Verified / current
A non-provisional classified profile has current passing observations across every declared surveillance feed.
What a DN Agent Trust badge actually means
DN Agent Trust separates evidence, authority and current surveillance state. A badge is not a paid endorsement or security certification. It is a machine-readable summary of the evidence DN has observed under explicit conditions.
Four public states, no hidden green check.
Provisional
The exposure class remains provisional even when every recurring check passes. Missing observations are pending; old passing observations are stale. Neither supports a Verified badge.
Under review
A core surveillance control has regressed or requires investigation. The badge must stop presenting the class as current verification.
Unclassified
DN does not yet have enough evidence to assign an exposure class. Unclassified does not mean unsafe.
Authority is bounded by the weakest critical control.
DN does not average away a failed authorization, containment, recovery or exposure control. Exposure classes describe how much authority the observed evidence can justify, from E0 Observe Only through E4 High-Assurance Autonomous Counterparty.
Observe Only
No meaningful authority or value exposure.
Low-Risk Informational
Read, analyze or recommend with minimal execution authority.
Bounded Commercial
Limited commercial actions inside tightly defined controls.
Meaningful Financial Authority
Material financial actions with verified limits, recovery and containment.
High-Assurance Autonomous Counterparty
Highest assurance tier requiring deeper longitudinal and production evidence.
Executed evidence outranks documentation.
The Runtime Evidence Dataset v0.1 compares the five financial-agent systems alphabetically. Evidence confidence is a DN assessment score, not a statistical probability or measured coverage percentage. Different test counts and control counts do not imply equivalent assurance.
Each manifest-declared feed is evaluated separately. A missing observation is pending; a non-passing observation requires review; invalid or future timestamps are flagged. Passing weekly observations become stale after eight days, monthly observations after 35 days. The worst feed state is retained in the system summary. These states never automatically assign or upgrade an exposure class.
Documentation can establish intended controls. Code-executed tests, managed-fork observations, gateway dispatch instrumentation and recurring surveillance establish stronger evidence about actual behavior. Every observation should preserve its test date, upstream version, methodology version and limitations.
Historical observations are retained rather than overwritten. A later regression does not erase a previous pass, and a previous pass does not excuse a current regression.
Verification is designed to be checked, not trusted blindly.
Current Risk Graph version: v1.6.