Exact ANN recall on the Bitcoin address vectors

You are looking at Bitcoin value-flow graphBitcoin address behavioural vectors, disk-resident in one Semurg substrate. Ask it to run the ANN recall benchmark and it runs the real query live on this box, on data you can download and re-run yourself ▸.

Click an example in the agent rail, or ask recall at 128 bits, 256 bits, or 128 bits for the binary-density line. Every run is measured end to end, on this box, on the real dataset. Nothing is pre-computed.

the live Bitcoin value-flow graph, transactions and addresses linked by value-flows (prev-tx to tx to address), parsed straight off the P2P network · Bitcoin mainnet P2P (our own block-parser node, no bitcoind, no REST). An exact float-cosine scan over the Bitcoin address vectors recovers 100% of the true neighbours (recall 1.0), no index and no recall cliff, and the vectors stay disk-resident. Measured live.

the vector-DB leader, same vectors

RECALL CLIFF, in RAM. An approximate index trades recall for speed and must hold the graph in memory; as the corpus grows past RAM the index, not the data, is the wall.

semurg, one engine

100% recall@10, disk-resident. The exact topK over the stored address vectors IS the ground truth, so no neighbour is ever missed, and RAM is the pipe, not the store.

Exact ANN recall over the Bitcoin address vectors, off disk.

Bitcoin address behavioural vectors, derived from the live value-flow. Measured live on this box once the deriver has materialized them.

ask the agent to "search", the recall@10 and QPS render here

Honest scope: the recall number is measured against the exact float-cosine topK the engine computes in the same pass (recall 1.0 by construction), so the binary code's recall is the density lever. The same substrate also serves graph, search and key-value.

Seen enough? Every number here is measured live on public data, and moves every week. Talk to the team ▸