Elliptic temporal network

Bitcoin transactions, 49 time steps

Each node is a Bitcoin transaction. Each edge is a flow of Bitcoin from one transaction's output into the next transaction's input. Transactions are labeled by the real-world entity behind them: licit (exchanges, wallet providers, miners, licit services) or illicit (scams, malware, ransomware, ponzis, terrorist organizations). 77% are unlabeled.

The dataset spans 49 consecutive ~2-week snapshots. Edges in Elliptic don't cross step boundaries and no node appears in more than one step — so the brush below picks one step at a time, and the graph you see is exactly that step.

Drag the brush, click a bar, or use ←/→ to move through time.

Illicit (4,545 txs, 2.2%)
Licit (42,019 txs, 20.6%)
Unlabeled (157,205 txs, 77.2%)
Color nodes by
Ground truth
Prediction
Only 22.8% of transactions are labeled. Switch to prediction to color the unlabeled 77.2% with an XGBoost classifier (F1≈0.78 on temporal split).
Step—
Visible nodes—
Visible edges—
Illicit (true)—
Licit (true)—
Unlabeled—
Predicted illicit—
Predicted licit—
Step has many transactions — simulation may be slow.
Brush bars show transaction counts per step; red overlay = illicit share. ← / → moves to the previous / next step. Scroll to zoom, drag nodes to pin.