Almost every Polymarket millionaire is a coin-flipper. Here's the autopsy.

Open Polymarket's leaderboard on any given day and the story writes itself: accounts up seven figures this month, category kings in politics, sports, crypto. The obvious move, the one every copy-trading product is built on, is to follow them.

We wanted to follow them too. So before wiring anyone's trades into our engine, we did what SPOILER does to its own record: we made the claim take an exam.

The test

Skill is not profit. Skill is beating the price. For every wallet we tested, we pulled their actual fill history from the chain and asked one question per trade: at the moment you filled, did your side beat the market's own price, consistently, across many independent events?

Then we made it hard to pass by accident. We count independent event clusters, because five bets on legs of the same match are one bet wearing five hats. We split every history in two, so a wallet had to pass on its early trades and again, out-of-sample, on its later ones, yesterday's luck cannot grade its own homework. And the pass bar is statistical, 95% confidence with multiple-testing corrections, not a vibe.

The results

First, the whole venue: 332,424 wallets, 10.2 million fills. Only 88 had enough eligible history to be judged at all. Five passed.

Then the leaderboard itself: 90,989 leaderboard rows across every timeframe and category, 31,628 unique wallets frozen and ranked, and the 429 highest-ranked accounts deep-scored over their own recent histories. Three passed. Four hundred twenty-six did not.

And inside those ranks sat the names everyone wants to copy: 81 tested accounts had made at least a million dollars. Zero of them passed. Not most failed. All of them.

The autopsy

The failures split into two clean piles. 149 wallets simply did not have enough independent betting history to be judged, their fortunes rest on too few uncorrelated events for anyone, including them, to know if skill was involved. The other 277 had plenty of history and failed the only question that matters: their fills did not beat the market's own prices. Exactly one tested account fit the market-maker profile, earning on flow rather than foresight. The rest were not businesses. They were streaks.

Nothing on the leaderboard is fake. The money is real. What is fake is the inference, because a leaderboard is a machine for finding lucky people: sort a third of a million gamblers by outcome and the top of the list fills itself with heaters. Profit is the only column, and profit is the one number that cannot tell you whether it happens again.

What passed

Eight wallets, out of everything we tested: five found by the venue-wide scan, three more from deep inside the leaderboard's ranks, mid-list names nobody screenshots, including specialists whose edge lives in a single category. Their edges are small, fractions of a point of calibration, but they are statistically undeniable, surviving both halves of their own history and every correction we could throw at them.

We are not publishing the names. The eight are ours, found by math, and they now sit on the bench of our next engine as a sharp-flow signal family, with their own pre-registered exam to pass before a single live call ever uses them. Same rulebook we apply to ourselves: nothing trades on a story, everything qualifies on evidence, and the exam is written before the answers arrive.

Until then, the leaderboard will keep minting millionaires, and we will keep grading endings in public either way. One of those is entertainment. The other is a record: spoiler.bet

Method notes and our own graded record, losses included: spoiler.bet/methodology and spoiler.bet/receipts.