Wallets that keep buying together.
Some wallets buy the same new tokens within 30 seconds of each other, again and again, far more often than chance gives two wallets as busy as they are. Those are listed here as groups, under a number.
Buying together is not proof of coordination. One person with several wallets looks like this, and so do strangers copying the same public source.
The groups
4 groups, buying everything gives 24%
Group 12
Joint buys that ran: 11/664
4 wallets11 of 664 joint buys ran, 1% to 3% at 95% confidencefirst in 193 of 400 it shared with other scored wallets
Group 43
Joint buys that ran: 5/226
3 wallets5 of 226 joint buys ran, 1% to 5% at 95% confidencefirst in 68 of 226 it shared with other scored wallets
Group 1
Joint buys that ran: 17/1,214
2 wallets17 of 1,214 joint buys ran, 1% to 2% at 95% confidencefirst in 472 of 1,176 it shared with other scored wallets
Group 61
Joint buys that ran: 7/71
2 wallets7 of 71 joint buys ran, 5% to 19% at 95% confidencefirst in 2 of 70 it shared with other scored wallets
The figure on the right is the group's joint buys that really ran, out of all its joint buys on the tokens we read. Each square is a joint buy: green when the token ran, grey when it did not. The range beside it is where the true share plausibly lies given how few buys there are; a group whose range reaches down to 24% has not been shown to do better than buying everything.
How a group is found
For every pair of wallets we look at, we count the tokens both bought early within 30 seconds of each other. That count is compared with what chance gives: two wallets that each buy many tokens share many by accident, and crowded tokens are shared by everyone. Two machines that buy every launch are together on every token and are not a group.
A pair is linked when it has at least 5 such tokens, at least 3 times what chance gives, and the odds of that arising by itself are under 1 in 100 across all the pairs looked at together. Linked pairs become a group only when most of the wallets in it are linked to each other, not through a chain of acquaintances.
Wallets funded from the same source were already counted as one voice. A group is counted as one voice too: its wallets cannot raise a signal between themselves.
What this does and does not show
It shows that these wallets have repeatedly bought the same tokens at the same moment. It does not show that they are run by one person, that they talk to each other, or that they know anything the public does not. The data cannot tell those apart.
A group's record is counted on the same buys that made it a group, on the tokens we read: every token that ran and a sample of those that did not. It is a description of the past, and a past record says nothing certain about the next buy.
Who is in a group is not published, and no wallet's address is. A followed wallet's own page, reached from the board, says whether it is in one. Nothing here is advice.