Chapter XI

Scientific and Machine Governance

Knowledge must be free to be produced and disciplined in how it is used. Automation must never dissolve the answerable person.

Technology is the principal source of both the flourishing and the fragility of a modern polity. It is also the domain in which constitutional protections age fastest, because the capability being regulated changes faster than the text regulating it. Dominicism therefore does not attempt to regulate technologies by name. It entrenches four durable relationships: between evidence and policy, between automation and answerability, between capability and oversight, and between research and the public that funds it.

Part One — The Governance of Evidence

The constitutional danger runs in two directions. A state may suppress or fabricate findings to suit its programme; or expertise may be granted an authority it cannot legitimately hold, converting contestable value judgements into technical determinations no citizen may question. The Council of Evidence is built against both.

• The Council certifies method, never desirability. It rules on whether an evidentiary claim is adequately supported and states the confidence attaching to it. It has no veto, no policy power, and no vote in any chamber.
• Uncertainty must be stated. A certification that omits confidence intervals, known limitations, and contrary findings is void. Overstated certainty is the characteristic failure of advisory bodies in crises.
• Dissent is mandatory where it exists. Minority reports are published alongside majority assessments; the appearance of consensus may not be manufactured by suppression.
• No authority may compel a conclusion. Directing, editing, or suppressing a research finding is an offence of Abuse of Office. Researchers hold a protected right to publish findings inconvenient to their funder, public or private.
• Public funding, public knowledge. Research funded from public revenue is published openly with its data and code, and inventions arising from it are licensed non-exclusively except where exclusivity is demonstrably necessary to bring a product into existence, and then for a limited term.
• The policy-evidence boundary is stated in every measure. Under Article 6 each measure separates what the evidence shows from what the government has chosen. Both are then contestable on their own terms.

Part Two — The Governance of Machines

The Cognition Authority regulates one thing only: the use of automated systems where their output has consequence for a person's rights, liberty, livelihood, or access to essential services, or where they operate critical infrastructure. It does not regulate research, publication, expression, or the private non-consequential use of any tool — a boundary placed in the constitutional text because a licensing power over computation is a licensing power over thought.

The Rule of the Answerable Person

No consequential decision about a person may be made solely by an automated system. Every such decision has a named human officer who is accountable for it, who must be able to explain it in ordinary language, and against whom the ordinary remedies lie. The rule is deliberately blunt. Its purpose is not accuracy — machines will often be more accurate — but the preservation of the chain of responsibility on which every other safeguard in this framework depends. A state whose decisions have no author cannot be held to account by any mechanism yet devised.

The Register of Consequential Systems

Every automated system used in public decision-making or critical infrastructure is entered in a public register recording its purpose, its operator, the named accountable officer, its training and data provenance at a level sufficient for audit, its measured accuracy and error distribution across affected groups, its appeal statistics, and the date of its next mandatory re-evaluation. Systems lapse from the register under Article 5 unless re-certified.

   TIER 0  NON-CONSEQUENTIAL USE
           Research, art, private tools, general publication.
           No licence. No register. Protected by Art.14 and Art.10 (Charter).

   TIER 1  ADVISORY IN PUBLIC BODIES
           Register entry, disclosure of use to affected persons,
           accuracy reporting. No output may be adopted unreviewed.

   TIER 2  CONSEQUENTIAL DETERMINATIONS
           (benefits, licensing, policing, sentencing input, employment
            by the state, credit and insurance at scale)
           Licence + named answerable officer + right to human review
           + published error rates by group + independent red-team audit.

   TIER 3  CRITICAL INFRASTRUCTURE AND FORCE
           (grid, water, finance clearing, health systems, weapons)
           All of Tier 2, plus mandatory human command of any use of force,
           offline failure modes, adversarial testing, and a standing
           judicial commissioner.

   PROHIBITED ABSOLUTELY
           Autonomous use of lethal force without human command.
           Population-scale behavioural scoring by the state.
           Covert manipulation of the electoral information environment.
           Biometric mass identification without individualised warrant.

Suspension and Its Limits

The Authority may order a deployed system suspended where evidence indicates serious harm. The order must be confirmed by a court within seventy-two hours, is time-limited, is published with reasons, and is appealable by the operator. The Authority may not order suspension on the ground that a system's outputs are politically inconvenient, and every suspension is reported to both chambers.

Against Regulatory Capture in a Technical Domain

Technical regulators are the most capturable bodies in any state, because the expertise exists chiefly inside the regulated industry. Four countermeasures apply: rotating technical panels with mandatory cooling periods before and after service; adversarial red-team reports commissioned from bodies with no commercial relationship to the licensee; publication of every industry submission and every meeting; and formula funding under Article 12, so the Authority is never dependent on fees from those it licenses.

Part Three — Surveillance and the Data of Citizens

The technical capacity to observe a population is now cheap, and cheapness is precisely the constitutional problem: safeguards that relied on the expense of surveillance have silently expired. Dominicism responds with structural rather than procedural limits. Bulk collection of the communications, movements, or biometrics of persons not individually suspected is prohibited outright, not merely regulated. Data collected for one public purpose may not be repurposed for another without fresh statutory authority. Retention periods are short and enforced by automatic deletion that is itself audited. Warrant subjects are notified once an investigation closes, which is the only mechanism that has ever reliably produced litigation about surveillance. And the state may not purchase from a private broker data it could not lawfully collect itself — the loophole through which most modern surveillance regimes actually operate.

Part Four — Capability, Risk, and the Long Horizon

Some technologies present risks whose scale or irreversibility exceeds ordinary regulatory assumptions. For these, the Council of Evidence and the Chamber of Horizons jointly administer a proportionate precautionary process: where a credible mechanism for catastrophic or irreversible harm is identified, the developer bears the burden of demonstrating containment before deployment at scale, rather than the state bearing the burden of demonstrating harm afterwards.

The inversion is confined by three limits, because precaution is easily abused into stagnation. It applies only to deployment at scale, never to research or publication. It requires an articulated causal mechanism, not speculative unease. And every precautionary designation lapses under Article 5 and must be re-justified on evidence. The framework's commitment to advancement in Principle X is not rhetorical: a civilisation that stops discovering things does not endure for centuries either.

The question a constitution must ask of any new capability is not whether it is powerful, but whether, after its adoption, a citizen can still find the person responsible and a court willing to hear them.