Let users flag toxic users. Use the resulting toxicity score to filter out messages based on each reader’s toxicity tolerance level. Both the toxicity score and the tolerance level are driven by user flagging, with score decay over time to account for improved behavior.