An update to this point, I am actually one game away from winning all the games in my bracket and that final game is one where I am massively ahead. I guess that it further proof that using the “general rank” to calculate the brackets, instead of the appropriate sub-rank for the tournament’s time setting is not optimal.
A similar thing has happened to a lot of the tables of the tournament, by the way, where there is a player in the bracket that has won all or almost all games. The most extreme cases are these though:
The players in the brackets are all DDKs, except the two top ones that got all those wins and rose to 9k. However, their “correspondence rank” is 6.9k and 8.2k, respectively.
The 6.9k fellow shouldn’t have been playing versus DDK players and the 8.2k fellow had no business being placed in a bracket that had even a couple of 25k opponents in it.
It is not their fault this happened. This is the bracket they got, they played and they won (as was expected), but that certainly doesn’t help the balancing of the server rankings.
That also makes sense. I have never made a tournament myself, so I have never looked into on how those options work and I didn’t notice that setting, at all.
(Edit: there was a post 66 above mine, to which I was replying, which seems to be gone)
I would like to remind everyone that I have mentioned similar issues. To fix this I’ve created a second account for correspondence games only and I only use my first account for live games from then on. The differences between my rankings in different time settings is too big to be of any use in either finding a suitable opponent for a game or, as I see now in this case, entering a tournament. I appreciate all the effort that has been put in to create a balanced rank but to me it seems the result has serious flaws in practical, day to day use. But it appears to be working fine for the majority of players on OGS
Yeah I’ll have to look into it more, before I can really blame it properly. Like some of the top groups do look like they roughly group players by similar rank but way down there’s 10k and 23k in the same group regularly, so I’m not sure what that’s about.
I was chatting with Pianotorious on this topic, who pointed out that the ranks of bots they play appear to have declined.
They have played as one example doge_bot_4 over the last year, and (by filtering for just that bot) you can see in their game history the steady decline of that bot’s rank from 5-8k in 2024 to 10-12k now.
Similar with noob_bot_2: 8-9k in 2024, 11-13k now.
I know that bot-ranks are notoriously variable, but these do seem to be long-term trends…
Do these bots offer handicap games? If so, then maybe people take advantage of them like that.
The larger the handicap, the faster the game goes out of the “optimal scope” of the bot, so it is much easier to win, than the ranks would suggest. And this might lead to situations where the bot is “8k” in gameplay, but ranked at 12k in OGS, so when a 12k player goes to play a fair game against them, they get a bot that is 4 ranks higher.
Consequently, if an 8k player goes against the bot and has to give handicap to the bot, they are probably never going to win, either.
There is a potential problem there though – What I’ve found with some of the weaker bots, is that some of them play much better when they’re behind, so one gets the opposite effect: the handicap actually gets cancelled out a bit by the bot going into what feels like superboss rage-mode.
I am a high SDK player at best though (8k on Fox for ex), so I don’t entirely trust my own subjective experience here. It might just be deficiencies in my own play. But it does seem plausible, in the sense that local fighting is relatively “easy” for an engine to calculate, right? So if it decides to pull no punches when invading/reducing because it’s way behind overall…
Perhaps it will be useful if I share some example handicap games I’ve played against the currently high/mid-SDK strength bots.
(A loss, very typically what happens is that I get into a high stakes situation, lose a big group, and then the bot doesn’t seem to make any mistakes that might allow me to catch up): Pianotorious vs. Deutzia
One where I actually manage to hang on to the handicap advantage throughout:: Pianotorious vs. Deutzia
In this one – going off the AI analysis – the 6k bot never makes any significant mistakes, and needless to say I do: Pianotorious vs. Echinops
Here’s a non-handicap one where Kata_web, ranked 10k at the time (!) played near-perfect against me from start to finish: Freundschaftsspiel
It makes me wonder if, perhaps, there’s a pool of players who have learned – intentionally or not – how to manipulate these bots into situations where they play poorly? I’m guessing this is a known thing.
Edit, to add another point that’s been on my mind: Assuming bots don’t actually change their playing strength over time (ex. by storing game history or refining based on that?), I naively wish they would just have an established rating that didn’t change over time. Of course that wouldn’t work because then a slightly stronger player could “farm” win off of them almost indefinitely… Could players have an individually calculated RD against each bot they play, or something? That seems like the kind of thing that would just make the database explode in size, but maybe someone cleverer than me knows a way to accomplish it…
IMO, when handicap is larger than 3, bot games should be unranked. Most bots have been trained with even games. Low handicap games are not very different from even games, but bot may make unexpected moves in high handicap games.
