Alpha 12

[TomCat]

I think it's so nice that we do not have to wait for a match with this new system, but this style of game is not that satisfying. At least for those who understand how it works. For example, I defeated Mcompany yesterday. But I can not have the right satisfaction because he does not know about it. And if he does not look at MY statistics, he will not even know. And in any case, he will not care. Because I actually did not defeat him, but some of his who knows how much outdated ghost. And I do not want fighting ghosts. I'd rather fight people. With the knowledge that the game I'm watching is at the same time an opponent watching with me. Or he/she will see it later. But in any case, he/she will see it. And if I win, my opponent will pay me his points. The opponent and not some of his ghosts.

So how about dividing this RANKED into two parts? One with the actual matchmaking for people who don't want to wait. And one with some queuing for matches with real players only.

[GFX47]

Sorry for the late reply, I've been pretty busy lately...

1. We always join on the right side. Considering the fact that LR was bugged to where the left side always won the middle resource, I think it is a valid worry that there might be some imbalances with always joining the same side

This should be fixed.

2. No matches that I didn't volunteer for are shown to me. However, I think it is very important that I be able to see the latest loses I have, no matter if I won or loss

Matches may use old versions of you AIs.
It could be misleading and incite players to make unnecessary fixes/tweaks.

3. Furthermore, if I deploy an AI to test a trick against specifically those in my league, this forces me to either used unranked or private matches to do so, lest I want to affect the elo of everyone else

It was already the case with the previous system, right?

4. Speaking of the elo, I feel like such a system is way more prone to hyperinflation or hyperdeflation in the scores for any active player which constantly wind or constantly loses. At least in the auto deploy method, if you lost, you didn't give your opponent a million opportunities to get to 2000 for free

It will not be a pure elo system anymore, indeed, but it's the best solution to guarantee a constant activity for everyone.

The goal of the *ranking* system is more about evaluating your level relatively to other players, than finding a precise score value.

Also don't forget that the "ghost" AIs are impacted by win/loss and their elo score will vary depending on their relative strength. So the score of a test AI should rapidly decrease and not be matched against players of your rank anymore.

And I'll talk again later about seasonal resets to fight this inflation/deflation problem.

5. For anyone with multiple accounts, this makes it possible to run into each other even if they try to take all precautions to avoid each other

Players should not use multiple accounts anyway.

Imo, I feel like this system will result in more games, but otherwise be worse than auto queue

What's important here is that everyone can have fun with the game and no one looses points for "involuntary" matches.

[GFX47]

Yeah, I think I got it. The only thing is we're not really talking about an ELO rating anymore if the score is only adjusted for one player, for some matches, but I'm not sure that's super important.

See my previous answer to mcompany

I do see now that the number of matches has gone up dramatically, which to me trumps pretty much everything else.

+1

[edit] One question I have: if I queue a deployment, and then another one, will other players automatically only get matched against my newest deploy, or will both deploys be out there? I'm also not entirely clear how the 10 deployments limit and pushing several deploys of the exact same AI work in this brave new world.

Players will be matched to the newest deployment that matches the matchmaking rules (elo range, recent matches against you, etc).
And I don't think I can make another sentence with more "matches" occurrences than this one.

[GFX47]

Anyway in this way the elo will be creates out of thin air and this will likely cause the veterans to accumulate more and more of it. This may disrupt easily the 400 points of distance for matching other players. Have you given a thought on this?

See my previous answers

[GFX47]

So how about dividing this RANKED into two parts? One with the actual matchmaking for people who don't want to wait. And one with some queuing for matches with real players only.

Let's not divide something that is already not so big ^^

[mcompany]

1. We always join on the right side. Considering the fact that LR was bugged to where the left side always won the middle resource, I think it is a valid worry that there might be some imbalances with always joining the same side

This should be fixed.

I'm partially taking a break from the game because I'm too lazy to actually fix my AIs, but I am curious whether or not LR had a bug, or if the issue of one side always winning a middle resource still is prevalent (though less noticing). I mean, imo, always starting on one side isn't a massive issue, mainly just the worry that all sides might not be equal

2. No matches that I didn't volunteer for are shown to me. However, I think it is very important that I be able to see the latest loses I have, no matter if I won or loss

Matches may use old versions of you AIs.
It could be misleading and incite players to make unnecessary fixes/tweaks.

On the flip side, I don't change my AIs much, and any loss, regardless of the age of the AI, is still very important to track in case there's a bug within both AIs that is only noticeable with the old one or in case there's a player I haven't played against that found an exploit or in case I'm not playing against the updated version of my opponents. Sure, the AI is old. Sure it might be confusing for some. But for me or Chris, such losses are still necessary to keep track of

3. Furthermore, if I deploy an AI to test a trick against specifically those in my league, this forces me to either use unranked or private matches to do so, lest I want to affect the elo of everyone else

It was already the case with the previous system, right?

Not to anywhere near as high as a degree and on the previous system it could be limited further by choosing not to auto queue or purposely losing/drawing. On the current system, unless you stay 100% active and try to override old deployments as much as possible, it will affect the score for days to come

4. Speaking of the elo, I feel like such a system is way more prone to hyperinflation or hyperdeflation in the scores for any active player which constantly wind or constantly loses. At least in the auto deploy method, if you lost, you didn't give your opponent a million opportunities to get to 2000 for free

It will not be a pure elo system anymore, indeed, but it's the best solution to guarantee a constant activity for everyone.

The goal of the *ranking* system is more about evaluating your level relatively to other players, than finding a precise score value.

Also don't forget that the "ghost" AIs are impacted by win/loss and their elo score will vary depending on their relative strength. So the score of a test AI should rapidly decrease and not be matched against players of your rank anymore.

And I'll talk again later about seasonal resets to fight this inflation/deflation problem.

First of all, considering that a test AI from a RC version lasted a week to be matched against a top 5 player when I gladiabots more recent examples, I have my doubts that this system is working properly right now. Also, I do not believe that anyone has found the current system to be changing the elo of the ghost setups

5. For anyone with multiple accounts, this makes it possible to run into each other even if they try to take all precautions to avoid each other

Players should not use multiple accounts anyway.

And yet, some players still do for a variety of different reasons

[Ed-209]

I was using second acount on pc to test ideas against myself and occasionally try the new changes against others. Haven't in ages.

[mumpsimus]

...

Also don't forget that the "ghost" AIs are impacted by win/loss and their elo score will vary depending on their relative strength. So the score of a test AI should rapidly decrease and not be matched against players of your rank anymore.

I'm pretty convinced this part isn't working: I've been playing against the same AIs over and over for people who haven't played in a week or more, and the AIs always have the exact same score. See:

https://stats.gladiabots.com/player?&na ... ay=matches

and look for matches against MrChris (always at 1995) or McCompany (always at 1903).

[pier4r]

Gfx if the ghosts expire after some days (say 10. 10 days are a lot) and they change also their score with losses and wins (in telegram the reports say that this is not true); it is true that new score is created in the system (the ghosts don't start with 1000 but with something else) , but the inflation (as there are way more people producing ghosts with score 1000+) will be barely noticeable over time. Maybe after some months one will see inflated scores but nothing that we didn't have before. So the elo will still work somehow.

If the ghosts stay forever and do not change their score , then the effects may appear faster.

I really dislike autocorrect and touchscreen keyboards.

[GFX47]

OK let me investigate this bug.

[GFX47]

It should work as expected now.
Could you please confirm?

[GFX47]

1. What team setups are saved to be deployed against other players? Is it possible to change what AIs are setup for any given player, and is it possible to continue fight a heavily outdated AI like in my case?

All setups are saved.
Player can't change it.
Players with lower ranking could me matched against your old AIs if they have similar elo scores.

2. Why is it setup this way? I know in the chat that you say that the system is to allow players to play against AIs in their elo range, however (especially when deciding to match specifically against a losing AI) there is always a chance that there is an accidental deployment, a test AI, or a demotion AI, which has no baring on the elo of the player at the time of deployment. Likewise, if all of the elo ranges changes meaning (like they will start to do under this system), an AI over a week old is almost guaranteed to not match the rest of the elo range by being so old and being tested against so often.

As said earlier, and when there's no bug ^^', the "ghost" AIs have their elo score updated when winning or loosing.
So test/bugged AIs should go down pretty quickly and not be matched against players of your rank.

3. I the case that someone does play through all of the matches that someone makes manually deployments, Are the manual deployments prioritized for everyone over auto generated matches as an alternative towards the system to avoid this mess and actually see my losses against a recent AI

Yes, recent manual deployments are selected in priority during matchmaking.

[GFX47]

I'm partially taking a break from the game because I'm too lazy to actually fix my AIs, but I am curious whether or not LR had a bug, or if the issue of one side always winning a middle resource still is prevalent (though less noticing). I mean, imo, always starting on one side isn't a massive issue, mainly just the worry that all sides might not be equal

Please tell me if you notice any new bug like this.

On the flip side, I don't change my AIs much, and any loss, regardless of the age of the AI, is still very important to track in case there's a bug within both AIs that is only noticeable with the old one or in case there's a player I haven't played against that found an exploit or in case I'm not playing against the updated version of my opponents. Sure, the AI is old. Sure it might be confusing for some. But for me or Chris, such losses are still necessary to keep track of

Getting more data requires you to play more in a sense.
It's kinda how it previously worked, isn't it?

Not to anywhere near as high as a degree and on the previous system it could be limited further by choosing not to auto queue or purposely losing/drawing. On the current system, unless you stay 100% active and try to override old deployments as much as possible, it will affect the score for days to come

That's a small disadvantage compared to the benefits of the new system in my opinion.

First of all, considering that a test AI from a RC version lasted a week to be matched against a top 5 player when I gladiabots more recent examples, I have my doubts that this system is working properly right now. Also, I do not believe that anyone has found the current system to be changing the elo of the ghost setups

It should work better now.

And yet, some players still do for a variety of different reasons

Do you see any good ones?

[mumpsimus]

It should work as expected now.
Could you please confirm?

I see ghost AIs' scores changing now, so yes, I think it's fixed, thanks!

[mumpsimus]

Think I have found a bug:

closest to an ally.PNG (1.17 MiB) Viewed 1968 times

This base is not closest to an ally by a long way, by any calculation I can come up with. (I verified by clicking on each of the allies, and for none of them is the base on the right closer than the base on the left.)

Replay ID: 870658

[GFX47]

It should be fixed in 12.1.

[mcompany]

On the flip side, I don't change my AIs much, and any loss, regardless of the age of the AI, is still very important to track in case there's a bug within both AIs that is only noticeable with the old one or in case there's a player I haven't played against that found an exploit or in case I'm not playing against the updated version of my opponents. Sure, the AI is old. Sure it might be confusing for some. But for me or Chris, such losses are still necessary to keep track of

Getting more data requires you to play more in a sense.
It's kinda how it previously worked, isn't it?

Before auto queue, I guess. After autoqueue, not really. With auto autoqueue, I could just set up a few matches, wait a day, and have well over 10 matches with quite a bit of data of how certain things work and what works, and what doesn't.

Not to anywhere near as high as a degree and on the previous system it could be limited further by choosing not to auto queue or purposely losing/drawing. On the current system, unless you stay 100% active and try to override old deployments as much as possible, it will affect the score for days to come

That's a small disadvantage compared to the benefits of the new system in my opinion.

Me and you have a different opinion of things. I mean, at most on the new system, people probably get to still play against me despite the fact that I quit over a week ago due to feature creep. For all of the players playing against me, either they win against the AI and it isn't even worth watching the large number of repeat matches because it is the same thing with no change from "me" and no need to change from them. for anyone who does lose, there really is no real difference, except that you lose more often.

And yet, some players still do for a variety of different reasons

Do you see any good ones?

I know there has been multiple times where a player has wanted to try a different idea without affecting their rank too much. The main caes of this I know of is Nullpointer switching to T-800 to test how well generic AIs worked, Me switching to theletter7 to try hyper aggressive AIs, Miojo switching to "Good bot" to test a low node AI, and Thekidpunisher switching to "TKP" to try a full AI with all bot classes. In all examples, no player played both accounts at the same time, and only Null permanently switched accounts with everyone else fulling switching accounts. I also know that with the level of feature creep that this new version brings, there is a case to be made that it is better to just start over back at 1000 and raise back up with a new AI. I'm pretty sure I'm one of the only players to deploy a blank AI to demote myself, and I can tell you, it is an unbelievably slow process. It is better and faster just to make a new account to do so

[mcompany]

2. Why is it setup this way? I know in the chat that you say that the system is to allow players to play against AIs in their elo range, however (especially when deciding to match specifically against a losing AI) there is always a chance that there is an accidental deployment, a test AI, or a demotion AI, which has no baring on the elo of the player at the time of deployment. Likewise, if all of the elo ranges changes meaning (like they will start to do under this system), an AI over a week old is almost guaranteed to not match the rest of the elo range by being so old and being tested against so often.

As said earlier, and when there's no bug ^^', the "ghost" AIs have their elo score updated when winning or loosing.
So test/bugged AIs should go down pretty quickly and not be matched against players of your rank.

In the context of when I said that, and when you replied, this wasn't true

[mumpsimus]

Still something not quite right about the ghost ELO:

This player only has a single game:

https://stats.gladiabots.com/player?&na ... ay=matches

and their AI is 4 assaults that flee from everything, so they should have pretty much bottomed out, but I keep seeing the AI lose ELO points and then reappear with a higher score. If multiple matches had been played I could explain this as multiple ghosts (though even then I'd expect the ghost to rapidly descend below 1000 ELO), but with one AI only and it never winning or even tying any matches, it should not be regularly jumping up by as much as 40 points.

[edit] if you look at https://stats.gladiabots.com/?&league=0 ... vePlayers= you see the ghost AI lose match after match, never winning any, yet occasionally jumping up in score.

[pier4r]

that's a great source of new elo. I presume GFX will fix it soon. Nice report!