Each day, players try to spot the real definition of an obscure word among the fakes submitted by other players the day before. Fake definitions rank on a daily leaderboard.
I love the traditional version of the game, but it's not so easy to get a session going. An asynchronous, global format is my attempt to make it more casual and accessible.
It's live for about a month now: ~100 daily players, 5000 total, some 20 people on 10-day streaks.
I spent way too much time trying to automate moderation (grouping related submissions, judging similarity, removing offensive submissions) using sentence-transformers and w2v. Finally gave up and settled on using deepseek to moderate for 0.002€ a day.
One thing I'm currently working on is improving the exploration vs exploitation balance in matchmaking (selecting the fakes to present to a player). I want sessions to contain good fakes, but every fake must be played a few times to judge its quality. As a first step, I'm replacing the ELO system for definitions with one that accounts for uncertainty (like Trueskill)