Table of Contents

• Read Context if you want a quick overview of what we're talking about.
• Read The Problem if you want to read my stance on the limitation of Ranked 2.0.
• Read Solutions to check out my proposed solution(s).
• Read Technical Details for an in-depth history and analysis of the SBMM in R6:Siege.

Context

With Operation Solar Raid, Siege is revamping the Ranked experience with Ranked 2.0. If you’re not sure what this entails, this CoreRoss video explains the whole thing.

In my humble opinion, the change is generally a great idea! It is a big improvement for the vast majority of players (the casual- to intermediate players), whilst barely impacting the competitive/high-level players.

An overview of the pros and cons in my opinion:

As you can see, many more pros than cons!

The Problem

The new Ranked 2.0 system hides the Skill-Based Matchmaking value completely (same as it has always done for Quick Match and Unranked). Instead, the visual Rank you see is based purely on the number of matches you have won, and the Skill-Based Matchmaking is independent of the visual Rank.

You are left with only being able to measure your personal performance on metrics such as K/D, Winrate, HS%, KOST, etc.

These are great metrics, don’t get me wrong. But they are all CONTEXTUAL metrics. A K/D of 1.2 of a Ranked 1.0 Silver player is not comparable to a K/D of 1.2 of a Ranked 1.0 Platinum player.

More importantly, if your K/D last month was 1.2, but this month it’s 1.1, you can’t really say you got worse without putting it in the context of which Ranked 1.0 rank you were playing in during both months. If you were in the same Ranked 1.0 Rank, then yes, you got worse. But if you were playing in Silver last month, and Platinum this month, then it’s not fair to say you got worse.

The problem then, with Ranked 2.0, is that we lose all context for performance metrics, as the actual COMPARATIVE skill measurement (SBMM “MMR” value) is no longer available.

The Ideal/Hopeful Solution

I cannot know for sure if I can “expose” the hidden MMR value next season. Not until the season is launched. If I can, that would obviously be the perfect way to keep tracking performance and “Ranked 1.0 Rank”, or SkillRank as I’ll likely call it.

The Alternative Solution

In the (altogether likely) event that the hidden MMR value will not be exposed client-side at all next season, I have a proposed alternate solution, which I hope to get your input on!

In a perfect system, I can simply keep tally of all matches played next season, and run the same algorithm (or a close-enough approximation) to keep track of all wins/losses and raise/lower a MMR value for all accounts, and show you that.

However, this assumes I can track all matches, which is simply not true. My app is used a fair amount, but not enough to track literally every single match. Thus a solution must be devised that can approximate the hidden MMR with incomplete data.

Here’s my proposal:

I have current MMR data on practically every (PC) player currently playing. I keep this data (no “MMR reset” for next season). I doubt Ubisoft will reset either, as it’s better not to reset. (See below for technical details on the failures of the current Ubisoft MMR implementation, and how I hope they improve these.)

At the start of a match, I calculate the average MMR of all players in the match. Let’s call this AVGM. I also find the maximum and minimum MMR of a player in the match: MAXM and MINM.

At the end of the match, I track the win or loss, and set the new estimated MMR (eMMR) as:

On win:
If eMMR < AVGM then eMMR is set to AVGM.
If eMMR > AVGM then eMMR is set to MAXM.

Similarly, on loss:
If eMMR < AVGM then eMMR is set to MINM.
If eMMR > AVGM then eMMR is set to AVGM.

The advantages of this simple implementation are:

If a player’s hidden real MMR is greatly different from the eMMR, this will very quickly update eMMR to a much better approximation.

If a player’s hidden real MMR is close to the eMMR, it will only increase/decrease slightly (and in the correct direction) which should maintain a “near enough” approximation.

Limitations:

Obviously this is an imperfect guestimate of the true hidden MMR. Mostly because I have imperfect information. However, it should be very serviceable as a metric to compare against other players, and especially to compare your own performance over time.

Use in the app:

The point of this whole implementation, is to maintain a visual representation of the skill-based metric within the app. I intend to KEEP THE RANKED 1.0 MMR STEPS (so no Emerald Rank, and no Platinum 4/5 etc.) and call the changed metric SkillRank to differentiate it from the “VisualRank” of Ranked 2.0.

Your input:

I absolutely 100% understand the need and want of the game developers to hide the true SBMM MMR value. In a perfect world, there is no need to hide it. But in the real world, where people attempt to manipulate the matchmaking to “win more”, it is better to hide it.

So here are some questions I would love your input on:

• Is it desired to have a SkillRank metric in the app?
• Is it ethical?
• What about making it an opt-in metric, where you can only see it for accounts that opted in?
  • It would still have to be tracked (hidden) for all accounts in order to work properly.
• What are your thoughts on keeping the Ranked 1.0 ranks as visual “translations” of the eMMR value?
• Any other thoughs/feelings you have!

Again, please join the Discord or reply to this Tweet!

Below is a load of information that you probably should read, but likely won’t. It’s up to you!

Often Misunderstood Technical Details

There is a lot of confusion about how the current Ranked 1.0 system even works, and also what changes with Ranked 2.0. It’s surprisingly difficult to explain, so I’m not surprised! I also suspect that Ubisoft is intentionally a bit vague here and there to hide some of the downsides/weaknesses of the systems.

Ranked 1.0

In the current Ranked 1.0 system, your rank is a category based on your Matchmaking Rating (MMR). This is a system value that determines who you are queued up with- and against. Skill-Based Matchmaking (SBMM) as it’s called, create match-ups based on similarly-skilled players. The SBMM algorithm R6:Siege uses is a custom variation on Microsoft’s TrueSkill2.0™ system. Here’s a paper explaining the details, which is actually fairly easy to parse.

Basically, TrueSkill assigns a value (MMR) to each player, and this value goes up on wins, and down on losses. How much it goes up or down by, is a function of how certain the algorithm is in the accuracy of the skill-value on that player. The accuracy goes up as you play more matches, as with more information, the algorithm becomes better at predicting your performance. Thus, you gain or lose fewer MMR points as you play more matches.

Example:

Let’s say you start with 2500MMR. It matches you against other players around 2500MMR. You win, and as this is your first match, the system has no idea of your actual skill level, so it adds a lot of points. Let’s say +500, so you are 3000MMR now.

You are matched against other 3000MMR players, and now you lose.

The system now removes points, and it’s still a lot, as it is not sure yet. But it’s more sure then before. -400 MMR points.

On-and-on we go:
Win, Loss, Win, Win.
+300, -200, +150, +120

You are now 2970MMR, and you start winning about 50% of your matches.

This is the goal of the system, bounce you back and forth in smaller and smaller increments, until you are in the skill-area you should be.

This is important, so let me repeat:

The goal of Skill-Based Matchmaking is for you to win 50% of your matches.

The Problem With Ranked 1.0

The system works. The problem (as with most things) is people.

Part 1: People are idiots that want to win, more than they want to have a good game

The moment you show this MMR, people attach prestige to it. They think they are “better than other people” when they have a higher Rank. And thus people start lying to the system, intentionally doing whatever they can to “break” the system and get matches against players that are lower skill then themselves, so that they can “win more”. Even though the whole point of the system is to play a fair and exciting match! If you win 50% of your matches, each match is a great competitive experience.

But people want to win 100% of their matches, so they want to be queued against players worse than themselves. Yet they choose to play a SBMM game?

People started queueing with accounts much lower ranked than them, in an attempt to get into matches with worse-skilled opponents. Thus the developers had to institute a MMR squad-restriction.

Part 2: People are impatient and over-estimate their own skill

Players often feel “stuck” in a lower rank than they think they “deserve” to be in. So it’s an industry standard to reset the MMR of all players every season, to give players a chance to rise to the rank they “deserve” quickly.

And then, obviously, inevitably, everyone (after playing a whole bunch of unbalanced matches) ends up in practically the same rank they ended in last season, because that’s actually the correct rank that they should be in!

Part 3: People want to progress and “get something” for winning

A couple years ago, players started complaining that, as they were playing more and more matches, that they were only getting very small amounts of MMR. They wanted bigger rewards for winning.

So the game developers instituted a 25MMR change minimum, to keep their most engaged players playing. This was a really dumb idea, and a clear case of “People don’t know what they want!”

The whole point of TrueSkill is that the MMR change becomes smaller as the system becomes better at predicting your skill level. By instituting this custom large minimum change, a very large variance was added into the matchmaking. Now a lucky streak of several wins in a row could put you out of your skill-depth by such a wide margin, you are destined to go on a losing streak.

In fact, your opponents may be so much better than you, that you might begin to suspect they are cheating, because you don’t have the game-knowledge to fathom anyone being that good at the game.

This minimum MMR change adjustment caused a cascade of issues.

False cheater reports skyrocketed. Complaints about “bad matchmaking” grew drastically.

But, conversely: People played more Ranked and for longer. Because they actually felt like they could rise ranks quickly with just a couple of wins! Wow, progression is a strong motivator…

Enter Ranked 2.0!

Ranked 2.0

The new Ranked version hides the Matchmaking Rating. Instead, this feeling of progression is created by simply winning matches. It’s the best of both worlds. People play more as they feel a sense of progression and earned rewards for winning (exactly the intended purpose of Ranked 1.0), without having to make concessions on the quality of the SBMM algorithm to make it compatible with the needs of players’ fragile ego’s.

MMR squad-restriction can also be lifted.

Basically, it's still the matchmaking of Ranked 1.0, but with a separate progression.

And because the matchmaking can be done hidden, it doesn't have to be dysfunctional in a way to keep people playing.

So here is my plea to the game developers:

Revert the SBMM algorithm changes you’ve made to placate the masses and generate a sense of progression. Now that the sense of progression is coming from Ranked 2.0, you can use the best version of the SBMM algorithm: the unaltered version!

Also please don’t do MMR resets anymore. The only thing those resets do, is throw out the door all the certainty that the system has gained about the skill of each player. If players improve their skill, they will quickly rise up the ranks, and be rewarded for their learning by a nice long winning-streak.

Some thoughts on using metrics outside of match win/loss for SBMM purposes

The game developers have publically mentioned that they tried various variations of SBMM algorithms (likely variations on TrueSkill2.0) that incorporated KOST, K/D, SRV, and various other metrics to do a personal performance MMR adjustment. Their conclusion was that, in the end, all that did was add much more noise and variance in a system that was already perfectly serviceable.

At the end of the day, winning or losing as a team is a great all-encompassing metric for measuring not just team performance, but individual player performance as well. (Your performance not just as a player, but as a teamplayer, in a team-based game!)

Thanks for reading, and hit me up on the Discord if you have any questions, comments, or concerns!

Tweet at me @R6AnalystApp.