What is a good win rate on Polymarket?

The distribution, by evidence depth

Method in one line: each wallet’s resolved positions are graded against the price actually paid, only positions held to resolution on one side count as forecasts, correlated markets in one event family count once, and a record lands in the ahead or behind column when its excess wins clear the 95% interval chance produces at those prices, in that direction. Full method on the methodology page.

The median wallet loses to its own prices

In the 20 to 99 event tier, the median wallet wins 60.9% of its events while its entry prices implied 61.4%. Measured wallet by wallet, the median record runs 0.9 wins per hundred short of what its own prices predicted, and only 45% of these wallets beat their prices at all. The significance test runs in both directions, and the split is lopsided: 1.3% of these records are distinguishable from their own prices and ahead of them, while 12.1% are distinguishable and behind. For every thin record that provably beat its prices, about nine provably trailed them. The middle tier fares worse still: at 100 to 999 events the median record runs 2.2 wins per hundred behind its prices, and 32% beat them. The 2026 SSRN study “Who Wins and Who Loses in Prediction Markets” found the same shape from account-level data: the median participant loses a little, and a small minority drives the accuracy. Our wallet-level data shows the same shape independently.

A perfect win rate can mean almost nothing

The 90th percentile win rate among wallets with 1,000 or more graded events is 99.9%. About one in ten of the deepest graded records has lost about one event in a thousand held to resolution. A record that close to perfect over that many events in practice only happens at extreme favorite prices, where winning was already close to certain, and the margin left over those prices is small: across the whole tier the median excess is 0.8 wins per hundred. Meanwhile the folk thresholds you will find elsewhere, 60% or 65% win rate as a mark of skill, sit comfortably inside what chance plus favorable prices produces. If someone quotes a win rate at you without the prices behind it, they have told you nothing yet.

Evidence depth separates provable skill from noise

The ahead shares climb from 1.3% at 20 to 99 events, to 4.1% at 100 to 999, to 42.8% at 1,000 or more. Deep records make real edges provable and lucky streaks impossible to sustain. This is also why the leaderboard gates on independent resolved events: a short hot run, however profitable, cannot carry the evidence a ranking claim needs.

We retracted the July 2026 versions of these shares. The earlier figures, 6%, 27%, and 45%, were computed as the ahead tail alone, although the wording around them did not say so. They also came from a test that counted events where a wallet held both sides, or entered and exited before resolution; scoring those against a single entry price credited hedging and profit-taking as forecasting accuracy. On the corrected test, run on 2026-07-26, the same ahead-tail statistic read 1.3%, 4.5%, and 44.5%. The thin tiers fell to about a fifth and a sixth of their published values while the deepest tier held. Counting both directions, 13.3%, 34.3%, and 66.9% of records were distinguishable from their own prices one way or the other, and below 1,000 events the behind tail dominated. The table above carries the current figures, which drift a little as the population grows. Mixed events were the raw material of thin records that looked provably ahead; deep directional records carry their evidence either way. The full revision history is in the correction log.