A TON gaming infrastructure project ran a public whitelist registration in late 2025. Within six hours of opening, 14,000 wallets had registered. By the time the registration window closed, 9,200 of those wallets had been created within the previous 30 days, 3,100 shared funding transaction patterns with other registered addresses, and 1,800 appeared to originate from the same five source wallets across a coordinated address-generation script. Without scoring, every one of those addresses would have received an equal allocation opportunity — representing close to 50% of the whitelist filled by addresses that were, in effect, controlled by a small number of actors farming the sale.
Anti-sybil scoring is the mechanism by which a launchpad assigns differential allocation weight to wallets based on behavioral signals rather than treating every registered address as a unique, independent participant. The term "Sybil attack" refers to the practice of creating many pseudonymous identities to gain disproportionate influence over a system — in a public token sale context, this translates directly to wallet farming: deploying dozens or hundreds of wallets to accumulate allocation slots that would otherwise be distributed across genuine participants.
Why On-Chain Behavioral Signals Beat Captcha
CAPTCHA and other bot-detection approaches address a different threat model: automated form submission by scripts. They do very little against coordinated farming by humans — or by low-sophistication bots that can solve modern CAPTCHA variants in under a second using commercial services. A wallet farming operation does not need to be automated to be effective; a single operator manually creating 50 wallets across a weekend represents a meaningful concentration of allocation capture at modest effort cost.
On-chain behavioral signals are more structurally resistant because they cannot be manufactured quickly. The signals that matter most in the TON ecosystem include:
- Wallet age: The date of the wallet's first on-chain transaction, measured against the whitelist snapshot date. A wallet created two weeks before registration opened carries substantially less weight than one with 18 months of on-chain history.
- Transaction diversity: Whether the wallet has interacted with multiple protocol types — TON jetton transfers, staking through nominator pools, DEX interactions on STON.fi or DeDust, TON DNS registration — rather than a single funding transaction and nothing else.
- Nominator participation history: Wallets that have participated in TON validator nomination — staking TON with validator nodes — demonstrate a level of ecosystem commitment that is difficult to fake at scale. The minimum staking thresholds for nominators create a cost barrier that most farming operations do not clear.
- Telegram account age: For projects distributed through Telegram Mini Apps or the TON ecosystem's Telegram-native touchpoints, the linked Telegram account's creation date provides an additional signal. Account age correlates weakly but meaningfully with genuine user status versus freshly-created farming accounts.
- Cross-wallet correlation patterns: Network analysis of funding chains — whether multiple registered addresses received their initial TON balance from the same source address within a short time window — is the most direct indicator of coordinated farming. A cluster of 20 wallets all funded from a single originating address in the same hour is not statistical noise.
How Scoring Translates Into Allocation Weight
The output of sybil scoring is not a binary pass/fail — it is a numeric score that feeds into an allocation tier. Tonstarter's scoring produces a value between 0 and 100 for each registered address. This score determines which allocation bracket the address qualifies for:
"Scoring does not decide who participates — it decides how much weight each participant's claim carries. A low-scored address can still register; it receives a proportionally smaller allocation within its bracket."
The tier structure typically looks like this: low-score addresses (farming-risk profile) receive minimal guaranteed allocation with heavy pro-rata compression; mid-score addresses receive standard tier allocations; high-score addresses receive preferential allocation access. The exact score thresholds and tier allocations are set per sale by the project team in collaboration with the launchpad — they are not fixed platform constants.
The New-User Penalty Problem
There is a genuine tension in wallet-age scoring that deserves direct acknowledgment rather than dismissal. A legitimate new TON user — someone who created a wallet three weeks ago, funded it through Tonkeeper, and has begun using the ecosystem — will score low on age signals regardless of their genuine intent to participate in a sale. This is not an edge case: TON's user base has been growing, and a meaningful fraction of any sale's target audience may be relatively new to the ecosystem.
Scoring systems handle this through two mechanisms. First, diversity signals can partially offset age penalties: a wallet that is three months old but has made jetton transfers, interacted with at least one DEX, and linked a non-freshly-created Telegram account demonstrates behavioral patterns inconsistent with a farming address. Second, tier floors exist: no registered address scores zero unless it meets explicit disqualification criteria (funding directly from a blacklisted cluster source, for instance). The floor means new legitimate users receive some allocation access, even if reduced relative to established wallets.
We are not saying that anti-sybil scoring penalizes new users unfairly — we are saying the penalty is structural and partially unavoidable, and that the tier floor design is specifically intended to keep the system accessible while reducing farming capture. Projects should communicate this to their communities before registration opens, not after.
What Scoring Is Not
Anti-sybil scoring is a behavioral assessment of on-chain patterns. It is not identity verification. A high sybil score does not mean the wallet owner has been identified, vetted for AML compliance, or cleared under the FATF travel rule. It means the wallet exhibits behavioral signals consistent with a genuine, non-farmed address.
This distinction matters for how projects design their compliance stack. For token sales above applicable thresholds — where KYC is legally required for the project's jurisdiction and sale structure — sybil scoring and KYC are both necessary and serve entirely different functions. Scoring reduces manipulation surface and adjusts allocation weighting; it does not replace KYC due diligence. The two mechanisms operate in sequence: sybil scoring determines allocation eligibility and tier; KYC determines compliance eligibility to participate at all.
Wallet Clustering Detection: What the Algorithm Actually Looks At
Clustering detection runs a graph analysis across all registered wallets for a given sale. Each wallet is a node; a directed edge is drawn from any wallet to any other wallet it funded. Clusters are identified using connected-component analysis: a set of wallets that are all reachable from a common origin address within two hops constitute a potential farming cluster.
Not every cluster is disqualifying. A legitimate user who gifted TON to a family member, both of whom then registered for the same sale, will generate a two-node cluster. The system's response is proportional: a two-node cluster with established wallet ages and diverse transaction histories is treated as a low-risk signal; a 40-node cluster with freshly created wallets, identical funding amounts, and no transaction history other than the initial funding is treated as a high-risk farming indicator.
TON network gas fees and finality are governed by TON consensus, not by Tonstarter — which means the on-chain data feeding the scoring system is transparent and independently verifiable. Any wallet owner can check the signals their address generates. This transparency is intentional: the goal is to make farming economically inefficient by imposing on-chain cost, not to make participation opaque.
For projects considering the Tonstarter platform, anti-sybil configuration details — including scoring signal weights, tier thresholds, and the clustering detection parameters — are covered during the onboarding review. You can start the application process at apply.html, and the anti-sybil framework is described in more depth on the Anti-Sybil Scoring page.
