two papers landed in my dms last night. who actually wins polymarket, 588 million trades. both end on the same line: about 3% of accounts take almost everything, the rest funds it.

what stuck was how they separate a skilled trader from a lucky one. lucky winners make their money on one or two events and call it edge. skilled ones make it across many, spread thin. they even put a number on it. lucky clusters around 0.64, skilled around 0.24.
i was going to write the clean version of this. numbers, a chart, the usual.
then i ran the same test on my own book.
mine is about 90% one cluster. one theme i read well at the right time. by their own number that isn't skill, it's a conviction bet that happens to be working. and most of the green is still on the screen, not in the bank.
so here's the unflattering bit. i don't have a general edge yet. i have one machine that reads one cluster well, and a recovery that lives or dies on whether that cluster resolves my way.
so i'm slowing down. a hot cluster is not a repeatable edge. weight follows where the edge is alive right now, not where it printed last month. and bolting a new domain onto the book because you're bored is how the 97% stay the 97%.
what keeps me in: the edge was never "i know this topic." it was reading how a market actually resolves against how everyone else reads the headline. someone tweets, the crowd buys yes, nobody checks whether the contract pays on the tweet or on two governments actually signing. that part travels. travels in theory and proven twice are different things though, and so far i've done it once.
the next trade is easy. the hard part is sitting flat until there's one worth taking.