The Inquiry and Its Failure Conditions
The first task is to test the premise. SIRDA will publish the argument, invite criticism, and record how the evidence changes it. The result can be a constitutional proposal, a narrower coordination mechanism, an improvement to an existing arrangement, or a reasoned decision to stop pursuing the design. These outcomes are distinct. If an existing or less ambitious arrangement solves the problem more safely, that finding counts against government without territory and is recorded as such. The inquiry remains useful when its hypothesis fails.
Begin with people who can challenge the premise
Initial review will include authors whose work is discussed, technical evaluators, governance researchers, legal scholars, and people concerned with public rights. Contributors can address one mechanism or the entire premise. Publication of responses requires permission, with personal and organizational contributions attributed separately.
Record the method before the experiment
Before a multi-model convention begins, the project will publish its questions, source corpus, selection criteria, prompts, sampling plan, and decision rules. Repeated trials will record configurations and seeds where providers expose them. The record will include variation and deliberate attempts to change conclusions through prompting.
People retain responsibility for interpretation and institutional choices. AI systems assist with drafting, criticism, comparison, and simulation. Agreement among models is a result under documented conditions; its meaning depends on shared assumptions, evidence, and prompt sensitivity. Each run will be logged, with advance rules for privacy, security, and provider terms. Withheld material and redactions will receive recorded reasons.
Publish failure conditions in advance
The following seven conditions organize the proposed testing protocol. Before a design is evaluated, each needs specified scenarios, coalitions, resources, evaluators, evidence, and decision criteria. Changing a criterion creates a new protocol version; earlier results stay attached to the version under which they were produced.
Legitimacy. Test who authorizes each power, how affected people participate, and how determinations become consequential for actors who never joined. A design fails when agreement among the regulated actors is its entire basis for authority, or when its decisions reach outsiders through any route other than powers that states and other parties already lawfully hold.
Measurability. Test varied models, resources, and deployment configurations against declared classification criteria. A design fails when interested parties can shield consequential claims from independent examination or error rates exceed limits set in advance.
Anti-capture. Test whether plausible coalitions can acquire lasting control of standards, evidence, appointments, and implementation. A design fails when its own granted powers allow that coalition to remove meaningful independent challenge.
Continuity. Test withheld funding, appointments, evidence, and access. A design fails when one participant can disable core review and the stated fallback fails to restore a functioning process.
Bounded authority. Test jurisdictional expansion through technical rules, exceptions, and emergency interpretation. A design fails when the body exercising power can enlarge its mandate without effective independent authorization and review.
Due process. Test an erroneous allegation, sensitive evidence, urgent restrictions, and a participant with limited resources. A design fails when a materially affected party lacks a practical route to reasons, challenge, and correction.
Effectiveness. Compare the design under the same scenarios with four comparators: an international standards regime in the style of the Basel Committee, strengthened national regulation, a treaty institution, and voluntary coordination among developers. A design fails when a less ambitious comparator closes the same gap, when it offers no credible risk-reduction mechanism, when it relies on compliance precisely where the scenario assumes defection, or when it gains effectiveness by sacrificing the preceding safeguards.
A failed test applies to the design and assumptions examined. The inquiry must explain whether the result calls for redesign, a narrower mandate, a different approach, or abandonment. Passing a limited exercise warrants further scrutiny, including real-world evidence, legal analysis, and public participation.
Leave a record other people can use
The project should produce clearer disagreements, documented failure modes, and procedures that other researchers can examine. Its value also depends on whether criticism can change SIRDA's own direction. The closing principle therefore applies to the institution being proposed and to the process proposing it: power cannot become the only witness to itself.