The common assumption is that once funds hit a mixer or cross a bridge, the trail is gone. In practice it rarely is — it’s just expensive to reconstruct by hand.
We train models to do the reconstruction:
- Cluster addresses that belong to the same actor from behavioural fingerprints.
- Score counterparties by risk and by how directly they connect to a target.
- Re-link flows across mixers and bridges using timing, amount, and graph structure.
The output isn’t a heatmap to admire. It’s a ranked set of leads — this address, this exchange, this off-ramp — with the evidence that put each one on the list.