How to Audit a Blockchain Wallet: A Worked Example
A worked audit of one Ethereum wallet over three months: count the transfers, strip out spam tokens, reconcile one token to the unit, and explain the outflow that looks like a loss.

A wallet audit asks one question: does the ledger agree with the books? The blockchain gives you every transfer, and the company gives you its records. The work is matching the two and explaining every gap.
Below we audit one real Ethereum wallet over three months, the same wallet and period as the first version of this guide. Every figure comes from the Bitquery MCP server, queried through an AI client on 30 October 2026. Each step shows the question we asked and the tool the AI called.
Why wallet audits matter
In 2023 the High Court of Singapore ruled on a case brought by Bybit against a former employee. She kept the Excel files listing the addresses where staff were paid in crypto. In September 2022, Bybit found eight unusual USDT payments made between 31 May and 31 August 2022, to four addresses that were not on any staff list. They totalled 4,209,720 USDT. Bybit sued and asked the court for summary judgment.
The gap showed up when the company compared what it had paid on-chain with what its records said it should pay. That comparison is the core of every wallet audit.
The wallet and the period
- Wallet:
0x7BFEe91193d9Df2Ac0bFe90191D40F23c773C060on Ethereum. Bitquery's label data marks it only as an externally owned account namedpunk2498, with no company tag. - Period: 1 March to 30 May 2024.
- Assumption: treat it as a company wallet, with a company ledger to check it against.
Step 1: Count the transfers
Prompt: How many transfers did 0x7BFEe911…C060 send and receive on Ethereum between 1 March and 30 May 2024, and with how many counterparties?
Tool:
eth_address_flow_summary
| Direction | Transfers | Counterparties |
|---|---|---|
| Received | 1,308 | 444 senders |
| Sent | 577 | 208 receivers |
The 1,308 received transfers carried 428 different tokens. This count covers top-level transfers only. Adding ETH moved by contract code inside other transactions brings the received count to 1,440. Decide which of the two your ledger records before you compare. Otherwise, the two sides never match. The 2024 version of this guide used an older API and counted 1,403, which shows how much the method moves the number.
Step 2: Strip out the spam
A count of 1,308 means nothing until you remove what no one asked for. Active wallets attract unsolicited tokens, and this wallet got plenty of them.
Prompt: Of the tokens this wallet received in that period, how many have a name that reads like a link or a claim offer?
Tool:
eth_transfers_raw_sql(a token-name pattern search)
74 of the 428 tokens carry names such as "Claim $AIOZ reward at https://…" or "claim rewards on …". These are phishing lures: almost all name a look-alike site, such as etherfi.gift or kelplabs.org, set up to get the owner to sign away tokens. Another 41 tokens carry AI or tech brand names such as NVIDIA, Tesla and OpenAI, often sent in amounts in the billions.
None of them belongs in a balance sheet. Keep a list of the token contracts you accept, and reconcile only those. Never filter by symbol: several of the spam tokens reuse real symbols such as BOME, NOT and REZ. The transfer tools also return a price status for each row. A token marked "no_price" needs a second look before it enters the books.
Step 3: Reconcile one token to the unit
Take GRAI, the stablecoin of the Gravita lending protocol, contract 0x15f74458ae0bfdaa1a96ca1aa779d715cc1eefe4.
Prompt: For GRAI (contract 0x15f7…efe4), how much did this wallet receive and send in the period, in how many transfers?
Tool:
eth_address_profilewith the contract filter
| Transfers | GRAI | |
|---|---|---|
| Received | 49 | 420,070.72 |
| Sent | 10 | 420,009.92 |
| Net | 60.79 |
These figures match the 2024 version of this guide, which used a different API, to within 0.01 GRAI. If the company ledger shows the same net change, GRAI reconciles. If it shows a different figure, the transfer list tells you which entries are missing.
Step 4: Explain the outflow that looks like a loss
A reviewer who reads only the counterparties would stop here. Of the 420,009.92 GRAI sent, 393,354.72 went in nine transfers to the zero address, 0x000…000. No one controls that address. On a first pass it looks like money thrown away.
Prompt: What happened in transaction 0x3ee9754c…480a?
Tool:
eth_tx_transfers
The transaction called burn() on the GRAI contract to destroy 104,694.06 GRAI. In the same transaction the Gravita admin contract sent 541.43 GRAI back to the wallet. Gravita refunds part of its borrowing fee when a loan is repaid early, so this fits the pattern of repaying a Gravita loan. A burn of the borrowed stablecoin is how that debt is cleared. Confirm the loan in the company's own records before you book it that way.
The other 26,655.20 GRAI went to a wallet labeled as the 0x Exchange Proxy flash wallet: the wallet sold GRAI through the 0x aggregator.
Step 5: Check where the funds came from
Prompt: Who sent this wallet its GRAI in the period, and what are those addresses?
Tools:
eth_address_flow_summary, thenlabels_for_addresses
| Sender | Label | GRAI |
|---|---|---|
| 0xcf55…2559 | Odos router | 181,320.23 |
| 0x084c…11eb | no label | 92,847.12 |
| 0x3175…c0dc | Curve GRAI/FRAXBP pool | 50,269.21 |
| 0x5db3…d36d | Uniswap v3 GRAI/USDC pool | 34,629.09 |
| 0xdef1…ee57 | ParaSwap (Velora) router | 30,455.91 |
| 0x4f39…56a6 | Gravita default pool | 25,944.32 |
| 0x4928…3d14 | Gravita admin contract | 3,963.13 |
About 70% of the GRAI came through DEX routers and liquidity pools, and most of it was then burned to repay debt. The one unlabeled sender, 0x084c…11eb, is where the audit should look next. Ask the company what that address is and why it sent 92,847 GRAI.
Step 6: Trace further when something does not fit
If a transfer has no explanation in the books, follow the funds.
- In the browser: MoneyFlow Lite draws a flow chart of the funds going into and out of an address, with labels and dollar values. It covers 17 chains, Ethereum among them. It draws on your Bitquery API quota, so you can try it on the 7-day trial.
- With an AI client: ask the MCP server to trace the next hops, as in the steps above. Tracing and labels sit on the AI Investigation MCP plan, $149 per month on the pricing page.
- If funds were stolen: Bitquery's crypto investigation service handles cases that need a written report.
Step 7: Write it up
The audit file should let someone else repeat the work:
- Scope: the wallets, the chain, the period, and whether internal transfers count.
- Sources: the data provider, the tools or queries used, and the date you ran them.
- Token list: the contracts you accepted, and the spam you excluded.
- Reconciliation: for each token, opening balance, inflow, outflow and closing balance, set against the ledger.
- Exceptions: every gap, how you explained it (such as the GRAI burn), and what is still open (such as
0x084c…11eb).
Run the audit yourself
Connect the Bitquery MCP server to Claude, Cursor or ChatGPT with one URL, https://mcp.bitquery.io. The setup guide covers each client. Then ask the questions above about your own wallet.
Related reading
- How to trace a Bitcoin transaction and address
- FATF red flag indicators, checked on-chain
- Cryptocurrency money laundering explained
First published in June 2024 by Divyasshree and Emmanuel Ajala. Rewritten on 30 October 2026 by the Bitquery team, with help from AI tools: the GraphQL examples and the Coinpath link were replaced with steps run live through the Bitquery MCP server on the same wallet and period, and the Bybit case summary was corrected (it was a civil claim, not a prosecution). Bitquery sells the MCP server, the API plans that MoneyFlow Lite runs on, and the investigation service named above. This article is for information only and is not accounting or legal advice.
Subscribe to our newsletter
Subscribe and never miss any updates related to our APIs, new developments & latest news etc. Our newsletter is sent once a week on Monday.