TL;DR on the TL;DRs: Clan Wars would be a better experience and considerably more competitive if we had player ratings and adopted Derfellios/hedja's proposal.
TL;DR re:hedja: Strong +1 to hedja's and Derfellios' proposal. It doesn't have the flexibility of min weight matching but we don't need min weight matching if we are only interested in quality matchmaking w.r.t. absolute rating differences (rather than something fancier like logarithmic rating difference) because rating differences are commutative, making CW matchmaking a simpler/narrower problem than general min-weight matching, with nice properties we can use.
TL;DR re:Ercole: Player ratings would also make Clan Wars more competitive by making high-skill players equally valuable for lower-skill clans as for higher-skill clans. Right now, only predominantly-high-skill clans actually benefit from high-skill players. In non-elite clans, elite players' wins become their clanmates' losses; this creates an incentive for elite players to migrate to elite clans (if they're perfectly rational and motivated solely by CW).
To add to what hedja said and to merge a discussion on that algorithm proposal (which, afaict, is the same as Derfellios' proposal) from WZ Public Chat Discord (
https://discord.com/channels/204926708795572226/888350086894063627/888362365895200779):
1. If there is a concern about match quality, the un-pairings in the second stage (unmatching & rematching to accommodate leftover players) could be gated by a maximum rating difference threshold. I.e., if you have the following sign-ups, in the notation of (clan name)=(rating):
A=100 B=90 B=90 C=70 D=50 D=50 E=30 F=20 F=20 F=20 F=20 F=20, and a maximum threshold of 40 (applicable only to the second stage)
The greedy algorithm would pair AvsB, BvsC, DvsE, DvsF, leaving 4 F's unmatched.
The second stage (Derfellios' and hedja's proposal, constrained by the threshold limitation proposed by me and Derfellios) would break up DvsE to create 2 new matchings (DvsF, EvsF) but not break up BvsC since BvsF and CvsF would be too "low-quality" matches.
(
That said, I don't think match quality warrants this additional complexity. The greedy algorithm does not guarantee match quality, so having it uniquely guaranteed in the second stage would be incongruous. Plus, as Balthromaw pointed out, Clan War Ratings guarantee
low match quality for players whose skill levels are significantly higher or lower than their clans. But as we can see, a trivial threshold-based adjustment to Derfellios'/hedja's proposal would resolve any match quality concerns.)
2. Per Derfellios' reasoning, this approach has provable optimality w.r.t. creating the maximum possible number of games (subject to the match quality constraint, if we incorporate the proposal in #1). Although it
does not guarantee a minimum-weight graph matching (in the above example it does, and in many cases it would, thanks to the properties of rating differences) because, e.g., it would still pair AvsB even if A's rating were 1000.
This is beneficial, though, because from a design perspective it prevents a clan with a runaway high rating from getting free wins and essentially getting kicked out of Clan Wars for being too good. (Since free wins would increase the rating gap, so unlike with too-low ratings, this wouldn't self-resolve.) Note that we could also prevent this runaway-rating scenario if we use a min-weight matching algorithm, by just adjusting our weight-generation logic to guarantee the highest-rated clan gets games. However, this type of problem shouldn't exist and highlights the general drawbacks of rating at a clan level rather than a player level; player ratings tend to be
much more smoothly distributed than clan ratings.
To add to what Ercole said about personal ratings and merge in a discussion from Global Chat, here's another benefit of player-level ratings:
Player-level ratings will add parity and competitiveness to Clan Wars.
Take the case of Ursus. When Ursus plays for TSFH/Harmony, TSFH/Harmony does not benefit from his skill. Ursus' wins raise TSFH's Clan War Rating, meaning that TSFH's other players now face tougher opponents and are more likely to lose.
Ursus' far-above-average skill does not translate to an elevated win-rate!But if Ursus leaves TSFH for MASTERs, Ursus
will help MASTERs. He will raise MASTERs' CW Rating, but that will have minimal effect on MASTERs' matchmaking. His wins will not be efficiently offset, and so his elevated win-rate will contribute to an elevated win-rate for the clan. Ursus (hopefully) hasn't realized this, but he has more to benefit from Clan Wars if he joins a predominately-high-skill clan because only predominately-high-skill clans can efficiently benefit from Ursus' elevated skill. This creates an anti-competitive incentive and improves poaching efficiency. In conjunction with the 40-player-limit clan handicap introduced in Update 5.12, this contributes to a net-harmful consolidation effect on the Clan ecosystem's incentive structure.
If we rated and matchmade at the player level, we would have
players at the margins with stable >50% win-rates instead of
clans at the margins with stable >50% win rates. Right now, clans wind up with stable >50% win rates (or stable <50% win rates) when they break matchmaking- i.e., when their rating is significantly higher or lower on average than the cluster of viable competitors, so matchmaking gives them games where they would expect to win >50% or <50%.
With player ratings, Ursus' wins won't turn into Strangesmell's losses! When Ursus plays for TSFH, it generally creates someone on the other side of TSFH (skill-wise) who is eating losses for Ursus' wins (or a group of such people, depending on the particulars of TSFH's skill distribution). Note that the vast majority of players would still have 50% win rates and in fact the stable 50% win-rate objective is easier to provide while matchmaking at a player level because there are vastly more players than clans and so we'd have fewer rating-gulfs. There would also be considerably greater
variety in matchmaking and reduce the number of sacrificial lamb players who keep joining Clan Wars only to lose almost all their games (because, even if their skill is average relative to the player pool, if it's significantly worse than their clans' participation-weighted average skill, CW matchmaking will keep matching them into low-parity beatdowns).
Clan-level rating/matchmaking also creates a closely-related manipulation scenario. Elite players (e.g., Python) could do what yolo swag has been accused of doing: hop on over to a low-rated clan and use their deflated rating to get easy matchmaking and an elevated win rate (until their CW rating catches up to their actual skill). Of course, this is not a good reason to keep the broken present matchmaking algorithm- if someone actually wanted to exploit CW this way, they could just choose a clan with a low rating but not as abysmally low as yolo's. And the benefit of this manipulation scenario is marginal for Python-tier players who would have an elevated win rate anyway even once their rating stabilizes. But this and several other CW manipulation scenarios only exist because of the clan-level rating/matchmaking for which Activision's stated reasoning (CW being a clan competition) is a non-sequitur.
If we're actually serious about match quality and experience, then rating and matchmaking at the player (rather than clan) level is an obvious choice.
Edited 9/17/2021 19:32:31
