Ranked Wars should be both fun and lucrative, rather than fun or lucrative. There is unlikely to be one single change that will achieve that, so we are looking at a whole range of tweaks. The first thing we are looking at in this programme of work is trying to improve matching and ranking.
We intend to change how matching and ranks work with the goal of ending up with a system where:
- The vast majority (>90%) of a faction's matches are winnable matches against factions with a similar stat range
- Factions should continue to get matched the vast majority of times they enlist
- Rankings should reflect -- as far as possible -- who we think can beat who
We're posting this to get feedback from the wider community on this change, specifically. We will read and consider all feedback, I will try and respond to as much of it as I can and think is productive. We won't make everyone happy, sorry. We're not aiming for perfection, we're aiming to improve over the current system.
High-level plan
Currently, matches are chosen randomly for each faction from a pool of factions with a similar number of members, from ±1 division, not taking into account a faction’s stats strength. Factions are promoted or demoted by a division depending on if they win or lose a war (with a couple of special cases), and rankings inside of Diamond Rank are based on a system of positional swaps.
We will move instead to a system where every faction has a hidden numerical rating, where higher is better. Rating is based on multiplying a faction’s stat-score (using a very, very similar formula we use to calculate Hidden Rank number) with a performance variable based on their performance in wars. Performance will be adjusted over time using a system derived from ELO.
Rank and division will be determined solely by a faction’s rating relative to other factions, although we will keep the member-count requirement for Diamond.
For matching a faction, we’ll pick an opponent at random from their pool of similar-rated peers. There will be a hard cap on how mismatched a fight can be, but with enough natural variance that you won’t see the same handful of opponents on a fixed cycle.
The factions who are the largest in their player-size bracket tend to be much, much bigger than the factions below them. It is very hard to find fair matches for CR/NS/MHQ (>90 members), for Monarch Retardation and [Redacted] (~40 members), and impossible for Inglorious Basterds (~25 members). For these factions, we will aim for harm-reduction for their opponents by trying to fairly rotate the number of times they have to face them (getting "sacrificed"), keeping it at less than 10% of the time. As part of the algorithm for this, most factions will also occasionally be paired (ideally <10%) with significantly weaker factions (getting "coddled"). We will reach across at most 2 divisions (eg Plat III->Plat I) to achieve this.
Pictures
😎 Sunglasses on because I don't use dark-mode 😎
Below we have comparisons of the last 40 wars each faction was in, and what the next 40 would look like under the new system. Note that these are arranged by colour, not chronologically. The rating given is how it looks after the last 40 wars the faction was in. We use the historical enlistment rate to predict future enlistment rate.
Orange and yellow mean a match-up was "too easy" for a faction (coddled), pink and red mean it was "too hard" (sacrificed). Green means we think it's a reasonable match-up. White means the faction didn't enlist, and purple means that they enlisted, but we didn't find a match for them.
Here's the top 20 factions by rating, before = left, after = right
400 to 420:
Here are twenty weeks of matches for Just Fer Khaos (#11), historical on the left, simulated on the right:
(nb: this was mislabeled as being The Next Level on the original post. sorry)
For The Revenant (#263):
Shout-outs
Particular thanks to Kamikaze-Tool, Qfiffle, Omanpx, Helmstetter, and other players who've helped discuss this with me, including the many hundreds who responded to DMs about their RW experiences, esp the rest of the staff team.
Questions and answers
How does this handle terming / can this be gamed?
Even though we are pretty good at detecting termed wars, the algorithm will assume that all wars are real. Your rating is made up of your stat-score and your performance. You can't really game the stat-score component, because additions or subtractions from your roster should genuinely affect your fighting ability. You can to some degree game the performance side of it by terming, but your reward is that you get slightly harder or easier matches, and when you stop terming then that will drift back to its natural state. The ELO algorithm is self-healing: if you've termed your way to an unrealistically high performance, your performance score will see a relatively big adjustment when you lose to team your rating thought you should easily beat. It's worth mentioning as well that we cap the performance variable, so there's a limit to how high you can go if you don't have the underlying stats for it (and how low).
You know this won't fix terming, right?
This will not stop factions terming, but, we hope it will cause a moderate reduction in terming, as we will largely remove the ability for one side to dictate terms, and also strongly reduce the number of times factions are faced with an unbeatable opponent. It does nothing to address the cost of wars, which is an issue at the higher end, for example.
I still see some unfair matches you've marked as fair
Not every match-up will be easy. Some will be hard. They should all be plausible, but part of the algorithm learning is that you will occasionally get a very hard or very easy match-up that we thought at the time was fairly well matched. Factions change over time, and a factions performance variable will update to match it. For a faction that doesn't term, over time their win-rate should trend towards 50%.
How exactly will rank and division be determined?
Strictly by rating. Say we decided we want 100 Diamond factions across the four Diamond divisions (which is probably around where we'll end up) then the top 25-rated factions with (>=75 players) will be Diamond III, the next 25 Diamond II, etc
If you're using an ELO-like system, what are the coefficients you're using? How exactly does it work? etc
We don't want to remove the pleasure of reverse engineering the numbers from our more analytical players. Plus we are likely to be tweaking them over time.
Didn't it used to work like this before "Open Letter to Ched"
That system did indeed take into account stats for making matches, however, it caused stronger factions who had few peers to be starved for matches, and meant the ranking didn't reflect who could beat who. This new proposed system should fix both those issues.
Does this mean smaller-stat high-activity factions won't ever get matched against truly hard factions?
So the text does explicitly mention a "a limit to how high you can go", but in modelling the current limit doesn't get hit in practice. As an example of how that might work more concretely, a faction that's incredibly good at fighting and has players with an average of 1bn stats could theoretically get matched to a faction that's never won a war, and has 25bn average stats. Another way to see it: in theory, the system would permit a Monarch Leviathan who won all of their wars to match with an NSIII who lost all of their. In short: I don't think this will be a problem in practice, but it's a number that's very easily tweaked if we see factions who do manage to get their performance rating high enough.
When is this getting released?
I don't know. Implementation shouldn't be a huge piece of work now the majority of the simulation work has been done. It's both a high priority and has people actively working on it. This isn't a future project, this is an under active development project.