Rating does not reflect current skill, only what you have done previously. When there is a pool of players sharing a rating it becomes meaningless to try to get any good predictions out of it since people improve drastically and at different speeds, same for opponents. At the same time the players might even be swapped out for new faces.

Wisdom has been spoken this day

I agree, even knyte will agree this is wisdom

I agree with the observations but not the conclusion (collective ratings appear to have excellent medium-run predictive value, since the constraints are usually met). But that debate sounds exhausting and I'm not going to gather data, so I just dropped the thread. Anecdotally, you see this in the relative stability of most clans' CW rating & the win rates generally converging to 45-55%. Ratings are real.

Edited 10/31/2022 21:41:31

We all agree on all the points that matter.

Ratings are good enough to separate clans, and if the same members continue playing clans of near-ish ratings over large periods of time, the ratings become accurate.

Members may change over seasons, but more often than not, the volume of changes isn't sufficient to radically change the rating. And where it does, there's an adjustment period, but it will over time once again converge a good enough indicator of overall clan skill.

For the M'Hunters internal ladder, I have set the K factor to a relatively high 40 in order to get some action in the ratings. I don't know the exact amount of points 2 equally rated clans exchange and there is maybe some fudge going on due to the certanity factor but out of experience, a bad CW day for our has easily lost us over 100 points. So the notion that the rating is not accurate due to players swapping clans looks a bit strange to me. In the past it seemed to take a skilled hyper active clan maybe 2 to 3 days to reach a stable rating while starting from even below 0.

Edited 11/2/2022 16:55:53

My point was more that you wouldn't be able to predict a win rate or placement based on the CW rating at the start of a season. Ratings might stabilize at the middle or at the end of a season and have a more or less accurate CW rating for the remainder.. (Win rate starts off high / low and then it goes towards 50% but that is not based on the original CW rating at the start of the season, but rather the new one you have achieved from playing during it). In my opinion it is a good tool to match clans but not for prediction. ~ Why not reset the rating at the beginning of a season and have a very high K-factor for the first few games? That way underrated clans won't have a big advantage over an overrated clan in the beginning of the season.

Yep, and as Norman points out, ratings can fluctuate wildly just due to pure chance, and this is going to vary clan-by-clan. Trying to capture, in a single score, the performance of a group of dozens of people across a variety of tasks is going to have a lot of noise.

Why not reset the rating at the beginning of a season and have a very high K-factor for the first few games?

Because past seasons' performance has significant predictive power; it's not zero information, and because the convergence period is going to advantage high-skill clans even further by creating essentially random matchmaking for a bit. TrueSkill doesn't have a K-factor per se but it also has parameters you can use to control the learning rate: setting the learning rate too aggressively can cause wild fluctuations, instability, or converging on the wrong optima.

That way underrated clans won't have a big advantage over an overrated clan in the beginning of the season.

Well, they would. If everyone starts at the same rating (with the reset), then half of clans will be underrated and half will be overrated. This proposal would exacerbate the advantage.

Imo, the silver bullet solution is dropping CW rating (at least for matchmaking) and matchmaking based on individual rating. The idea that we must have CW ratings because this is a clan competition is a non sequitur based mostly on word association. Individual level ratings are just so much more stable & predictive than clan-level ones, and this would solve the problem where only high-skill clans meaningfully benefit from high-skill players (since anywhere else, a high-skill player would just cause their teammates to have more losses).

Edited 11/2/2022 19:50:22

Yeah that's a really solid point actually knyte.

Individual ratings make way more sense. And winrate would actually probably scale down closer to 50% if it was implemented (since there are more possible opponents). This would make CW more of an activity based sport than it already is.

...so actually I don't want that change lmao

Win rates wouldn't converge to 50%, they'd just fail to converge for a different group.

Right now, your win rate (individually) strays from 50% if your skill level deviates significantly from your clan's rating (≈ the participation-weighted mean skill estimate of your clan). Clans' win rates stray from 50% when their opponents' mean skill level deviates significantly from their mean own skill level (≈ their own Clan War rating). This means that only a few clans can sustain significantly-above-50% win rates.

With individual ratings, your win rate (individually) would stray from 50% if your skill level deviates significantly from your opponents': i.e., the same way Rufus or alex have >>50% win rates on ladders, because they break the matchmaking. Clans' win rates would stray from 50% when a significant portion of their players breaks the matchmaking. This still advantages high-skill clans, because you can't face your clanmates.

We'd simply redistribute matchmaking error.

Clan Wars would remain skill- and participation-based, where your results reflect # of games played * win rate. But the distribution of clans' win rates would become smoother: there'd be more clans with win rates between 50% and 80%, rather than the sharp jumps we see on the margins. We'd probably see the competition become more enjoyable for more players, due to fewer lopsided matches. We'd probably see it become more competitive overall, because clans would see a smoother ROI curve for developing player skill.

M'Hunters, Myth Busters, and Optimum would face more competition for podium spots, so in that regard you'd stand to lose from increased parity. Right now there's a step function for return-on-skill, because there's a rapid jump from 50% to 67%+ from clustering of high-skill players. Instead we'd see a smooth curve where clans with one elite player would hit 50% and change, clans with 5 or 6 would get 60%-ish, and so on. MASTERs would still get 80% just because all of them are great, although this change might attract Python/etc. to participate again. But on the other hand, the competition would become more enjoyable & dynamic.

Actually, if I had to suggest a single change to Clan Wars, it'd be this one.

edit: so I Uservoiced it- https://warlight.uservoice.com/forums/77051-warzone-features/suggestions/45930352-matchmake-clan-wars-based-on-individual-ratings

Edited 11/3/2022 10:32:08

What's Optimum?

https://www.warzone.com/Forum/141949-happened-optimum

That's Optimum. Don't you remember it? You played in IPL (https://tinyurl.com/infinity-premier) from those days. After nuking the clan for the first time (as an intoxicated social experiment), I repurposed it for other projects. It was a diplomacy clan under Miego before collapsing (getting nuked). Then it was a newbie recruitment project on top of Open Clans (https://bit.ly/optimum-prime) before- you guessed it- getting nuked, by me. Then it was a casual hangout clan before I handed it over to forksandwich, who- as a gesture of goodwill- nuked it for me. Now it is the home for a breakaway faction of M'Hunters, the Empire of Laconia to their collapsing Mars Congressional Republic, as the gates opened by Clan Wars transform Warlight society & reshape the future. Generations of builders may have fallen to threats beyond their comprehension, but their powerful structures remain...

Edited 11/3/2022 23:16:21

Oooh, I remember! The logo changed, so I didn't recognize it. Anyways, support your local M'Hunter. When you see one walk, why not say hi or give them a hug? Together we can make the world better.

norman* stinks or else he could have all the hugs and kisses ever

Edited 11/4/2022 12:50:12

norme is you clanmate. haha edited. classic error norme/norman

Edited 11/4/2022 12:51:24

if Norman hugged me I wouldn't have daddy issues :(

Will any of last season's CL players play for MH in the upcoming season? Or all of them for Optimum?

Will any of last season's CL players play for MH in the upcoming season? Or all of them for Optimum?

Optimum is the continuation of M'Hunters CL team. Optimum has claimed M'Hunters CL spot and M'Hunters has accepted this.

Also, sorry for coming forward so late but I have in depth information regarding a cheating incidend in M'Hunters. Back like 6 years ago, when M'Hunters first joined the CL, the players were pretty insecure how to play their templates, so Norman has offered to help his mates in private. I know of at least 1 or 2 incidents where Norman has helped some players during their games. If requested, I can give screenshots from the mailing lists between Norman and the then CL team.

Anyhow, I now request the CL administration to be consequent with their former ruling:
--> Give Norman a 2 season ban.
--> Give the players who have received the help a similar ban (once they come back, they haven't played for years).
--> Give Optimum a 15 point penalty.

For those that don't know that is an Alt of Norman. The biggest give away is that it hasn't earned any points in the last 30 days. You know points would require you actually play.

Its one of a few Alts of his that have some how found their way into MH lately, it sad when he thinks dummy accounts might look like his clan is thriving again.

Edited 11/15/2022 21:53:25

For those that don't know that is an Alt of Norman. The biggest give away is that it hasn't earned any points in the last 30 days. You know points would require you actually play.

One thins is kinda sus. For [IM]YouMustBeKidding it shows "Plays from Hungary" (due to the long username, needed to dig into HTML) and for Norman it is "Plays from Germany".
Ofc I know what a VPN is, but usually it switches the locations to countries, which have good datacenter infrastructure: Germany, Ireland, Netherlands. So unless that is a very deliberate conspiracy, then the switch from Germany to Hungary was not very expected.