NOT A SINGLE FUCKING ONE OF THESE 0/20/0 BY 8 MINUTES SUB HUMANS

Parties need to go die in several fires. 99% of them just ruin and abuse games they "play"

 

[squabbernack]

Yup parties ruining the game just like dota.

250 solo queue games, 250 of them with 3-5 kids on MY team going -10 KDA or worse (while I still managed a fucking 20 KDA on my mains, come the fuck on)

IN 250 solo queue games, the enemy has had NOT A SINGLE FUCKING ONE OF THESE 0/20/0 BY 8 MINUTES SUB HUMANS

Parties need to go die in several fires. 99% of them just ruin and abuse games they "play"

Im not saying parties are not a part of the problem but yesterday I had a Bebop in my team that played literally his 2nd game of deadlock ever. And he was with and against people in ritualist with 150+ games. And I dont think he was in a party with anyone, he apologized to all of us in the end.

 

[testingdeadlocks]

Hello,

We may have discovered a bug(?) that completely nullifies the concept of MMR, by playing as a Party.
Hear me out: Our party of 3-4 literraly lost dozen or even hundreds of Rank # places in the leaderboard really fast.
It would be fine if we did actually lose a lot of games given the magnitude of it but... We did win 15 games for 4 losses in the last two days.

Sure we also factored that there was a few games which were somewhat unbalanced and would potentially win us very low amount of MMR (If the current ranking system is alike dota2)
However in the past, those type of unbalanced games also happent in SoloQ and the MMR gained/lost felt fine.

Right now, we are thinking of not playing together & play solo to actualy try to regain our ranks, I feel like this is really bad for the future of the game, along the fact that this is possibly a tool that'd be used by smurfs.
I am not sure if this is reproducible or is even happening to other parties/solo players right now.

Here is some more information if you could investigate it with your internal data:

Match won id: 66436786 (Party: Yamato/Mo&Krill/Drifter/Wraith)
Match lost id: 66655977 (Party: Pocket/Warden/Yamato/Drifter)
Full list of the 19 games: 66710419 66699276 66688784 66676918 66666091 66655977 66644638 66635118 66626793 66617284 66436786 66428047 66413074 66402964 66394118 66376827 66360602 66352116 66341975

For myself I went from roughly about top 53 in Asia leaderboard to top 99 after these 19 party games ~ which is close to going from low E6 to high E4 for reference. (This would usualy take me about 15-30 losses in a row for it to happen)

Please contact me if you want further informations.
View attachment 91125
View attachment 91124 View attachment 91123

You are forgetting one of the major change in ranking system from Dota 2 to Deadlock

TLDR
% percentile relative ranking system
Check your games with statlocker

Deadlock uses % percentile relative ranking system over the numerical number based MMR in Dota 2 so if the player base of respective rank for your case its Eternus gets better or worse depending on your performance you will lose or gain rank without even playing a single game

Due to the Deadlock "relativity" ranking system your can go up and down in rank faster than static numerical ranking of up (+25 to 50) and down (-25 to 50) found in Dota 2

I'm surprised an Eternus like you haven't started using Statlocker dot gg site it will make the ranking system dead set obvious
I'm sure you are serious about this go ahead check the PPP of your winning and losing matches combine, compare and calculate against the total it should addup and makes sense

Due to relativity % system to gain the same rank back after losing 50% you need to win 100% more

 

[Hericho]

Deadlock uses % percentile relative ranking system over the numerical number based MMR in Dota 2

This is where it can get blurry...
The exact yoshi's words : "We are updating rank badges globally today. Previously badges were tied to fixed skill ranges on the backend, rather than percentiles."

With that information, I'm pretty sure they are still actively using MMR behind the scene, only the rank shown is based off Percentiles.
(They've done this in Dota aswell, but I think in a different manner, as they ajusted it in "waves" rather than it being a fixed x % = x rank)

Due to the Deadlock "relativity" ranking system your can go up and down in rank faster than static numerical ranking of up (+25 to 50) and down (-25 to 50) found in Dota 2

This is where it didn't line up with what I experienced. Sure in my case you can factor in the relativity since I'm comparing Rank places on the leaderboard BUT...

Here is two facts:
In the past going from top 100 to top 50 was about a 15-30 game win difference in Soloq.
I did not play for about 2 weeks and my leaderboard rank barely moved (5 places)

For me to lose 50 leaderboard ranks, I would need to have my MMR be lower than the 50 dudes that "overtook" me.
So while I was winning (+11 win difference) there is absolutely no way that the 50 people below me won : enough games to offset my 11 wins, add to that 10 to 20 wins (if I base it off my previous top 50 to 100 data while being generous) ALL OF THAT over the same time period which was over 30h or so ~.
And the same could be said for other members of the party where they also lost leaderboard ranks

Statlocker dot gg site it will make the ranking system dead set obvious

Statlocker uses PPP, this is their internal skill measuring system, it is powerful and "mostly" accurate. But the data they use is like the name suggests: Personal Performance (which is their own maths to judge how a player performed) coupled with the "lobby's rank" which they get from the matchid's.
Statlocker is a great site and tool, but at the end of the day PPP is only an estimation and not the real deal.

 

[testingdeadlocks]

This is where it can get blurry...
The exact yoshi's words : "We are updating rank badges globally today. Previously badges were tied to fixed skill ranges on the backend, rather than percentiles."

With that information, I'm pretty sure they are still actively using MMR behind the scene, only the rank shown is based off Percentiles.
(They've done this in Dota aswell, but I think in a different manner, as they ajusted it in "waves" rather than it being a fixed x % = x rank)

This is where it didn't line up with what I experienced. Sure in my case you can factor in the relativity since I'm comparing Rank places on the leaderboard BUT...

Here is two facts:
In the past going from top 100 to top 50 was about a 15-30 game win difference in Soloq.
I did not play for about 2 weeks and my leaderboard rank barely moved (5 places)

For me to lose 50 leaderboard ranks, I would need to have my MMR be lower than the 50 dudes that "overtook" me.
So while I was winning (+11 win difference) there is absolutely no way that the 50 people below me won : enough games to offset my 11 wins, add to that 10 to 20 wins (if I base it off my previous top 50 to 100 data while being generous) ALL OF THAT over the same time period which was over 30h or so ~.
And the same could be said for other members of the party where they also lost leaderboard ranks

Statlocker uses PPP, this is their internal skill measuring system, it is powerful and "mostly" accurate. But the data they use is like the name suggests: Personal Performance (which is their own maths to judge how a player performed) coupled with the "lobby's rank" which they get from the matchid's.
Statlocker is a great site and tool, but at the end of the day PPP is only an estimation and not the real deal.

I find statlockers PPP to be close enough accuracy similar to Deadlock hero skill level

Statlocker - Deadlock Stats Tracker & Match Analytics

Track your Deadlock matches, analyze performance, and improve. Real-time stats, hero tier lists, build guides, and rank tracking.

statlocker.gg

Match won id: 66436786 (Party: Yamato/Mo&Krill/Drifter/Wraith)

Statlocker - Deadlock Stats Tracker & Match Analytics

Track your Deadlock matches, analyze performance, and improve. Real-time stats, hero tier lists, build guides, and rank tracking.

statlocker.gg

Match lost id: 66655977 (Party: Pocket/Warden/Yamato/Drifter)

Full list of the 19 games:
66710419
[URL unfurl="true"] https://statlocker.gg/match/66710419/summary[/URL]
Win
66699276
https://statlocker.gg/match/66699276/summary Win
66688784
https://statlocker.gg/match/66688784/summary Win
66676918
https://statlocker.gg/match/66676918/summary Win
66666091
https://statlocker.gg/match/66666091/summary Win

66655977
https://statlocker.gg/match/66655977/summary Loss
Ascendant V team loss vs Ascendant 4 enemy
66644638
https://statlocker.gg/match/66644638/summary Win
66635118
https://statlocker.gg/match/66635118/summary Win
66626793
https://statlocker.gg/match/66626793/summary Win

66617284
https://statlocker.gg/match/66617284/summary Loss
Losing as Eternus 2 team vs Ascendant 6

66436786
https://statlocker.gg/match/66436786/summary Win

66428047
https://statlocker.gg/match/66428047/summary Loss
Ascendant 6 team losing vs Ascendant 2 enemy


66413074
https://statlocker.gg/match/66413074/summary Win
66402964
https://statlocker.gg/match/66402964/summary Win
66394118
https://statlocker.gg/match/66394118/summary Win
66376827
https://statlocker.gg/match/66376827/summary Win
66360602
https://statlocker.gg/match/66360602/summary Win
66352116
https://statlocker.gg/match/66352116/summary Win
66341975
https://statlocker.gg/match/66341975/summary Loss
Ascendant 4 team loss vs Ascendant 3 enemy

Ok I checked your basic losses and wins
I think your losses were hard on you and your party due to relative system losing ascendant games meant you lost tons of internal rank points if you hover your losses its often phantom rank PPP

Ideally Eternus party should have Eternus teammates and enemies to maintain balance but I know Asian servers don't have enough players compared to EU and USA so there is big wide gap for lower rank players to join Eternus level games

I do have to remind you that you can swing very wide ranks with a single loss compared to Dota 2 if we were to use Dota 2 scale its -250 to -1k

I'm going to show you recent matches for example
https://statlocker.gg/profile/47563731/matches/75724850/summary Losses C+ -11PP - 13PP Ascendent 1 Team loss vs Ascendant 1 enemy
https://statlocker.gg/profile/47563731/matches/75552752/summary Win S +11PP +9PP Ascendant 2 win vs Ascendnat 1 enemy

Notice how big your PP swing was during your losses

The biggest problem with % tile system is that when you often lose on your worst conditions games you often have worse hero performance which swings the total rank and hero rank which deadlock certainly uses source Deadlock hero rankings

For your specific case since you were losing Ascendant rank games the ranking system for Total rank and hero rank assumed that you were lesser than Eternus which is 1 rank lowers than your

My assumption of was to regain rank back you have to win twice from within the same rank tier but if its over 2 ranks then its x4 the number of games to win but like you were saying there is clearly something wrong with the ranking system because even with the x4 numbers of games to win your party were clearly winning so that leaves to party ranking variable

I still haven't figured out how the party ranking system works unless someone else wants to add some info or knows better or until the Valve API for Match, Ranking gets fixed for Statlocker to work properly

 

[Hericho]

I do have to remind you that you can swing very wide ranks with a single loss compared to Dota 2 if we were to use Dota 2 scale its -250 to -1k

Yes IT CAN BE, But I still feel like you are assuming a thing that is not entirely correct there.

Valve uses some form of MMR, which is your exact skill score, and it places you precisely on the leaderboard.
(Ex: 7000 MMR = Top 40, so 39 players are above you)

We can probably say that the leaderboard is just a continuous line of players sorted by that score.

Percentiles used here don’t move you up or down... they simply cut that line into sections and assign a rank icon to each "slice". In short view them as Categories.
(Ex: Eternus 5 = players between the Top 1% and Top 4%)

Do you see the difference now?

For my case :
On the leaderboard, moving from top 100 to top 50 requires a noticeable jump in MMR. In reality and from experience that usually means something like 15-30 wins of progress.

We can see it there on this statlocker's graph :

From march 14 (after bug): I was sitting at +15 net wins
To april 15 (my last match), I was sitting at +36 net wins

I did the jump again (from top 99 to top 55) after this bug, so it needed +21 win for me to achieve it.


I'll also add PP since you seem to like it, here is the graph:

Between March 12–14, I was around Top 50'ish in game, and on March 14 I was Eternus 3 (PP-wise).
And from March 14 to April 15, I progressed from Top 99 to Top 55, and on April 15 I was Eternus 6 (PP-wise).
Throughout the entire period shown in the graph, I remained Eternus 5–6 in-game.
So this is widely inaccurate, don't refer to PP ever as valuable data in this case.

Ok I checked your basic losses and wins
I think your losses were hard on you and your party due to relative system losing ascendant games meant you lost tons of internal rank points if you hover your losses its often phantom rank PPP

Yes, I mentionned it in the first post where I was aware that the game were not completely balanced every time

Even if in my case loses did cost more and win gave less.
If we go from a usual data that valve uses in dota (that also use MMR)
A balanced match will give or take : 25 MMR
An unbalanced match will give (25+x) MMR or take (25-x) MMR
Lets say that x is about 20 MMR in my case as the matches were highly unbalanced
(Which is unrealistic imo)

Let's say that all of my matches were highly unbalanced (all games were not, but for the sake of explaining we'll do it there)
I won 15 games: 15*(25-20) = 75 MMR
I lost 4 games: 4*(25+20) = 180 MMR
Which results in a total of -105 MMR

If we compare it to balanced matches
105 mmr is about 4 wins or loses worth of balanced matches
Even if I were to win 1 mmr and lose 49 mmr for each 19 games (as it should be the strict minimum)
It would returns 181 MMR, which is equal to 7-8 balanced games

In my case this wasn't just a small adjustment of 8 games, THIS WAS A WIDE SWING OF 20 games worth of MMR.
If I lost 8 games right now (balanced & slightly unbalanced) I'd probably lose about 10 to 20 ranks and not 50.

My assumption of was to regain rank back you have to win twice from within the same rank tier but if its over 2 ranks then its x4 the number of games to win but like you were saying there is clearly something wrong with the ranking system because even with the x4 numbers of games to win your party were clearly winning so that leaves to party ranking variable

So yes, we both came to the same conclusion, that there might be a party Q rank variable/multiplier that is causing this bug or incorrect behaviour.
Either this or the MMR range given/taken in a single match is WAY too wide , but I don't think this is the case, since I'm playing in Asia and didn't feel like I encountered such wide gaps in SoloQ (It's easy to get a grap of it since I can compare it on the leaderboard)

 

[van17ino6]

Yup parties ruining the game just like dota.

250 solo queue games, 250 of them with 3-5 kids on MY team going -10 KDA or worse (while I still managed a fucking 20 KDA on my mains, come the fuck on)

IN 250 solo queue games, the enemy has had NOT A SINGLE FUCKING ONE OF THESE 0/20/0 BY 8 MINUTES SUB HUMANS

Parties need to go die in several fires. 99% of them just ruin and abuse games they "play"

take your meds