What a signal is
A signal is a category of behavior the score pays attention to. The wallet lane and the token lane look at different signals, because the questions are different, but both feed the same tiers and flags. Two things hold across all of them. No single signal decides a read, a score weighs many together, so gaming one dimension does not move the result much on its own. And what follows is the what, not the how. We describe the categories openly. The exact criteria, the windows, and the cutoffs stay internal, for the reasons on the overview page.Wallet signals
When TrustGate scores a wallet, it is asking whether this address has earned trust through what it has actually done.History and depth
How much real activity the address has, with recent activity carrying more weight than activity from years ago.
Age and longevity
How long the address has been active. Time is one of the harder things to fake.
What it interacts with
The range and quality of the contracts and protocols it has actually engaged with.
Whether it builds
Deploying contracts is a strong, costly-to-fake signal of a real builder rather than a passing wallet.
Counterparty quality
The standing of the addresses it transacts with. Counted as part of its own behavior, never propagated into its score.
Human or automated
Whether the rhythm of activity reads like a person or a script. Automation itself is not punished, the pattern is what gets read.
Value that moves
Whether real value flows through the address, rather than dust meant only to pad a history.
Repeated exit marks
Wallets that keep showing up in coordinated token exits pick up a durable mark that follows them into later reads.
Token signals
When TrustGate scores a token or contract, it is asking whether the thing is what it appears to be, and whether the activity around it is honest.Who holds it
The makeup and trust standing of the holders, not just how many there are.
How holders got in
Whether holders actually bought in or were handed tokens to manufacture the look of a crowd.
How it trades
Whether the trading looks organic or coordinated.
The deployer behind it
The track record of the wallet that launched it, since a deployer carries its own reputation.
Independent interest
How many distinct, unrelated participants engage with it, versus a small cluster moving in sync.
Real, recent volume
Recent and genuine volume counts. Stale or wash-like volume does not carry the same weight.
Temporal token signals
Token Shield now looks at how holders actually behaved over time, not only at a snapshot.Hold duration
How long buyers actually held before selling. Abnormally short holds across many wallets are a pump-shaped signal.
Exit ratio
What share of a purchase each wallet sold back, and whether many credible wallets dumped the majority inside the same window.
Wash trading
Bidirectional pairs that trade the same token back and forth. One-way router flow is not treated as wash.
Honeypot volume
Sell volume near zero relative to buy volume after the initial distribution period.
Coordinated long-hold exit
Credible wallets that held, then sold the majority together. The sync of the sell is the signal, not how long they sat.
Temporal flags already surface on batch and Token Shield reads. The numeric weight of temporal deltas on the published score is calibrated separately and may be zero on a given environment. Flags still fire.
Contract and NFT signals
Non-token contracts and ERC-721 / ERC-1155 collections have their own local scoring path. They do not go through the paid ERC-20 oracle. What they look at, in categories: whether the source is verified, how old the deployment is, how many unique interactors exist, whether the deployer has standing, and for NFTs, whether transfer patterns look like wash trading.Staking signals
Staking is a commitment signal layered on top of wallet and deployer reads. See Staking intelligence. The rules that matter: a stake younger than 7 days earns nothing. Self-staking (staker is the token deployer) earns nothing. Staking a BLOCKED token earns nothing. Unstaking right after a deploy, circular staking rings, and coordinated-exit participation void or reset the stake age.Two-tier explainability
Hardened oracle responses split the explanation on purpose.- Public layer (
publicExplain): short headline and lines a DEX can show a retail user. - Protocol layer (
protocolExplain): structured risk categories a backend can branch on. Still no formula weights.
These are categories, not a formula. Knowing that a read looks at, say, how holders got in does not tell you the line between healthy and suspicious, and that is by design.
Next
Flags
The named patterns a read raises when signals line up the wrong way.
Confidence and states
How sure a read is, and how new tokens are handled.