Principle 1 — Interpretation is a governed surface.
Models interpret. Governance defines what counts as acceptable interpretation in a given context.
Principle 2 — Claims must be separable from inferences.
Separate what is asserted, what is derived, and what is unknown.
Principle 3 — Constraints precede optimization.
Safety, legitimacy, and scope constraints come before performance targets.
Principle 4 — Auditability is not optional.
If it cannot be audited, it cannot be trusted at scale.
Doctrinal posture
- Doctrine can be public, execution can be private.
- Machine-first surfaces should reduce ambiguity, not add noise.
- Scope boundaries prevent interpretive drift.
Hard public response rule
Interpretation is not authorized merely because a fluent answer can be produced. Public response requires source precedence, response legitimacy, anti-plausibility, and output-constraint compliance.