← back to blog

Referral trees and why self referring backfires

What a referral tree actually tracks

A referral link looks simple from the outside. You get a code, someone signs up or connects a wallet through it, and you both get some kind of credit toward a future distribution. Underneath that, though, every referral program is building a graph. Each referral event is an edge between two identities, whether that identity is a wallet address, an email, a device ID, or all three tied together. String enough edges together and you get a tree: one root wallet, a set of wallets it referred, the wallets those referred, and so on.

Projects keep this graph because it’s useful for two very different purposes at once. It’s a growth mechanic, since referral trees are one of the cheapest ways to get real users to pull in more real users. But it’s also a detection tool, because the same graph structure that shows organic growth also shows synthetic growth. A tree where one wallet refers 40 wallets that never do anything except claim a token and refer nobody else looks completely different from a tree where a real community is inviting friends. Anyone building an airdrop referral program with a claim in mind designs the reward structure and the fraud detection at the same time, because they know from prior airdrops that the referral mechanic is the first thing people try to game.

The funding trail problem

The most basic thing a chain analysis process checks is where the gas came from. Every wallet needs native token to pay for its first transaction, and that native token has to arrive from somewhere. If wallet B and wallet C both received their opening gas from wallet A in the same block range, and A also happens to be the referrer for both, that’s not a coincidence a clustering algorithm needs much sophistication to catch. It’s a two-hop link: A funds B and C, A refers B and C. Even without any other signal, that pattern alone is enough to flag the group for manual review on most airdrop-aware analytics tools.

This is why funding a farm’s wallets from one hot wallet or one exchange withdrawal, in a tight time window, is one of the most common reasons a batch gets zeroed out. The referral tree just makes the pattern easier to find, because the referral edges point analysts straight at the wallets worth checking for shared funding in the first place. Without a referral link between them, an analyst has to go looking. With one, the project handed them the search query.

Device and network fingerprints add another layer

On-chain data is only half the picture. Most airdrop campaigns that run through a web app also log request-level signals: IP address, TLS fingerprint, browser fingerprint, and session timing. A referral tree built from a signup form, not just a wallet address, ties an email or account to a device the same way funding ties a wallet to a source.

If ten “different” referred accounts all load the claim page from the same subnet, present the same canvas and WebGL fingerprint, or share timezone and font list quirks, that’s a second, independent signal pointing at the same operator, and it correlates with the referral graph rather than contradicting it. Analysts don’t need either signal alone to be conclusive. They need the two to line up, because coincidence in one dimension is common but coincidence in two independent dimensions is not. This is why running multiple wallets credibly is a browser and network problem as much as a wallet problem: distinct wallets behind identical fingerprints undercut the separation the wallets were supposed to have, and a referral tree connecting those same wallets removes any ambiguity that’s left.

Timing patterns compound the risk

Clustering isn’t only about who is connected to whom. It’s also about when things happen. A referral chain where every downstream wallet claims within minutes of the one above it, in the same order the tree was built, reads as scripted rather than organic. Real referral adoption is messy: people sign up over days or weeks, some never finish onboarding, some show up late after seeing a friend’s post. A farm that fires off its whole tree in one script run compresses all of that into a pattern with almost no natural variance, and variance is exactly what a project’s fraud model is trained to look for the absence of.

The same applies to task completion order within a campaign. If every wallet in a tree does the identical five actions in the identical sequence with a similar gap between each, that sequence itself becomes a fingerprint, independent of gas source or device. It’s one more axis a clustering pass can group wallets on, and it stacks with the others instead of replacing them.

What happens when a wallet gets flagged

The practical cost isn’t abstract. When a project’s sybil review flags a cluster, the usual outcome isn’t a warning to the top wallet. It’s exclusion of the whole connected component, because the point of building the tree in the first place was to make the group’s shared identity visible. A wallet that did nothing wrong on its own, but sits two hops from a flagged root through a referral edge, can lose its allocation purely on graph proximity. Some programs are stricter than others about how far the exclusion radius extends, and that detail is rarely published, which means a self-built referral tree is a liability with a blast radius the operator doesn’t fully control.

This is also why a self-referral tree is worse than simply running unconnected wallets that happen to get flagged individually. Individual false positives are a normal cost of any farm at scale. A referral tree turns those individual risks into a shared one, because flagging any single node gives an analyst a direct, labeled path to every other node the operator built.

Referral trees and cluster contamination

There’s a second-order cost worth naming directly: a self-referral structure doesn’t just risk the wallets inside it, it can contaminate wallets that were never meant to be connected. If a farm reuses a proxy exit, a browser profile, or a funding wallet across separate “unrelated” identities and also runs a referral tree among a different set of wallets, an analyst who links the shared infrastructure now has a bridge between two clusters that were supposed to be independent. What was meant to be several small, disposable clusters becomes one large one the moment any edge, on-chain or off, connects them. Referral programs are one of the easiest ways to accidentally create that edge, because they’re explicit, timestamped, and often stored in a project’s own database rather than inferred from behavior.

Running multiple wallets without building a self-referral tree

None of this means referral programs are off limits, and it doesn’t mean multi-wallet participation is inherently a problem, since plenty of real users hold several wallets for legitimate reasons. The distinction that matters is whether the wallets are structurally connected to each other, on-chain through funding, off-chain through device and network fingerprints, and through explicit graph edges like referrals, in a way that a real independent user’s wallets wouldn’t be. A referral link between wallets an operator controls is one of the most direct ways to create exactly that kind of connection, because it’s voluntary, logged by the project itself, and rarely worth the marginal reward compared to what it costs if the cluster gets reviewed.

The honest framing is that a referral program rewards bringing in participants who wouldn’t otherwise have shown up. Referring your own wallets doesn’t do that. It just adds a labeled edge to a graph that was already being watched, on a mechanic that projects built specifically because they expected people to try this.

The bottom line

Referral trees are a growth tool that doubles as a fraud detection tool, and self-referring collapses that distinction in the worst possible direction. It creates funding links, timing links, and often fingerprint links between wallets that were otherwise separable, and it does so through a channel the project logs on purpose. Treating referral rewards as something to farm with your own wallets tends to cost more in cluster exposure than the reward itself is worth.

If you want more breakdowns like this on how airdrop eligibility actually gets evaluated, and tested notes on the tools involved in running wallets the right way, head back to the Airdrop Farming home page.

Get new guides and videos first — join the Telegram channel.

need infra for this today?