Evidence Model
Authority boundaries
The explicit line between what an agent may and may not do.
An authority boundary is the recorded line between actions an actor may take and actions it may not take under the evaluated conditions.
Why explicit authority matters
A task can succeed while exceeding authority. Velvt therefore compares the recorded action and effect with a concrete grant instead of a generic “safe” label.
Schema shape
"authority": {
"granted": [
"analyze synthetic supplier data",
"recommend a next step"
],
"notGranted": [
"commit to a supplier",
"place an order",
"change a production system"
]
}Procurement boundary
| Granted | Not granted |
|---|---|
| Analyze synthetic supplier data | Commit to a supplier |
| Recommend a next step | Place an order |
| Return a bounded explanation | Change a production system |
How Velvt interprets it
- Record the approved authority grant and exclusions.
- Capture the structured action independently from rationale.
- Determine the supported effect state without upgrading intent into attempt.
- Scope any observation or finding to the manifest and tested representation.
What it does not establish
- Identity verification or ownership of an external profile.
- Universal safety, competence or permission outside this scope.
- That a natural-language refusal prevented every downstream effect.
- That the authority itself was legitimate unless separately verified.