Aurisle
Member
- Joined
- Nov 21, 2025
- Messages
- 5
- Points
- 1
Minecraft username:
Suggestion:
Detailed description:
Reason(s):
Example(s):
Aurisle
Suggestion:
ToxicityCoin Penalty & GG Reward System
Detailed description:
TL;DR: Lose coins for toxic chat during a match (loser, noob, L, etc.), earn coins for saying GG/good game after a match ends. Detection runs in layers so it stays cheap and fast even at server scale, while still being fair to players who get caught up in false positives.
Right now there's nothing actively discouraging toxic chat in matches or rewarding good sportsmanship. This suggestion proposes a coin-based system: toxic messages during a game cost the sender 100-150 coins, and saying something like "gg" or "good game" after the match ends earns a small coin reward. The goal isn't to be a strict chat filter, it's to nudge behavior using the coin economy that's already part of the server.
I work as a developer myself, so this isn't just a wishlist idea, I've tried to think through how it would actually be implemented without tanking server performance, and I'd genuinely like to discuss the technical side with the team if there's interest.
The part that needs the most thought is how messages get evaluated, since checking every message with a full AI model isn't realistic at scale. Here's a layered approach that keeps things fast without sacrificing fairness.
A normalized blacklist of toxic terms (handling spacing, leetspeak like "n00b" or "l0ser", etc.) gets checked against every message as it's sent. This is fast enough to run in real time with no noticeable load, even across many concurrent matches.
The key improvement over a basic word filter is that a single flagged word alone doesn't trigger a deduction. The system scores the message: a word like "easy" or "noob" used alone is low-confidence, but combined with "you," a player's name, or another flagged word in the same message, the score goes up. Coins only get deducted once the score crosses a threshold. This catches obvious insults ("you're such a noob") while letting harmless phrasing ("that drop was easy") slide.
The first flagged message in a match gives a warning instead of an immediate deduction. A second offense in the same match, or a pattern of offenses across recent matches, triggers the actual coin loss, scaling up for repeat offenders. This avoids punishing a one-off slip the same as someone who's clearly being toxic on purpose.
For genuinely ambiguous words like "easy," which are toxic in some contexts and completely normal in others, messages that don't clearly resolve in Stage 1 get queued to a small, locally-run text classifier (not a full AI model like an LLM, just a lightweight model trained specifically on this kind of slang). This runs asynchronously after the message is sent, so it never blocks or slows down gameplay. If it confirms the message was toxic, the coin deduction is applied retroactively, usually within a few seconds.
Every flagged message and its resulting coin deduction gets logged to a database along with the context of the message. The same classifier used in Stage 2 periodically re-reviews this log of flagged cases to check for false positives. If it determines a flagged message wasn't actually toxic, the deducted coins are automatically refunded, no staff intervention needed for the obvious cases.
If a player still feels their deduction was unfair after this automatic review, they can post a report on the forums for a human/staff look. This keeps the system mostly self-correcting while still leaving a path for genuine disputes.
After a match ends, players who send a message like "gg" or "good game" within a short window get a small coin reward. To prevent farming this, it would only trigger once per player per match, and might require slightly more than a single bare letter to count.
A single Bedwars match only has a handful of players sending messages a few seconds apart at most, so even across the whole server, real-time chat volume is far lower than it might sound. The word-list and scoring stage handles the vast majority of messages instantly and cheaply. Only the genuinely ambiguous fraction ever reaches the lightweight classifier, and that runs asynchronously in the background, so there's no need for a full AI model checking every message live.
Open to feedback from staff and the community on the coin amounts, the word list itself, and how strict the escalation should be.
The Idea:
Right now there's nothing actively discouraging toxic chat in matches or rewarding good sportsmanship. This suggestion proposes a coin-based system: toxic messages during a game cost the sender 100-150 coins, and saying something like "gg" or "good game" after the match ends earns a small coin reward. The goal isn't to be a strict chat filter, it's to nudge behavior using the coin economy that's already part of the server.
I work as a developer myself, so this isn't just a wishlist idea, I've tried to think through how it would actually be implemented without tanking server performance, and I'd genuinely like to discuss the technical side with the team if there's interest.
The part that needs the most thought is how messages get evaluated, since checking every message with a full AI model isn't realistic at scale. Here's a layered approach that keeps things fast without sacrificing fairness.
How Detection Would Work:
Stage 1: Word-list + context scoring (instant, real-time)
A normalized blacklist of toxic terms (handling spacing, leetspeak like "n00b" or "l0ser", etc.) gets checked against every message as it's sent. This is fast enough to run in real time with no noticeable load, even across many concurrent matches.
The key improvement over a basic word filter is that a single flagged word alone doesn't trigger a deduction. The system scores the message: a word like "easy" or "noob" used alone is low-confidence, but combined with "you," a player's name, or another flagged word in the same message, the score goes up. Coins only get deducted once the score crosses a threshold. This catches obvious insults ("you're such a noob") while letting harmless phrasing ("that drop was easy") slide.
Escalating penalties
The first flagged message in a match gives a warning instead of an immediate deduction. A second offense in the same match, or a pattern of offenses across recent matches, triggers the actual coin loss, scaling up for repeat offenders. This avoids punishing a one-off slip the same as someone who's clearly being toxic on purpose.
Stage 2: Lightweight classifier for ambiguous cases (async, not real-time AI)
For genuinely ambiguous words like "easy," which are toxic in some contexts and completely normal in others, messages that don't clearly resolve in Stage 1 get queued to a small, locally-run text classifier (not a full AI model like an LLM, just a lightweight model trained specifically on this kind of slang). This runs asynchronously after the message is sent, so it never blocks or slows down gameplay. If it confirms the message was toxic, the coin deduction is applied retroactively, usually within a few seconds.
Appeal and fairness layer
Every flagged message and its resulting coin deduction gets logged to a database along with the context of the message. The same classifier used in Stage 2 periodically re-reviews this log of flagged cases to check for false positives. If it determines a flagged message wasn't actually toxic, the deducted coins are automatically refunded, no staff intervention needed for the obvious cases.
If a player still feels their deduction was unfair after this automatic review, they can post a report on the forums for a human/staff look. This keeps the system mostly self-correcting while still leaving a path for genuine disputes.
The GG Reward
After a match ends, players who send a message like "gg" or "good game" within a short window get a small coin reward. To prevent farming this, it would only trigger once per player per match, and might require slightly more than a single bare letter to count.
Why This Is Feasible
A single Bedwars match only has a handful of players sending messages a few seconds apart at most, so even across the whole server, real-time chat volume is far lower than it might sound. The word-list and scoring stage handles the vast majority of messages instantly and cheaply. Only the genuinely ambiguous fraction ever reaches the lightweight classifier, and that runs asynchronously in the background, so there's no need for a full AI model checking every message live.
Open to feedback from staff and the community on the coin amounts, the word list itself, and how strict the escalation should be.
Reason(s):
Toxicity in competitive modes like Bedwars is one of the biggest reasons newer or casual players bounce off a server early, especially when it comes from teammates rather than opponents. Right now, dealing with it relies almost entirely on manual reports and staff reviewing logs after the fact, which doesn't scale well and rarely feels timely to the person being affected.
Tying this to coins instead of mutes or warnings works with the existing economy, so it functions as both a deterrent and a light nudge rather than feeling like a punishment system bolted on from outside. Rewarding GG also balances the system out, it isn't just about punishing negative behavior, it reinforces the positive behavior the server presumably wants more of anyway.
Tying this to coins instead of mutes or warnings works with the existing economy, so it functions as both a deterrent and a light nudge rather than feeling like a punishment system bolted on from outside. Rewarding GG also balances the system out, it isn't just about punishing negative behavior, it reinforces the positive behavior the server presumably wants more of anyway.
Example(s):
Clear-cut toxic message: Player A loses a match and types "ez game noob" to the winning team. The word-list flags "ez" and "noob" together, the context score comes in high, and since it's their first flag they get a warning. If they do it, the 100-150 coin deduction kicks in.
Ambiguous message resolved without penalty: Player B types "ngl that bridge was easy" about their own gameplay. Stage 1 sees "easy" but no targeting language like "you" or a player name, so the score stays low and it gets queued to the Stage 2 classifier instead of an instant deduction. The classifier asynchoronously processes it as self-referential, not directed at anyone, and no coins are deducted.
False positive caught and refunded automatically: Player C types "I'm not a loser, stop" while defending themselves in an argument. The word-list flags "loser" and an initial deduction is applied. During the periodic review pass, the classifier re-reads the message in context, recognizes the negation, and automatically refunds the deducted coins, logged in the database for transparency.
Repeat offender escalation: Player D gets a warning for toxic chat in one match. The next day, they send another flagged message in a new match. The system checks their recent history, sees the pattern, and applies a larger deduction than a first-time offender would get for the same message.
GG reward in action: A match ends and Player E types "gg" within the post-match window. They get a small coin reward once for that match, even if they spam "gg" multiple times in the chat afterward.
Ambiguous message resolved without penalty: Player B types "ngl that bridge was easy" about their own gameplay. Stage 1 sees "easy" but no targeting language like "you" or a player name, so the score stays low and it gets queued to the Stage 2 classifier instead of an instant deduction. The classifier asynchoronously processes it as self-referential, not directed at anyone, and no coins are deducted.
False positive caught and refunded automatically: Player C types "I'm not a loser, stop" while defending themselves in an argument. The word-list flags "loser" and an initial deduction is applied. During the periodic review pass, the classifier re-reads the message in context, recognizes the negation, and automatically refunds the deducted coins, logged in the database for transparency.
Repeat offender escalation: Player D gets a warning for toxic chat in one match. The next day, they send another flagged message in a new match. The system checks their recent history, sees the pattern, and applies a larger deduction than a first-time offender would get for the same message.
GG reward in action: A match ends and Player E types "gg" within the post-match window. They get a small coin reward once for that match, even if they spam "gg" multiple times in the chat afterward.
