Two different questions
It is easy to read a tier as the whole story. It is not. A read actually answers two separate questions. The tier and score say how the behavior looks. Confidence says how much that read is standing on. They move independently. A token can look fine on early signals while the read still sits at low confidence, because there is just not much history to judge yet. The two together are far more honest than either alone.Confidence
Confidence is a measure of how much there is to go on. An address or token with a long, rich history gives a read plenty to work with, so confidence is high. Something brand new, or thinly used, gives the read very little, so confidence is lower no matter how the early signals point. On the wire, confidence may arrive as a number0–100 or as a label ("low", "medium", "high"). Discovery is tolerant of both. Oracle and gating normalize to a number.
Low confidence is not a bad mark. It is the read being upfront that the picture is still forming. The right move when confidence is low is to lean harder on your own checks, not to treat the tier as settled. As real behavior accumulates, confidence climbs and the read gets sturdier.
Protocols that gate should set a minConfidence on their own ladder. A HIGH score at 20 confidence is not the same input as a HIGH score at 85.
Think of the tier as what the behavior says, and confidence as how loudly it is saying it. A confident BLOCKED and a low-confidence BLOCKED mean different things, and TrustGate tells you which one you are looking at.
How new tokens are handled
Tokens that launch through a launchpad pass through a lifecycle, and where a token sits in that lifecycle changes how much real behavior there is to read. TrustGate marks this with astate field on batch and discovery responses.
mining
mining
The early phase, before a token has a real open market. On the current batch path this is a holder count too thin to be a market. There is activity, but not the kind of genuine trading behavior a full read depends on.Scores at this stage are necessarily provisional. The number and tier describe the deployer wallet, not the token. Show it as deployer trust, clearly labeled. Confidence is capped.
graduated
graduated
The token has a real holder market. Now there is market behavior to read: who is buying, how it trades, who holds it. This is where a token score carries its full weight.
TrustBadge in the discovery kit handles both states on its own. You do not branch.
Snapshots
Oracle and batch scoring record a point-in-time snapshot (snapshotId, scoreStability, directionDrivers). You can list recent snapshots at GET /api/snapshots/{address}.
That completes the scoring core
The products
Where the score becomes something you can actually use.
Back to the overview
Revisit the concept now that the full picture is in place.