Two live datasets. Every domain.

The whole benchmark suite runs on exactly two public, live datasets: the Bitcoin blockchain and the Binance WebSocket. Not a pile of frozen academic corpora, two real feeds a skeptic can pull themselves, updating every second, that between them exercise every workload a universal substrate claims. One dataset, every face.

Why these two

1 · universally available

Anyone can pull the Bitcoin chain over P2P or subscribe to the Binance socket, for free, right now. The benchmark is re-runnable by any skeptic on the same public data, the one thing a fabricated number cannot survive. No licensed corpus, no "trust our export".

2 · constantly updating & dynamic

A new Bitcoin block every ~10 minutes, thousands of Binance trades every second. The graph and the stream are ALIVE, not a snapshot, so the numbers exercise live ingest, churn, streaming and consistency, the things a load-once academic file can never show.

3 · genuinely large · ~750 GB

The full Bitcoin chain is ~750 GB, on the order of a billion transactions and a billion addresses, one of the largest verifiable real graphs in existence. Not a toy: a value-flow graph big enough that the engine, not the dataset, is the thing under test.

4 · every modality on one dataset

The SAME Bitcoin data is a graph (value-flow traversal), a key-value store (address / tx point read), a SQL/OLAP table (aggregate the flows), a search corpus (pattern-scan tx & addresses), and an object store (tx bytes + metadata + links). Binance adds time-series + streaming. One dataset, every domain, which is the whole cross-domain moat, made concrete.

Every demo on this site now runs on one of these two live sources. Bitcoin for the graph, KV, SQL, OLAP, search and object faces; Binance for time-series and streaming. Every number is measured live, on this box, on data you can pull yourself.

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