Rejected Tying in battles should not effect elo and gxe at all and /offertie should be usable before turn 100

[yonmd]

Recently I've seen a replay in which two players were tied in an older gen. Instead of the more reasonable 0 change in elo and gxe the higher ladder player lost elo
and the lower ladder player won elo and gxe.

Here is a recreation on the gen 1 ladder:

another thing is offer tie, which is just really annoying as certain games just go on forever, especially older gen ones. Sometimes even though someone might win if both players are bored you should be able to /offertie when you want to. Maybe reducing the turn number to 50 as some people might try to abuse this and start spamming offertie before even getting to this state. The general consensus with a tie is that no one gains anything and no one loses anything so I feel like it also shouldn't affect gxe or elo at all when you accept the tie.

Test with offering tie in gen 1

 

[Adeleine]

Instead of the more reasonable 0 change in elo and gxe the higher ladder player lost elo and the lower ladder player won elo and gxe.

The ELO formula just does not work in this way. If you play another game using ELO ratings, like chess, draws will change your rating just like they do here.

The reason for this boils down to that ELO is a measure of skill (and these same general ideas apply to GXE as well). If you have a 1300 rating and your opponent has an 1100 rating, ELO thinks that, because you have a higher skill rating, you will perform better. People who perform better usually win, so it thinks it is more likely that you will win.

If 1300 you ties your 1100 opponent, you didn't perform better than them. You performed exactly as well as them. This means that you did worse than ELO's expectation that you'd outperform them, which means your rating must be lowered appropriately.

"You did different than expected, so your score needs to be changed" is the same idea behind rank changes in wins and losses. Pretend that 1300 you v.s. an 1100 opponent means you win an expected 70% of games. If you win the match, you've won 100% of the games, which is higher than the expected 70%, so you gain points. Winning one game is better than the "expected" 0.7 games.

Other competitive games also make ties meaningful, such as soccer and American football. The only time we'd expect to nullify a game on the ladder, to make neither side gain or lose points, is if there was a bug or glitch or whatever that prevented the game from actually being the correct game.

 

[Vileman]

To add to the previous post,
"GXE (Glicko X-Act Estimate) is an estimate of your win chance against an average ladder player. "
By that definition, why shouldn't gxe change after a tie? You are not winning against an opponent, rather you tied so "statistically" it shows you're not winning vs certain player.

 

[DaWoblefet]

I definitely do not support being able to tie before turn 100. If anything, I think it'd be preferable if the tie occurred at a later number of turns. Ties should only be used in situations where it is obvious the game is going to go to 1000 turns and both players know it. It would be an abuse to be able to tie between, say, two friends who don't want to play each other on the ladder and so offer a tie at Team Preview so that their ladder scores are only marginally affected. That's also the motivation for having GXE/ELO be affected; if there was literally no cost, it would be trivial to just offer a tie with your friends to avoid having to lose GXE/ELO. Now granted, if you wanted to abuse the system and switch for 100 turns until you could offer a tie and abuse the system that way, sure, you could, but it's at least inconvenient.

The intention of ties is only for stall mirrors that would ordinarily progress to 1000 turns and tie anyway.

 

[The Official Glyx]

The matter of ties and their purpose (namely in response to the above comment about whether it should become doable sooner or later) is something that would likely be better discussed in Policy Review, in my opinion, especially with regards to their validity in suspect tests. Until that thread is made (if it ever is), I'll just share my opinion on the former here.

I believe the ideal solution lies in either of two extremes, or one not-so extreme, being:

A: Ties (in their current implementation) are done away with completely.

B: Players are allowed to draw whenever they want.

C: Nothing is changed from the current implementation.

Regarding A, rather than necessarily allowing the potential of endless games to occur, I would suggest that /offertie be removed, and that the current 1000 turn rule instead be replaced with a more nuanced system that acts similarly to a combination of both the threefold repetition and 50 move rules in chess where battles that are resulting in no net loss of PP or other game influencing value over a considerable amount of time are instantly made into a draw, perhaps with an added warning in advance that lets players know that this will happen. Such a system would likely require more features than what I described here, though I believe the general idea is conveyed well enough.

Regarding B, this really just depends on your perspective of when and why ties should ever occur.

Regarding C, the current implementation of simultaneous timeouts resulting in a tie effectively allows players to tie whenever they want, albeit with the risk of one player betraying the trust of the other. With the functionality of ingame ties in mind, the logic is notably similar, where people could agree to a simultaneous forfeit for a draw, but then betray one another's trust to get an easy win. You could perhaps even go as far as to emulate this behavior on PS by removing /offertie and instead making forfeits be something that occurs when both players have selected something and making simultaneous forfeits result in a draw.

My personal preference leans more towards A, with simultaneous timeouts also defaulting to just coinflipping the winner instead of tying. I can still see the validity in preferring the other two options, though.

 

[DrPumpkinz]

It would be an abuse to be able to tie between, say, two friends who don't want to play each other on the ladder and so offer a tie at Team Preview so that their ladder scores are only marginally affected.

That sounds like such an unbelievably niche edge case that wouldn't even be a problem if it were common. If two friends don't want to fight from the start of the match, there's no reason they should get a slap on the wrist for getting unlucky with ladder matching.

 

[Adeleine]

Whether or not it is common enough to matter, I simply do not know at all, but...

If two players agree to manipulate the way points are handed out so that neither of them has to lose points (a rephrase of the situation you mention), there is a word for that. That word is "cheating the system". Even if it's just friends trying to be nice to each other and not trying to be manipulative. That should not be allowed.

E: The weaker player of the pair benefits by the ability to not lose points against the stronger of the pair. Being saved from having to lose points when you otherwise would lose them is comparable to gaining points.

 

[DrPumpkinz]

Neither party is gaining points either. No one benefits from drawing immediately, so any nefarious manipulation is going to amount to nothing.

 

[Tmi489]

That sounds like such an unbelievably niche edge case that wouldn't even be a problem if it were common. If two friends don't want to fight from the start of the match, there's no reason they should get a slap on the wrist for getting unlucky with ladder matching.

What if:
- In, an non OU ladder (lets say, Hackmons)...
- We're not talking about just 2 friends, but a whole group of top tier ladderers making a pact
- They always draw against each other to keep their elo (as fighting each other is the only way they lose), and abuse mid (or low) ladder to get free wins.
- The ladder becomes untoppable if you aren't in the pact, but also plain uninteresting if you are in the pact

 

[DrPumpkinz]

What if:
- In, an non OU ladder (lets say, Hackmons)...
- We're not talking about just 2 friends, but a whole group of top tier ladderers making a pact
- They always draw against each other to keep their elo (as fighting each other is the only way they lose), and abuse mid (or low) ladder to get free wins.
- The ladder becomes untoppable if you aren't in the pact, but also plain uninteresting if you are in the pact

Nothing is stopping them from doing that now besides a tiny point deduction. Most people on Showdown are more interested in actually playing Pokemon.

 

[Irpachuza]

Nothing is stopping them from doing that now besides a tiny point deduction

And, you know, having to wait 100 turns each game

 

[DrPumpkinz]

And, you know, having to wait 100 turns each game

Could always just open another game while the rigged game goes to time. The point is that worrying about a widespread conspiracy of people trying to rig Pokemon Showdown ladder placements is just absurd. Like Tmi489 said, rigging a ladder would be super boring for the people rigging the ladder.

 

[pants]

If you tie with someone of a different ELO (which implies a different skill level), that means you are closer in skill than what the ladder currently thinks, and the ladder adjusts by putting the two players’ ELOs closer together.
If someone is expected to win a game and they tie, that means they should be expected to win less games, which is reflected by lowering their rating by a few points.

 

[DaWoblefet]

Could always just open another game while the rigged game goes to time. The point is that worrying about a widespread conspiracy of people trying to rig Pokemon Showdown ladder placements is just absurd.

We actually have to ban people every so often for ladder boosting (that is, cheating on the ladder via forfeiting to a friend's accounts / their own alts to artificially increase their rank). It's not as uncommon as you might think; people really do care enough about their ranking to cheat to maintain or acquire it. We can and do retroactively punish those users by removing their accounts from the ladder if they're caught cheatng, but it's a good idea to not allow persons easier methods of cheating, whatever that might be.

I'll stress again that the point of ties is only for cases speeding up stall matches where the 1000-turn limit of Endless Battle Clause would almost certainly apply. I think Rosa put it well; any further change to the intent of the implementation of ties would need to be discussed in its own PR thread, in a similar manner to how ties were discussed before.

 

[Zarel]

I also want to add that Glicko and GXE are mathematical formulas that you can't just change to whatever you want.