Most Web3 growth metrics are not measurements. They are invoices submitted by sybil farms. The 2023–2025 airdrop meta industrialized fake participation: wallet farms running thousands of addresses through testnets, quest platforms, and points programs, inflating every top-of-funnel number a growth team could report. Then the token launches, the farms exit, and the "community" evaporates — the repeated pattern behind post-airdrop attrition that commonly runs 70–95 percent and post-incentive TVL drawdowns of 40–80 percent.
The industry knows. LayerZero's 2024 sybil screening — self-reporting windows plus bounty-driven hunts — disqualified hundreds of thousands of addresses from a single airdrop, and every serious points program since has shipped with forensics attached. By 2026, exchanges, market makers, and investors all discount raw user counts by default.
Our position, after running growth for 450+ projects since 2021: sybil resistance is not a post-hoc filtering problem. It is an acquisition design problem. You either acquire humans expensively-looking-but-cheap, or acquire bots cheap-looking-but-ruinous. This post covers the design levers and the measurement standard.
Why farmed metrics collapse
Incentivized acquisition has an adverse-selection engine at its core: the lower the friction and the more legible the reward, the more attractive the campaign is to professional farmers — who respond faster than genuine users, because responding is their job. An open quest with a click-based task list and a hinted airdrop is optimally designed for sybils and roughly neutral for everyone else.
The collapse is then mechanical. Farmed wallets have no use for the product; they hold exactly until reward realization, exit in the same week, and take your liquidity, your chart, and your credibility with them. Worse, the numbers poison decision-making for months beforehand — teams tune products against bot behavior and negotiate listings on user counts that will not survive contact with TGE.
Filter one: cost of action
The single most effective design lever is making each identity cost more than its expected reward. A farmer's economics are simple: expected reward per wallet minus cost per wallet, times N wallets. Raise the cost side and N collapses.
Practical mechanics, in rising order of friction: gas-bearing transactions rather than off-chain clicks; small non-refundable commitments (a paid boost, a Telegram Stars purchase, a nominal stake); time-locked participation that forces capital or attention to sit for weeks; and escalating quest chains where meaningful rewards only unlock after cumulative real usage. In our campaigns, even a 0.50–2 dollar cost-of-action gate typically removes the majority of farm traffic while barely denting genuine conversion — humans who want the product pay small costs; scripts do not.
The tradeoff is honest: cost-of-action suppresses top-of-funnel volume, and your dashboard will look worse before it looks real. Full proof-of-personhood (biometrics, KYC) filters harder still but at real privacy and drop-off cost; we reserve it for high-value allocations, not general acquisition.
Filter two: regional targeting
Sybil farms concentrate where distribution is broadcast and anonymous — global quest platforms, open airdrop aggregators, English-language campaign pages. They are structurally weak where distribution is relationship-gated: a Vietnamese Telegram group with active admins, a Chinese-speaking WeChat community, a Korean Kakao chat. Entering through native-language communities — the core of regional growth — is itself a sybil filter, because each acquired user arrives vouched by a context that scripts cannot cheaply fake.
Regional targeting also makes anomalies visible. A campaign run through 40 Vietnamese communities that suddenly shows Eastern European IP clusters and uniform wallet funding patterns is easy to flag; the same traffic hidden inside a global quest feed is not.
Filter three: KOC funnels
Key opinion consumers — small creators with 500–10,000 followers — are the acquisition channel farms imitate worst. A KOC wave of 100 small, historied accounts across target communities produces referral trees that can be audited creator by creator: humans refer bursty, socially clustered, variably-sized downlines; farms refer uniform ones. We attribute every campaign at the creator level and read 30-day retention per cohort, which turns the KOC roster itself into a quality-ranked asset over time — the operating core of our KOL and KOC programs.
Measure retained wallets, not signups
The measurement standard that keeps everyone honest is a ladder, with reporting anchored at the bottom rungs:
| Metric | What it counts | How gameable |
|---|---|---|
| Raw signups / quest completions | Clicks with an account attached | Trivially |
| Verified humans | Passed cost-of-action or personhood gate | Moderately |
| Activated wallets | First real product transaction | Weakly |
| Funded wallets | Wallet holds or deposits meaningful value | Weakly |
| D30 retained funded wallets | Still active a month later | Barely |
The north-star metric we hold campaigns to is D30 retained funded wallets, with D90 as confirmation. Everything above it on the ladder is diagnostic, not achievement.
What quality CAC looks like in 2026
Indicative ranges from our 2025–2026 campaigns, which vary widely by vertical and region: raw quest signups cost 0.05–0.50 dollars; verified humans 0.50–3; activated wallets 2–10; and D30 retained funded wallets commonly land at 8–40 dollars — cheaper in Southeast Asian consumer categories, more expensive for DeFi in Korea or Japan.
The psychological hurdle is that quality CAC looks expensive next to vanity CAC — 20 dollars per retained wallet against 10 cents per signup. The comparison is false: the signup pool retains a few percent at best, so its implied cost per retained user is often similar or worse, before counting the token allocation burned on farmers and the credibility cost of a post-TGE cliff. One mid-cap exchange we worked with cut incentivized signup volume by more than half after moving budget behind cost-of-action gates and regional KOC funnels; total signups fell, while retained funded accounts per dollar roughly doubled. That trade is the whole thesis.
FAQ
What is sybil resistance in Web3 marketing?
Designing user acquisition so that fake or duplicate identities are unprofitable to operate — through per-identity costs (gas, stakes, paid actions), relationship-gated regional distribution, and creator-attributed funnels — rather than trying to filter bots out of the data after an open campaign has already attracted them.
What retention is considered good for Web3 user acquisition in 2026?
For incentivized campaigns, D30 retention of 10–20 percent of verified users is solid and above 20 percent is strong; unincentivized product-led cohorts should do meaningfully better. If more than half of "users" never fund a wallet or return after week one, the campaign bought traffic, not users.
Is buying quest-platform users ever worth it?
Sometimes — for awareness, social proof, and testing funnel mechanics — if you price it as advertising rather than acquisition and never report it as users. The moment quest volume feeds a token allocation or a listing narrative, it becomes a liability. Our full acquisition framework is in the playbook; to pressure-test your funnel design, talk to us about user growth.
