The levels in AgenID v1.1.1
Each level has a fixed claim type and evidence type, and covers only its own claim and scope.
L1_REGISTERED is where every agent starts: a signed declaration the registry has accepted and anyone can re-verify. The step up to L2 is domain control, the one check whose evidence a third party can re-derive for itself. L3 waits on a written evidence standard and L4 on a deployment sampling method that does not exist yet.
| Level | What was checked | Issued today |
|---|---|---|
| DECLARED | An operator signed a manifest; not registered | Yes |
| L1_REGISTERED | The signed manifest is registered; still a self-declaration | Yes, the ceiling |
| L2_DOMAIN_VERIFIED | The operator controls a domain, via DNS | No |
| L3_ORGANIZATION_VERIFIED | The operator's legal entity was reviewed | No |
| L4_DEPLOYMENT_VERIFIED | A deployment was sampled | No; method not yet defined |
| L5 | Reserved name only | Not issuable at all |
Why it matters
Levels make weight explicit. L2's evidence is a DNS record anyone can re-query, so a mistaken L2 is externally detectable, which is why it is planned first. On the Verification Card and badges, L1 renders amber and an unverified agent renders neutral grey, never red, because absence of verification is not a negative finding.
Levels also have time limits. Every assertion carries verified_at and expires_at, so a level describes a window rather than a permanent mark, and editing the manifest makes an assertion inapplicable to the new version.
What it does not mean
No level means compliant, audited, certified or safe. L2 would mean domain control, not trustworthiness. No level says what an agent is allowed to do either; see declared, verified and authorized. A level is also not the agent's status: the status values in the schema are separate, and today every registered agent is ACTIVE.