Skip to content

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

13 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 

Repository files navigation

Bybit theft graph

Public Ethereum transfers around the Bybit cold-wallet theft on 21 February 2025. The cold wallet 0x1db92e2eebc8e0c075a02bea49a2935bcd2dfcf4 was drained to a single immediate receiver 0x47666fab8bd0ac7003bce3f5c3585383f09486e2, from which funds were dispersed across a network of intermediate wallets. The FBI later published a list of 51 destination wallets identified as part of the laundering operation. The graph also contains wallets from unrelated ETH transfers in the same time window — addresses with no connection to the theft — added to make the benchmark dataset more realistic. These are referred to as noise.

The download is public chain data from bigquery-public-data.goog_blockchain_ethereum_mainnet_us.

How the graph is constructed

1. Download: Direct ETH transfers, internal ETH transfers from contract calls, and ERC-20 token transfers are pulled from bigquery-public-data.goog_blockchain_ethereum_mainnet_us for the window 21 Feb – 7 Mar 2025 and written to out/raw/.

2. Local graph walk: A breadth-first walk traces funds forward from two starting points:

  • Cold wallet 0x1db92e2eebc8e0c075a02bea49a2935bcd2dfcf4 — starting at the moment of the theft, 2025-02-21 14:13:35 UTC
  • First receiver 0x47666fab8bd0ac7003bce3f5c3585383f09486e2 — starting at the time it first received funds from the cold wallet

At each hop, a wallet is expanded only once, and only transfers at or after the time it first received traced funds are followed. The walk goes up to 10 hops. A wallet with more than 50,000 outgoing transfers is treated as a high-volume router and is not expanded.

3. Enrichment: Token metadata (symbol, name, decimals) to decode ERC-20 amounts, and a scan of the 90 days before the theft to set an active_before flag per address.

4. Noise (optional): --noise samples a small set of unrelated ETH transfers from the same window and adds them to the graph. The addresses introduced are written to out/noise_nodes.csv.

5. Export: The graph is exported to three CSVs in out/:

File Contents
nodes.csv One row per address: hop distance, contract flag, prior-activity stats
edges.csv One row per transfer: amount (decimal-scaled), asset symbol, timestamp, gas info
tokens.csv Symbol, name, decimals for every ERC-20 in the graph

Neo4j

Addresses are nodes. A transfer is a relationship between two addresses.

A pre-built dump is hosted at s3://gds-public-dataset/bybit-hack-graph.dump. Load it with:

neo4j-admin database load graph --from-path=.

Run

pip install google-cloud-bigquery
gcloud auth application-default login
python3 download_graph.py --project YOUR_PROJECT_ID
python3 download_graph.py --project YOUR_PROJECT_ID --noise

--project is the Google Cloud project ID. out/ is generated and not committed.

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages