Winning Polymarket traders: what evidence to ask for before comparing

Correction (July 16, 2026): the previous version compared named wallets and published grade-cohort benchmarks without committing the production query and output needed to reproduce them. We withdrew those wallet grades, cohort counts, medians, and strategy comparisons instead of carrying them forward as current facts.

The URL is unchanged so earlier citations reach this correction. The old planning document is superseded and is not a source for this page.

A fair comparison starts by naming where each number comes from. Portfolio profit belongs to Polymarket's portfolio-level record. Market rows can describe distributions and outcomes, but summing them does not give a second account total unless the accounting rules say it does.

A claim about a group of wallets also needs a retrieval time, inclusion rules, a rule for missing values, and a saved read-only result. Without those, a median or grade count is a passing observation that nobody can check later.

Profit factor divides gross gains by gross losses. It means something only when the numerator, the denominator, which market statuses count, and the zero-loss case are all stated. Comparing medians across grades adds one more requirement: which wallets held each grade at retrieval time.

The earlier profit-factor medians, rolling Sharpe medians, and top-tier wallet count had no committed query output in our repository, so we removed them. A future cohort study has to publish its read-only query and saved result beside the article before any benchmark comes back.

Win rate counts how often a market row finished positive. It does not say how large the gains were, how large the losses were, how much capital was deployed, or whether the result came from a repeatable process.

The chart below illustrates that arithmetic with made-up points: similar win rates next to widely different profits. No point is a real wallet or a current grade.

A metric from a short history is fragile. Before comparing two traders, check the resolved-market count, the time span, and whether one event dominates the result. A careful estimate shrinks a thin record toward the average instead of treating a short streak as skill.

Many trades are not proof of edge either. The result has to hold across enough independent markets to survive ordinary variance.

A strategy label names a pattern in past activity. It does not show that the pattern caused the profit. Comparing strategy groups needs the same reproducible inclusion and retrieval rules as comparing grades.

Do not copy a strategy because an undated chart ranks it highly. Open the live profiles, check that the label matches the actual positions, and ask whether you could execute the same way.

Concentration shows where a trader operates. It does not show that specializing produces better returns. A claim that specialists beat generalists has to define the categories, the concentration threshold, the eligible wallets, and the profit basis.

We withdrew the earlier specialist comparison because none of that was committed with the article.

Search a username or wallet address from the leaderboard, and remember that the profile updates live. Read portfolio profit, market diagnostics, risk measures, and sample size on their own bases instead of copying one headline number.

When an article compares wallets, look for a dated source and a reproducible query. Without them, the honest answer is that nobody knows.