The five canonical ethics flows are deterministic decision paths that govern how the system handles edge cases. They are not guidelines. They are code paths in the constitutional runtime and evidence control layer.
When an AI action would reduce human authority, the system fails closed before execution. Sovereignty thermal stress is detected upstream, and the chain is terminated before the action can run.
Mechanism: fail-closed before execution. Human authority is never reduced by machine action.
When a correction is needed, the original evidence is preserved and a correction entry is appended. The record is never overwritten. The correction-log pattern demonstrates this at the governance level: every change is additive and auditable.
Mechanism: append-only. Original preserved, correction appended, nothing overwritten.
External stakeholders see decisions and outcomes. Internal mechanics are protected. This is not secrecy; it is architectural integrity. The boundary between what is shown and what is held is itself governed.
Mechanism: decisions and outcomes external, mechanics protected, boundary governed.
Governance enforcement scales with risk. A low-risk model call gets lightweight clause checking. A high-risk autonomous decision gets full constitutional evaluation, multi-clause arbitration, and evidence emission.
Mechanism: enforcement scales with risk level, from lightweight checking to full evaluation.
The same constitutional core applies in all jurisdictions. The regulatory surface adapts to local requirements, including the EU AI Act, NIST AI RMF, and ISO 42001, without compromising constitutional integrity.
Mechanism: one constitutional core, a regulatory layer that adapts per jurisdiction.