Talk: Matchmaking/Seasonal Rankings

Archives

MMR Estimates: Read this first[]

This page is for discussing the MMR needed for each medal.
Keep in mind that Valve does not disclose the exact rankings, and likely does not use a single value to determine medal rank.
This means that any numbers put forward will only be an estimate.
Do not post here just to say that your MMR/medal is within a few hundred points of the figures given on the page. That is normal.
Do post here if you have reason to believe the entire table is inaccurate, based on multiple accounts (i.e. not just your own MMR)

Discussion[]

Now a month and a half later, all data points to the cutoffs being the same and almost, if not exactly, accurate. Since the cutoffs are remaining constant we could potentially update the calculations/medals sections to cover the calculation methodology and suggest that the cutoffs are likely accurate rather than highly speculative? Aletheies (talk) 05:59, 25 November 2019 (UTC)

Done - Lemoncake(talk) 06:18, 25 November 2019 (UTC)

This should be the final adjustment for this set of data. Removing the divine data points from the Herald-Ancient set and incorporating responses on the survey generated over the night suggests that the distribution is 154 mmr per medal starting at 0 until divine and then 200 per medal until immortal. The fact that 154 * 30 medal pre divine = 4620, which was the forecasted divine 1 cutoff, further increases the likelihood that this is an accurate distribution. The raw data can be viewed here: https://docs.google.com/spreadsheets/d/1ulLP2AM6JMuwW7-T6Bz8z9liBbK2jxD1deK1IdXUBbY/ Aletheies (talk) 23:45, 11 October 2019 (UTC)

https://www.reddit.com/r/DotA2/comments/dgnwxx/season_4_matchmaking_medal_rankings_by_dota_2/ - Lemoncake(talk) 06:54, 12 October 2019 (UTC)

Nice, very cool to see - Aletheies (talk) 07:23, 12 October 2019 (UTC)

After analyzing additional responses from the survey there were 5 Divine data points with percents, surprisingly (or not surprisingly depending on how you look at it) they were along a nearly perfectly straight line with a slope of 200 and an intercept of 4620 for divine 1. The mean squared error for these 5 data points along this line is only 2.4 which combined with how round the 200 slope is suggests that these values are likely the cutoffs for the divine bracket with divine 2 at 4820, divine 3 at 5020, divine 4 at 5220, divine 5 at 5420, and immortal at 5620. These cutoffs also match every single data point for reports in the divine/immortal bracket. I've updated the article with these cutoffs for the divine bracketAletheies (talk) 08:36, 11 October 2019 (UTC)

Following up on the analysis of the divine bracket, I found the minimum and maximum non-outlier MMRs needed for each medal from the survey:

Divine 1 - 4669-4814
Divine 2 - 4867-4987
Divine 3 - 5000-5212
Divine 4 - 5295-5395
Divine 5 - 5497-5632
Immortal - 5624+

(Also I apologize if I'm not following wiki etiquette with talk pages, do you put newer stuff on top? Or older stuff? What about replies?) Aletheies (talk) 07:59, 11 October 2019 (UTC)

I collected data by adding random people from the dotabuff recent matches tab and by this reddit post: https://old.reddit.com/r/DotA2/comments/dg6br9/new_medal_mmr_survey_help_us_collect_data_on_the/ I was only able to get 8 data points but they range from Herald to Divine and are in such a straight line that I think it's enough to support my findings.

Herald 5, 82%, 723 MMR

Divine 2, 51%, 4923 MMR

Guardian 2, 32%, 977 MMR

Herald 4, 94%, 607 MMR

Archon 3, 51%, 2698 MMR

Legend 5, 13%, 3721 MMR

Legend 2, 10%, 3260 MMR

Archon 1, 88%, 2447 MMR

Finding a line of best fits suggests that the slope of the line is 156 MMR per medal with an intercept of -22 meaning the cutoff for herald 2 is 134 then add 156 mmr for each subsequent medal. This line of best fit has an R^2 value of 0.99988 and a mean squared error of 5.645. It is also independently supported by this post: https://old.reddit.com/r/DotA2/comments/dg36ke/badges_got_updated_4650_mmr_is_now_divine_i/ suggesting that divine cutoff is now 4650 where the line of best fit puts the divine cutoff at 4658. I've updated the page to reflect these findings, they're not perfect but it's a lot better than the page's current cutoffs. Aletheies (talk) 03:49, 11 October 2019 (UTC)

I already read that thread before making this page, your numbers seem low compared to some of the reports. 160 increments is the 'historical' distribution mentioned by Valve today, i think it's justified. We shouldn't take 4650 as Divine 1 based on one account. The numbers on this page are not supposed to be cutoffs, but approximations ~ (both higher and lower). Since Valve said the distribution will be unstable for the first few days, we'll wait and see. But i might have to change it back after everything stabilizes - Lemoncake(talk) 05:14, 11 October 2019 (UTC)

I went through the survey responses that did not include a percentage or were very likely erroneous: Additional my survey found the following percent-less/outlier data points:

Ancient 1, 22 (likely erroneous data point)
Ancient 5, 4590 (matches fit)
Archon 4, 1200 (likely erroneous data point)
Archon 2, 2497 (matches fit)
Archon 2, 2550 (matches fit)
Archon 3, 2750 (matches fit)
Archon 4, 2080, 50% (expected Crusader 4, 47% - most likely accidentally selected Archon instead of Crusader)
Crusader 1, 12112, 43% (likely erroneous data point)
Crusader 5, 2233 (matches fit)
Crusader 4, 4444, 45% (expected Ancient 4, 63% - turns out that Ancient 4, 45% is exactly 25 mmr from Ancient 4, 63%, maybe they selected the wrong medal before their percent updated? Either way 4k mmr being crusader is likely wrong)
Divine 1, 4800, 0% (matches fit)
Divine 4, 4250 (likely erroneous data point)
Divine 4, 5378 (expected Divine 5 - more on that below)
Divine 1, 0 (likely erroneous data point)
Divine 1, 4759 (matches fit)
Divine 1, 4703 (matches fit)
Divine 5, 5632 (matches fit, although it implies a jump between Divine 5 and immortal - more on that below)
Divine 5, 5497 (matches fit, although it implies a jump between Divine 5 and immortal - more on that below)
Divine 5, 5602 (matches fit, although it implies a jump between Divine 5 and immortal - more on that below)
Divine 5, 5503 (matches fit, although it implies a jump between Divine 5 and immortal - more on that below)
Divine 1, 5129, 67% (likely erroneous data point)
Divine 2, 4921 (matches fit)
Divine 3, 5000, 65% (matches fit for medal but the percent is high, it's unlikely they are exactly 5000 though so it's likely they rounded down)
Divine 4, 5360 (expected Divine 5 - more on that below)
Guardian 3, 0, 10% (likely erroneous data point)
Guardian 1, 234, 23% (likely erroneous data point)
Herald 1, 1200, 21% (likely erroneous data point)
Immortal, 6234
Immortal, 6105
Immortal, 6180
Immortal, 5624
Legend 4, 3600 (matches fit)
Legend 5, 3696, 13% (expected Legend 4, 83% although on the survey they indicated they had just lost a game and the percentage had not updated, the difference between Legend 5, 13% and 3696 is 46 mmr. This is more than you would expect for a loss, so this data point is concerning)
Legend 1, 3105 (matches fit)
Legend 4, 3621 (matches fit)
Legend 2, 3335 (matches fit)

Taking out the divine bracket and the data points that are off by hundreds to thousands of mmr from the rest of the data set, all but one data point from the survey matches with the fit. Even many of the data points that are off by hundreds to thousands can be explained with a misclick which makes them fit perfectly within the model. With this many data points I think the current cutoffs work very well for everything below divine. But you are right about the numbers being low compared to some reports - specifically the divine bracket which seems to play by different rules as the cutoff for the mid to high divine ranks and immortal is much higher than the linear model would predict. Aletheies (talk) 07:48, 11 October 2019 (UTC)

Okay, that's fine, thanks for working all this out. You may want to look at the numbers again in a week or so, though, in case there are any changes. - Lemoncake(talk) 08:43, 11 October 2019 (UTC)

5200 mmr divine 1, support tbd

Normal. - Lemoncake(talk) 07:11, 24 September 2019 (UTC)

5370 mmr divine 3. I think with the latest rank reset the numbers are all off Theino (talk) 17:18, 21 September 2019 (UTC)

Adjusted numbers upwards. - Lemoncake(talk) 10:04, 22 September 2019 (UTC)

4089 core mmr legend 6, support tbd - 49.144.234.38 11:39, 21 September 2019 (UTC)