Independent wallets, not big wallets
One actor running six addresses is one opinion. Getting that distinction right matters more than expanding coverage.
The hardest problem in this space is not finding wallets. It is knowing which wallets are the same person.
Every alert we publish rests on a claim: that several separate participants decided the same thing at roughly the same time. If that claim is wrong if the six wallets are one desk splitting an order then the alert is not a pattern. It is one trade, dressed up as consensus, and it is worse than no alert at all because it feels like confirmation.
This is why we weight independence as its own input rather than folding it into the wallet count. A cluster of six addresses that resolve to two actors scores materially lower than a cluster of four addresses that resolve to four. Both appear on the surface as 'multiple wallets buying'. Only one of them is information.
The instinct when building a wallet dataset is to start from known entities exchanges, market makers, public funds and grow outward from labels. We tried that and it stalled: the labels were accurate and the coverage was useless. What worked was inverting it. Group wallets by how they behave first, and treat identity as a secondary, optional question. You do not need to know who an actor is to know that two addresses are probably the same actor, and for our purposes the second fact is the one that changes the score.
We are not going to publish the specific signals we use to make that judgement. They are the part of the system a competitor would most like to have, and they are also the part most easily gamed by anyone who wants their flow to look like a crowd. What we will say is that the test is behavioural and continuous, not a one-off label, and that it is applied on all four chains by automated crawlers rather than by a labelling team.
The corollary is a bias we will defend. We would rather track fewer wallets we understand than a large number we do not. A bigger tracked set is an easy number to advertise and a poor proxy for signal quality the marginal wallet in a bloated set is usually a bot, a bridge or a duplicate, and each of those actively degrades the feed. Coverage grows every day, but it grows on the crawler's terms, not on a marketing target's.