The pilot rejects classes of failure such as underclassification, fake role separation, invalid evidence, ceremonial reviews, missing participation, invalid decisions, lifecycle relabeling, and semantically empty monitoring.
Proposed architecture · N0
From a public question to an accountable decision record
Separate what is known, what is contested, and who is allowed to decide.
NousPolis is designed to expand analytical capacity without collapsing research, deliberation, safeguards, and authorization into one opaque AI output. Each stage should create records that later stages can inspect and challenge.
Current boundary: the process below is an institutional design. Parts are now represented in an experimental machine-testable PDLS pilot, but NousPolis remains N0 and has no public-policy authority.
Reader-facing map
The canonical seven-stage explanation
Seven stages keep the path from concern to consequence inspectable.
This is the plain-language map. The executable PDLS pilot below uses a more granular state machine and validation gates, but it operationalizes the same constitutional separation rather than creating a second governance system.
1. Raise and register the question
Record the public problem, jurisdiction, mandate, requested action, source of the question, and relevant local knowledge. The agenda itself is part of the record.
2. Define the problem and affected people
Document competing framings, affected communities, rights, institutional constraints, classifications, and the consequences of doing nothing.
3. Gather and verify evidence
Distinguish sources from claims. Track provenance, methodology, uncertainty, contradictory evidence, and the specific propositions each source actually supports.
4. Compare and challenge options
Analyze the status quo and materially distinct alternatives. Use procedurally separated analytical passes and adversarial review to expose unsupported assumptions, omitted stakeholders, alternative causal models, rights risks, and failure modes.
5. Deliberate and publish separate outputs
Keep established findings, contested findings, uncertainty, distributional effects, normative choices, feasibility constraints, recommendations, and material dissent visibly distinct.
6. Safeguard review and human authorization
Review rights, legality, due process, equality, privacy, democratic integrity, conflicts, delegated authority, and AI risks. A legally responsible human authority may accept, reject, or modify the recommendation, with reasons recorded.
7. Implement, measure, challenge, and reopen
Link the chosen action to responsibility, milestones, causal assumptions, outcome measures, guardrails, and reopening conditions. New evidence, serious harm, rights breaches, implementation failure, or scheduled review can begin a new analytical cycle without erasing history.
Today · experimental N0 implementation
Policy Deliberation and Learning Standard
The architecture now has a machine-testable policy-process pilot.
NP-PDLS-001 v0.1 turns a policy question into a structured record with typed artifacts, lifecycle states, validation gates, conditional modules, adversarial findings, safeguard review, a human decision record, implementation trace, monitoring, and reopening logic.
The hardened pilot was merged on 17 August 2026 after independent adversarial re-review. Its suite includes one positive reference case and 26 adversarial regression tests. The validator also checks governing-corpus coverage.
The pilot is an N0 semantic conformance instrument. It does not prove external independence, real-world evidence, policy correctness, legal authority, or authorization for action.
Policy-process lenses inside the workflow
NousPolis uses established frameworks as diagnostic questions—not as rival constitutions.
The PDLS pilot operationalizes several policy-process traditions at the point where their questions are useful.
IAD
Who holds authority, which positions exist, what rules-in-use apply, what information and incentives shape action, and where are the vetoes?
Advocacy Coalition Framework
Which stakeholders and coalitions exist, what beliefs and resources organize them, and where are conflict or convergence likely?
Punctuated Equilibrium
How does the severity of a problem compare with political, administrative, and public attention?
Multiple Streams
Are problem recognition, solution readiness, and political or administrative readiness aligned?
Narrative analysis
Which materially competing frames and causal stories could change how the problem is understood, especially under polarization or manipulation?
Policy feedback
How might implementation change behavior, institutions, power, trust, constituencies, path dependence, and future policy options?
What travels through the system
Different records need different protections.
Citizen testimony, public evidence, analytical records, authority records, and outcomes carry different duties. Provenance should make later conclusions traceable without exposing protected material.
| Information type | Examples | Main protections |
|---|---|---|
| Public and affected-community input | Issues, testimony, priorities, objections, appeals | Consent, privacy, representation limits, minority preservation, anti-capture controls |
| Evidence and claims | Research, law, data, causal inferences, forecasts | Verification, provenance, claim typing, uncertainty, contradiction visibility |
| Analytical records | Framings, institutional diagnosis, options, assumptions, dissent | Procedural separation, reproducibility, adversarial challenge, audit |
| Safeguard and authority records | Rights review, blocking findings, authorization, rejection, rationale | Separation, legal authority, reasons, accountability, exception controls |
| Implementation and outcomes | Milestones, effects, complaints, new evidence | Predeclared measures, version history, monitoring, appeal, reopening |
Illustrative example
Urban planning
Should a city allow taller housing around railway stations?
This fictional case uses no real jurisdiction, resident data, measured NousPolis results, or actual policy recommendation.
Define competing problems
Housing supply, displacement, accessibility, transport, infrastructure, land value, emissions, and neighborhood continuity may all be defensible framings. Classification determines how much scrutiny is required.
Build evidence and alternatives
Separate source records from analytical claims. Compare the no-action trajectory, the proposed change, and materially distinct alternatives rather than treating one proposal as the default.
Challenge before synthesis
Independent analytical passes and adversarial review test causal assumptions, transferability from other cities, omitted groups, rights risks, implementation constraints, and contradictory evidence.
Human decision and reopening
The city authority records its reasons and accepted trade-offs. Monitoring conditions may later trigger review, but the analytical system does not automatically change policy.
Static comparison
Different governance mechanisms
NousPolis, Futarchy, and Liquid Democracy solve different institutional problems.
They can overlap, but they are not substitutes. NousPolis focuses on the institutional path from evidence and challenge to accountable authorization; Futarchy centers conditional forecasting against an authorized objective; Liquid Democracy centers direct voting and revocable delegation.
| Dimension | NousPolis | Futarchy | Liquid Democracy |
|---|---|---|---|
| Primary purpose | Institutional architecture for AI-assisted public reasoning, safeguards, decision records, and learning. | Decision mechanism using conditional prediction markets against an authorized welfare objective. | Democratic participation through direct voting and revocable delegation. |
| Who ultimately authorizes? | A legitimate human institution under the applicable constitutional and legal rules. | Depends on how the market-based decision rule and objective are constituted. | Citizens and delegates under the adopted voting rules. |
| Evidence provenance | Explicit architectural concern. | Not supplied by the mechanism alone. | Not supplied by the mechanism alone. |
| AI safeguards | Explicit architectural concern, including capability boundaries, review, provenance, and maturity gates. | Not inherent. | Not inherent. |
| Rights, appeals, independence | Intended as separate architectural layers and controls. | Not supplied by prediction markets alone. | Not supplied by delegated voting alone. |
Prediction markets or delegated participation could inform a future NousPolis institution. Neither forecasting nor delegation by itself supplies evidence provenance, safeguard gates, lawful authority, or institutional accountability.
Further reading: Robin Hanson’s Futarchy proposal · Chiara Valsangiacomo on Liquid Democracy · The Principles of LiquidFeedback.
Today
Limits and unresolved questions
What NousPolis has not solved yet
There is no deployed public-policy institution and no public-policy authority. N1 remains blocked until every mandatory N1 enforcement requirement passes and the required transition disposition is made. The experimental PDLS pilot is still controlled within the same repository environment, so it does not supply external roots of trust. Rules for citizen standing, identity, representation, voting, delegation, and ratification remain constitutionally unresolved. Participation infrastructure still needs anti-bot and anti-capture protections, privacy, moderation, accessibility, and due process. Real-world cost, speed, scalability, and performance remain unproven.
Inspect the records
Move from explanation to evidence.
Read Architecture v0.1, the PDLS executable-pilot record, the N1 enforcement requirements, the dated N1 evidence snapshot, the independence rules, and the architecture adversarial-review disposition.