Human oversight in AI-for-science must become structured scientific authorization.
Vague requirements like "human review required" are operationally weak. They fail to specify which expert, which artifact, which authority, which action, which duration, and which residual blocks remain in force.
SCOPE operationalizes review by binding it to:
- Typed reviewer roles
- Inspectable artifacts
- Explicit approval scopes with hierarchy
- Mandatory expiration
- Machine-readable grants
- Measurable review quality
The field-level change: review becomes enforceable, auditable, and improvable.
See scoped_scientific_authorization.md for the SCOPE v0.7 implementation.