Pending Make ELO Decay Less Harsh

[ABR]

So, a few days ago I was 2530 on the randbats ladder and checked back today to see I went down all the way to 2462 (I believe this was just 3 days of not playing). I knew how decay worked but this kind of reminded me how harsh it can be if you don't play (or if you play less than 6 games). As such, I would like to suggest some possible options to mitigate this. The options can be categorized as follows:

1) Increase the duration before decay kicks in - currently this is 24 hours, but could this be extended to something like 48 or even 72 hours? I feel like if the goal is to dissuade camping at a high rank, 48 or 72 hours isn't really long enough to need dissuasion.

2) Decrease the games played requirement - 6 admittedly feels excessive, I would considering it being 1-3 games needed to have 0 decay. When you're so high up the risk-reward in searching is already heavily skewed towards risk (say +7 for winning vs -30 for losing) by searching 1 game that playing And decaying feels overly punishing.

3) Decrease the amount of decay - maybe just halve it? What is currently 1 point per 100 becomes 1 per 200, and what is now 1 per 50 becomes 1 per 100.

As for which of these options I think is best, it would probably be options 1 + 2, or options 1 + 3 in tandem, all 3 together is probably unnecessary, and I would be pretty happy with any of these being implemented standalone.

Thank you for reading and I am curious to hear what other people think about it.

 

[Clas]

dont really have anything to add past affirmation, all of these would really help reduce the "necessity" of playing actively to retain elo. the biggest one there has to be the 6 games requirement, thats a bit rediculous a request especially at higher elos where games take a good 4-10min to queue and then like an average of 15-30min to finish. i do not have 3-4 hours a day to simply prevent elo decay, and this shouldnt be expected of people.

if i were to make a suggestion, though, implement the elo decay in 3 stages.

• stage 1) 1 point per 200, kicks in after 72h of less than 2 games (could be 3 games)
• stage 2) 1 point per 100, kicks in 96h later of less than 3 games (could be 4/6 games)
• stage 3) 1 point per 50, kicks in 1 week later of less than 5 games (could be 8/9/10/12 games)

this removes the ability to "camp" top ladder while still incentivising play, creating a drastic elo dropoff for those that havent played in a long time while not punishing those that simply havent played in the past could days.

 

[Zarel]

I'm currently considering making the rating period 48 hours. I think this would be better for Glicko as well as Elo.

edit: To be more specific: I would rather not change the minimum games played requirement below 6, because this is Glicko's preferred number of games per rating period. I would rather change the rating period to be whatever makes 6 the average number of games played, for someone in the top ~200. Would that be 48 hours or would that be 72 hours?

 

[ABR]

I'm currently considering making the rating period 48 hours. I think this would be better for Glicko as well as Elo.

edit: To be more specific: I would rather not change the minimum games played requirement below 6, because this is Glicko's preferred number of games per rating period. I would rather change the rating period to be whatever makes 6 the average number of games played, for someone in the top ~200. Would that be 48 hours or would that be 72 hours?

I admittedly am not sure how much the average person plays, 48 is an improvement but 72 feels preferable overall.

The glicko reasoning for not dipping below 6 is fair. Is reducing the amount of elo deducted on the table?

And thank you for the response.

 

[Zarel]

Reducing the amount of elo is a good next step after we get average 6 games per rating period, if the decay still feels too harsh. Halving rating decay is already huge, though, so unless there's a reason to rush, I'd want to get people's impressions of it before we do anything else.

Halving rating decay means someone at the top of the randbats ladder is losing around 300 Elo in a month of zero games played, keeping them still inside the top 500. This seems sane to me.

 

[Dark Pulse]

as someone who only plays 1-2 games on weeknights if I have the time, I like the idea behind all of this and support making changes that allow people to enjoy their lives without worrying about internet points, but I'd like to make sure we keep in mind that any reduction in elo decay will by definition increase the average rating of players at the top of the ladder, meaning people will have to grind for longer to reach it in the first place.

I don't have any stats or numbers on me cause I've only just read this in passing, but I think some kind of impact assessment would be worth doing
to ensure these changes don't have unintended negative consequences. 2500 is already a huge slogfest to reach (we already had plenty of people complaining about how many games were needed during the Randbats Ladder Tour) - are we happy with needing to go over 3000 to top the ladder? Are we happy that this 3000 player will still be above 2500 elo after a full month of 0 games?

At the end of the day, some equilibrium point between elo decay and elo rating will end up being reached for any system we make. So we need to make sure that said equilibrium point is something we're comfortable with being both achievable, and fair to those who've achieved it. All I ask is that we not be arbitrary with the numbers.

 

[Zarel]

Okay, Dark Pulse makes a good point. The top 500 is the same number of people no matter what. So tweaking numbers will only change who is advantaged or disadvantaged: do we favor people climbing the ladder more, or do we favor people already at the top more?

I'm suddenly very unsure. I'd appreciate more opinions on that.

 

[ABR]

I don't think 2500 would suddenly become 3000 or something like that, I think it would just be slightly easier to maintain 2500 instead of almost guaranteed losing it in a day. Particularly within ladder tournaments it can be very frustrating to have people decay daily when they play daily.

Regarding who we favor, weakened decay could also mean more fair matchmaking - if someone is at 1500 trying to climb then ideally they're paired with another "true" 1500, not someone who was 1800 but decayed down.

I do understand the concern but would like to see a change tested at minimum, the rating period extending to 48/72 feels neither too small nor too large a change to get started with.

 

[paysa]

I wouldn’t want a drastic change myself but I’d also prefer a system that favours players already at the top - my reasoning being that they are the ones who invest the most time on ladder and shouldn’t get disproportionately punished by the decay/matchmaking system (e.g. if you go 3-2 in the 2500s, you probably lose a net 20-30ELO and you get hit with decay the next day). A somewhat larger grace period lets high ladder remain populated with good players, reduced the chance of playing someone 300 ELO below them, and promotes competitiveness

I can understand why some may complain about how many games it takes to hit 2500+ but 1) the majority of players aren’t hitting 2500 or whatever number represents the Top 3/10/30 anyway and 2) using examples like RBLT or any ladder tournaments in general feels only somewhat relevant since they are not the primary drivers of ladder activity, and people who participate in those are usually willing to grind it out anyway (let alone the fact that even if you hit e.g. Top 6, you’ll likely have to keep playing throughout the week to fight decay/other grinders). Additionally, I think players that care about their ELO that much in the first place are the ones that benefit from these proposals the most, since they will likely be the ones who continuously play on high ladder to maintain their ELO, position, get games in vs. other good players, etc.

I don’t think we have to lean too heavily in favour of top ladder either to come to a favourable outcome - 48 hours between checks with reduced decay, i.e. options 1 and 3 (doesn’t need to be halved decay fwiw) would be a good middle ground in my opinion. Of course, if some napkin maths says that it’d actually do the opposite then I’d be glad to hear alternatives, but the current decay/check system is a common gripe and I echo the sentiment of wanting to see an improvement on that front.

 

[PigWarrior19]

What about a new decay benchmark? Not all ladders are equal as far as activity goes, (rands is 1k above most smogon ribbon tiers lol) this would incentivize stymied ladder players to get back into it because it’d be easier to get and maintain the rank required to get the level of games that are desired. This also circumvents the elo inflation issue I think? With more people laddering it’s to be expected to see a slight jump in the top 500s elo however that bump doesn’t seem as big as what reducing the amount of decay would result in, correct me if I’m wrong please.

a climb from 2000s-2500+ is way more enticing and less time consuming than a climb from 1500-2500+ you shave off 3-5hrs of needless laddering to get back to where you were.

Also 3 games a day or 5 in two to fight off decay completely sounds interesting, few less games or a larger grace period sounds nice.

 

[PokeChess]

Increasing the decay timer to 72 hours would be a solid improvement. Right now, the system favors those who can play every day, but ends up punishing consistent players who prefer to play in focused sessions every two or three days.

With a 72h window, the need for regular activity would still be there, but with more flexibility — which could better accommodate players who engage consistently but not necessarily daily. It also helps balance real-life routines with competitive goals, reduces the pressure of daily grinding, and brings more stability to ratings.

In the end, those aiming for the top will still need to stay active and perform well — they just won’t be penalized for short breaks.

 

[Knuckstrike]

Implementing a longer grace period, or being able to 'stockpile' played ladder games for later days, can all help with the more extreme cases of ladder decay, and I think are reasonable options. However, they fail to address the main problem which makes decay so overbearing specifically for Random Battles: the extremely high Elo ratings determining the size of the decay.

The rating system is relative, meaning only differences in ratings matter and not their absolute values: you could give everyone on the ladder 10,000 extra Elo tomorrow and it wouldn't change anything (other than decay). If we look at the ladder distributions as of writing (specifically top 500 where this matters), it's pretty clear that the differences between #1 and #10 or #100 or whatever are very similar:

Placement	Random Battles	OU	Ubers
1st	2512	2072	1813
10th	2442 (-70)	2003 (-69)	1753 (-60)
100th	2325 (-187)	1856 (-216)	1619 (-194)
500th	2167 (-345)	1707 (-365)	1512 (-301)

The main exception being whenever people clump together at 1500 cause that's where the decay stops.

However, despite their almost identical Elo differences near the top, their decays vary wildly: top 10 in Ubers decays by 7 points per day, OU by 12, and Randbats by 21.
The main purpose of ladder decay is to prevent users from being able to 'sit' on a high ladder rating without playing, all without draining Elo from the lower rated players. Because the differences are near identical between the most common ladders, Randbats players fall through the rankings 3x quicker than almost all other ladders, with OU being about 2x as bad. This goes far beyond the purpose of decay. In my opinion it doesn't make sense to punish players more severely when their preferred top ladder is at a much higher rating than other ladders.

The simple solution I would suggest is to implement a cap on daily ladder decay. For example, you can make everyone above 1900 lose the same for an even 10 Elo a day of not playing, or set the cap to a round 2000, it doesn't really matter. Either way this would equalize how quickly top players fall through the rankings when they are inactive for all ladders where decay is relevant.
Additionally, because 'most' users have a lower rating and are below the cap anyway, this will not affect the ladder decay for those users, making ladder inflation less of a risk than with a measure that is more generous with decay for everyone.

Of course this can be done alongside any of the other systems.