Stats

How well the rating predicts, where ranked traders sit, what they do, and what none of it can tell you. Everything here is from the build that produced the current board. How much of Polymarket is in yet has its own page.

 

Does the rating predict?

Ratings are fitted on the earlier part of the history, then checked against what each trader actually did later. This is the number the project lives or dies on.

—

Out-of-sample validation did not run on this build.

The two that should sit near zero are the point: a rating that tracked how much someone bets, or how much money they moved, would be measuring activity rather than accuracy.

Where ranked traders sit

1500 is no edge over the market price. Most traders are close to it, which is what an honest rating of a liquid market looks like.

 

In this build

 

What it can't tell you

  • Predictive power is weak.
  • Independent checks.
  • Coverage.
  • Big markets.
  • Profit is counted at resolution. Charts step when a market settles, not when prices move.
  • Flags describe patterns, not people. A quoting flag means the wallet reversed sides fast and often while holding almost no net position. A pairing flag means most of its matched trades met the same counterparty. Both are measured from public trades, both can be wrong, and neither means anyone did anything improper — only that the wallet is not ranked as a forecaster. Write to us if you think a flag is wrong.
  • Wallets, not people. Names are Polymarket usernames, and one person can run several wallets.
  • Inside information looks like skill. Public trades cannot show why someone was right.
  • Open positions do not count until their markets resolve.

The full reasoning behind each rule is on the rules page.