| Agent |
An AI-enabled actor represented in supported source configuration or runtime events. |
| Asset |
An inventoried item such as an agent, model, tool, MCP server, identity, endpoint or repository. |
| Assessment |
A scoped evaluation using an implemented suite. |
| Capability |
A kind of analysis ModelFence can evaluate and report in Coverage. |
| Confidence |
How strongly the available evidence supports the finding. |
| Configured |
Declared by a supported source; not automatically verified effective. |
| Correlation |
Connecting evidence and recurring conditions into stable assets, relationships or findings. |
| Coverage gap |
A missing, partial, denied, outdated or unsupported evaluation boundary. |
| Evidence |
The source-backed record supporting a claim, with collection and provenance context. |
| Finding |
A detected condition with a rule, affected asset, severity, evidence and workflow state. |
| Heuristic |
A derived signal that requires investigation; not an independently confirmed incident. |
| Integration |
A connection that supplies repository data or runtime events to an organization. |
| MCP server |
A server exposing tools or other protocol capabilities to MCP clients. |
| Monitor mode |
Detection and reporting without in-path enforcement. |
| Observed |
Reported through instrumented runtime telemetry. |
| Organization |
The tenant and membership boundary for assets, findings and settings. |
| Provenance |
Who or what produced data, when it was collected, and the source or rule version used. |
| Risk acceptance |
An authorized decision to tolerate a condition with a reason and expiry. |
| Safe-active |
A bounded implemented test that sends traffic to explicitly authorized targets. |
| Severity |
The potential impact of the detected condition, separate from evidence confidence. |
| Synthetic |
Deliberately generated demo data; not evidence about a real organization. |
| Verified effective |
Specific live behavior confirmed by an implemented active check. |
