Polymarket Whale Tracker

Tracking a whale is trivial; every Polymarket wallet is public. The hard part, and the part that determines whether the tracking is worth anything, is checking whether the whale's record separates from luck before you act on it. Convexly's free wallet check does that read on any public 0x address: realized entry edge with a bootstrap 95% interval, a 30-resolved-position readability floor, and a concentration screen. First wallet free, no signup; more wallets free with an account.

What a 100-whale calibration audit actually found

We ran a calibration audit on the top 100 Polymarket wallets by realized profit, a cohort holding $115.7M in combined realized profit across the 95 of 100 wallets with resolvable position data. The intuition a whale tracker sells is that big profits imply good judgment. The data points the other way: the Spearman rank correlation between Brier score and realized profit was +0.42 (bootstrap 95% CI +0.25 to +0.57), meaning the worse-calibrated whales tended to earn more, and the correlation held after removing the ten largest profit wallets (+0.42 to +0.41). The conditional-median version is starker: the 21 of 100 wallets whose Brier score was worse than a coin flip (above 0.25) had a median profit of $1,256,508, versus $625,593 for the other 79 wallets, a 2.02x ratio in favor of the worse forecasters. The full method, charts, and CSV are in the Polymarket whale calibration audit.

Most whale records are one event wearing a track record

Median single-event concentration across the 100 whales in the audit was 70.7%, and 46 of the 100 drew over 80% of their profit from a single event. 27 of the 100 had a 2024 US election market as their single biggest source of realized profit; those biggest-event wins alone add up to $33.8M, 29% of the cohort's combined realized profit, from one election cycle. When one resolved event dominates a record, the record cannot be separated from that one outcome, and no amount of additional volume repairs the read.

Clustering makes this worse, not better. In the 2024 US popular-vote markets, 8 of the wallets in the audit entered the same side in the weeks around election day (the middle half of first entries landed inside an 8-day span centered on October 31), risking a combined $7.2M and capturing $12.3M. A tracker feed shows that as eight independent whales agreeing. It was one bet, expressed eight times, at implied odds of roughly 28% to 39%. If you had mirrored the cluster you would have inherited the concentration, not the judgment.

The three checks that come before acting

1. Realized entry edge with its interval. Did the whale's entries resolve true more often than the prices it paid implied? Only a bootstrap 95% interval that clears zero on the positive side is consistent with an edge on the record. A point estimate without an interval is a screenshot.
2. The 30-resolved-position floor. Below 30 resolved positions the interval is too wide for any verdict, and the honest output is “too few resolved trades to read”. Whale-sized positions do not lower the floor; sample size is about counts, not dollars.
3. The concentration screen. If one event is at least 60% of the net result, the record is flagged as too concentrated to read, and that flag overrides any flattering number next to it. By this screen, most of the whale audit cohort is unreadable as skill.

Run all three on any whale in about 30 seconds with the free wallet check. Most records, whale-sized included, read “cannot tell”, and that is the answer working as intended: it stops you from paying for conviction that the data does not contain.

Diagnostics on public on-chain records, not investment advice; a past read is not a forecast.