One Person, Many Wallets: Linking Multi-Wallet Users Without Collecting Identity

Almost every web3 fraud, sybil airdrops, wash trading, allowlist gaming, multi-accounting, reduces to one problem: a single human operating behind many wallets. Solve the linking problem and most of the fraud collapses with it.

Wallets are free, infinite, and pseudonymous, so the wallet address is useless as an identity anchor. What does not multiply for free is the device and browser that connects them.

Why the wallet is the wrong unit

A wallet address is a keypair. Anyone can generate ten thousand in a second, and nothing on-chain ties them to a person or to each other unless the operator is careless with funding. Building fraud defense on the wallet address means defending a unit the attacker mints for free.

The strong unit is the environment: the device, browser, and network path behind the wallet’s activity. Those cost real money and are far harder to multiply, which is exactly why they make a better identity anchor.

How a stable visitorId links wallets

Prynt assigns each browser session a stable visitorId, a high-entropy device identifier built from the browser and network attributes that survive the usual evasions:

• Incognito mode does not reset it.
• Cleared cookies and storage do not reset it.
• A new wallet address does not reset it.

So when fifty wallets connect from one browser over a week, they share one visitorId, and the link is obvious. The identifier answers a narrow, powerful question, “is this the same device?”, without ever asking who the person is. That is what makes it a privacy-respecting alternative to document KYC.

Building the wallet-to-device graph

Turning the identifier into a defense is a matter of joining data you already have:

1. Capture visitorId on every wallet connect, quest, trade, or claim.
2. Store the wallet-to-visitorId mapping alongside your existing activity records.
3. Cluster wallets by shared visitorId to reveal multi-wallet operators.
4. Enrich with network signals, proxy and datacenter detection, so shared-device clusters routing through one pool score higher.

The output is a clustering of your wallets by the humans likely behind them, which feeds sybil scoring, referral-abuse checks, and allocation caps.

Distinguishing farms from legitimate sharing

Not every shared device is fraud. Households share laptops; couples share a desktop; a community may gather on shared campus wifi. The size and pattern of the cluster separates the cases:

• Two or three wallets on one device over months looks like a household.
• Eighty wallets appearing on one device in a qualification window is a farm.
• Coordinated tempo across the cluster, machine-paced actions, distinguishes automation from a family.

Weight clusters by size, network reputation, and behavioral regularity, and set proportional outcomes: full trust for small organic clusters, caps for medium, exclusion for dense farms.

Why cross-site reputation multiplies the signal

A device that operated a farm in another project’s campaign is not a blank slate in yours. A shared reputation network lets the visitorId carry history, so a burned farm arrives pre-flagged and the operator has to rebuild infrastructure instead of amortizing it across campaigns. That rising cost is the whole game in sybil defense.

Privacy by design

Linking by device rather than identity is not just convenient, it is the privacy-preserving choice. You learn that two wallets share a machine without learning the owner’s name, email, or documents. Pair that with explainable reason codes and an appeals path, and you get strong sybil resistance that communities can actually accept.

Start linking

Add the agent, log the visitorId on every wallet interaction, and run one clustering query. The multi-wallet operators in your data surface within days, and the shape of your sybil problem stops being a guess.

Explore what the identifier survives in the Prynt playground, and start on the free tier. As your activity grows, the pricing scales with session volume rather than the wallets you are trying to link.