I would like to raise some critical issues regarding the current bot system on OGS that make it extremely unfair and difficult for bot operators to maintain their bots properly. There are three main problems that need urgent attention:
Bot Visibility in the ‘Play Computer’ Screen
When a bot sets its handicap games to “Unranked” to avoid rating distortion, its name is grayed out on the ‘Play Computer’ screen. Users cannot click to challenge it directly and must manually search for the bot’s ID. However, if the bot’s name is too long or contains Chinese/special characters, the name gets cut off on the UI. This makes it impossible for users to search for the bot, causing the bot to receive zero game requests from new users.
Exploitation of Ranked Handicap Games
Conversely, if a bot enables “Ranked” for handicap games to stay visible, it becomes a target for malicious users. Some players intentionally lower their accounts to a two-digit kyu level (sandbagging) and challenge the bot with a 9-stone handicap. This causes the bot’s rating to plummet instantly (e.g., the bot ‘12bturbosai’ dropped all the way to 9k due to this issue).
Flawed Rating Calculation for Bots (Excluding Resignations/Timeouts)
Following the recent system update about a month ago, games won by bots via resignation or timeout are excluded from the rating samples. Bots naturally win most of their games when the opponent resigns after falling behind. Because of this rule, only the games where the bot loses are being accumulated in the rating samples. This structural flaw forces the bot’s rating to continuously drop.
Suggestions for improvement:
Allow bots operating in unranked match mode to be clickable/challenging directly from the ‘Play Computer’ screen, or fix the UI text cutting off long names.
Revise the bot rating system so that wins by resignation or timeout are properly counted, or implement a protection system against extreme handicap sandbagging.
[Simple Solution for the Handicap Abuse]
An easy way to fix this is to automatically make any handicap game with 6 or more stones “Unranked” by default. If the system forces 6+ stone handicap games to be unranked, bot operators can safely leave their bots as “Ranked”. Within 5 stones, bots can easily handle the game and maintain a fair rating, completely preventing the 9-stone sandbagging abuse.
Thank you for reading, and I hope the development team can look into these bot-related issues soon!
I don’t think this is the cause. There was an update to how ratings for bots are calculated, however this resulted in a rating spike for some bots:
Anoek said he’d update the system to try and fix it:
If you look at the rating graph for 12bTurboSai you’ll see that there was a sudden spike in its rating followed by a sudden drop. The date of Anoek’s post appears to line up reasonably well with the apex of the spike.
Regarding the announcement you mentioned, I am well aware that recent system updates mean games against bots no longer affect a human player’s rating. However, that only applies to humans. To calculate the bot’s own rating, the system still inevitably pulls data and samples from its matches against human players. Therefore, the bot’s rating is still highly active and affected by these games.
You are correct that Anoek’s recent system update caused a sudden rating spike and initial drop for many bots. However, the core issue I am raising is the structural flaw that drives the rating all the way down to 9k and keeps it there. Even if the initial drop was triggered by the system fluctuation, the combination of two specific factors makes it impossible for the bot to recover:
9-Stone Handicap Abuse: Sandbagging users are intentionally taking advantage of the system by forcing 9-stone games on the bot, causing massive rating losses that shouldn’t happen in a fair environment.
No Rating for Resignations: Every single game where a human wins against a bot is added to the bot’s rating samples. However, for the bot, only games won by counting (score estimation) are included. Since human players rarely play out a game until counting when they are losing to a bot, it is practically impossible for the bot to accumulate winning points.
So while the system update may have started the instability, the 9-stone handicap abuse combined with the fact that the bot only accumulates losses is what is actively destroying the bot’s rating.
While games won by a bot against a human player by resignations are excluded from the rating samples, games between two bots (bot-vs-bot ranked matches) are still fully reflected in the rating regardless of how they end.
Some might argue that a bot can just recover its rating by a ranked match with other bots. However, because a bot’s rating has been artificially dragged down to very low rank due to sandbagging, it becomes a “rating black hole.” When this underrated bot plays against other normal bots, it unfairly drains their points and drags down their ratings too.
Ultimately, this flaw doesn’t just affect one bot; it creates a domino effect that ruins the entire bot rating ecosystem on OGS.
Also the settings bot operators set are mostly ignored in the game creation screen. For example, my bots allow 5x5 games, however, when the users set their custom settings to 5x5, the interface shows:
