How Competitive Skill Rating Works (Season 14)

From the data you actually collected, you are gaining more SR on a victory than you lose on a defeat. Also, as of this moment, you have a 58.8% win rate.

These are climbing numbers. As long as you maintain them and continue to play, you will continue to climb.

The primary reward for ranking up is higher quality games. People that say that all ranks are the same are wrong.

If your skill is 250 above your SR, you will win more games than you lose and rank up over time.
If your skill is 500 above your SR, you will win most of your games.
If your skill is 1000 above your SR, you will hard carry. Games will be faster and the conclusion will almost never be in doubt.

Sure. Make sure to reference the current post (See above).

The significant majority of the evidence is that MMR is still in the game, but Scott is sloppy with his language. The section ā€œBut I just read this post from a developerā€¦ā€ is my current view on the topic.

Thanks for the data. Iā€™ll look through it in more detail, but my first impression is that Blizzard made a silent change. My data saying that there are no streak bonuses in low ranks is from season 6 (Stevo's Bronze to GM Analyzed - Google Sheets). Streak bonuses still do not exist in diamond and above (recent data: Streamer Data - Google Sheets ā†’ NerfLucio 13), but your data makes it look like streak bonuses now exist in plat and below.

Once small divergence is that Stevoā€™s run was an established account, while Teslaā€™s is a newish account.

Yep I failed to mention that once he accesses to Diamond, it stabilizes around 25. I would myself try to confirm it but it is very unlikely for me to win more than 5 games in row. Even more to keep the same kind of performance all the way! Maybe other players could help with that?

I managed to get 6 wins in a row yesterday. Hard to say if thereā€™s been any kind of buff after the 4th winā€¦ My data:
https://docs.google.com/spreadsheets/d/1cD4boQC1foCNoGYQeEJLGGYcKZRrQIUtvoY_QF64s6w/edit?ts=5c42abb6#gid=0

Note that in the first one, I played Brig to support a five stack playing Goats. Didnā€™t do much but still didnā€™t do that bad to deserve only 16SR. I believe the slight SR average advantage (+47) might have also been seen in MMR because there was a ā€œlong wait to find a balanced gameā€ and the 5 stacks normal better win chances coul Od have played a role here, taking maybe 2 or 3 SR away.

On my fifth win, I performed very well, Gold elim and card elims 63% kill participation with only one death. Confirmed by Oversumo and Overbuff. Normally I would have expect 24-25 for that kind of performance. I got 24. Since it was the fifth and looking at Tesla, I was expecting a 25% buff, so maybe 28-29SR. But nope.

Just a data simple, but it might show that for established accounts, the streak bonus doesnā€™t apply, or apply later than the 5th or 6th game.

BUT WAIT the players have concluded that luck doesnā€™t exist and people simply belong at their rank :joy:

That link is private. To make it public, go to Share (in the upper right), select ā€œAnyone with link can viewā€ and copy the link in the dialog.

Yep, sorry:

https://docs.google.com/spreadsheets/d/1cD4boQC1foCNoGYQeEJLGGYcKZRrQIUtvoY_QF64s6w/edit?usp=sharing

There appears to be some new, relatively sophisticated algorithm. DongQuixote (above) indicates that it can happen in a downward direction for established accounts.

My inferences for the rules are:

  1. If a player has very high statistical performance AND
  2. A high win percentage AND
  3. Is on a win streak AND
  4. Is below diamond
    He gets highly exaggerated gains.

The reverse also seems to be true:

  1. If a player has very low statistical performance AND
  2. A low win percentage AND
  3. Is on a loss streak AND
  4. Is below diamond
    He gets highly exaggerated losses.

But because of all the conditions, and the relative lack of data, it is difficult to be sure.

Actually the win streak portion used to be true but no longer seems to be. Below diamond, streaks only affect SR gains downward. I watched a recent bronze to GM from a t500 (Linepro), and he didnā€™t experience any winstreak bonuses; only performance bonuses.

This post:

https://www.teamliquid.net/forum/overwatch/512584-comprehensive-overwatch-competitive-play-guide

mentions win streaks, but if you do a quick search for it, a lot of people are saying theyā€™re gone now. It also fits with what I saw on the stream.

Loss streaks may also be gone as this was in season 11 ish that I witnessed these.

I canā€™t find patch notes for either.

I concur with your analysis.

It could even be more sophisticated by applying a streak buff relatively to the statistical performance anomaly. For Tesla it was after 4 games, then again after 8 apparently. For just above or below average performance, first buff could be applied after 8 or 10 games for example. Preventing that way most common lucky/unlucky streaks getting buffed but rewarding or penalizing players who might not perform statistically well or bad but who manage to impact matches results either way.

1 Like

I quickly parsed the first day of Lineproā€™s run, and win streaks are definitely visible: Streamer Data - Google Sheets ā†’ The 500IQ 14. By row 14, his gains are 70+ for 11 wins. He then loses a game (-18 SR) and wins one (+24 SR), so his win streak bonus reset.

His SR gains do bounce around a significant amount but he was memeing a lot (battle Mercy, playing heroes he is not great at, not getting all golds, etc.)

I also reparsed the data the ursolost made me aware of (TheGoodSucc 13), from last season, and it shows pretty obvious win streak bonuses, that reset each loss.

Including me, above. You canā€™t trust everything you read on the internet, especially when Blizzard silently changes the rules.

Yep.

I updated the original post. Thanks for your assistance.

I recently placed two accounts. So it might help you with the data.
I only played quickplay on the accounts and tryharded until level 25. By the end i was getting matched with diamonds and masters mostly with random gm showing up once in 3 matches.
This account:
Went 7-3 with tanks and got placed in 2699 while only playing tanks and supports. Died a lot (7-8/10min). However had good stats overall.
Other account:
Tryharded and was getting matched with plat/diamonds (no masters or gm showing up).
Went 7-3 as well during placements but this time played hammond, phara (14 minutes) mccree (8 minutes) and zen (4 minutes). However had arpund 29elm/10 minutes and 4 deaths/10 minutes. The account was placed 3056. Right now after 113 comp matches, itā€™s at 3206.

Thatā€™s interesting! I think what that data confirms is that no matter what someoneā€™s MMR is in QP, everyone starts at the same point in comp. Estimated at high gold I think. And performance, more than results, is what determine someoneā€™s SR after first placement matches.

Note also that your 7-3 could bring different results even with the same average performance depending if you had a streak or not. A 3 losses, 7 wins in a row would bring more SR than a 2w-1l-2w-1l-2w-1l-1w for example.

Actually I went w-w-w-w-l-l-w-w-l-w during the placements.

I think you should make it more clear that number is for 3000+ accounts because I can almost certify that the neutral sr gain, or when a game is perfectly balanced and your performance is close to the average of the players of your rank playing the same character, the neutral sr gain below diamond is lower than 24.

I have no hard evidence but I think it is about 24. Iā€™ve been pretty much the same SR for a fair while indicating Iā€™m probably average for that SR. Yes, sometimes I win or lose more or less, but on average itā€™s 24. I had a spreadsheet tracking my SR to figure this out.

Which account was this for?

Performance can make a significant difference, but so can win/loss order. The earlier games count for more.

Thatā€™s where the data comes from, but I think it is true for all accounts. Have you written down all your SR changes for a large amount of games? If yes, I can do analysis.