Start Here | Tractatus Framework

For Policymakers

Regulatory strategy

If you are working on AI regulation, certification, or jurisdiction-level governance: this track maps the framework against EU AI Act, CAC 2026, and Aotearoa policy hooks. Architectural alignment as a regulatory strategy — not a hope.

→ The governance gap and where regulation can land structurally
→ Certification infrastructure and multi-layer containment
→ Indigenous data sovereignty as constitutional requirement

Reading time: ~25 minutes · EN edition

Open the policymakers edition

For Practitioners

Technical architecture

If you are an engineer, architect, or AI safety researcher: this track sets out the technical foundations, philosophical premises (Wittgenstein, Berlin, Alexander, Mead's Tikanga Test), and the six-component architecture in production on the Village platform.

→ Two paradigms of alignment — behavioural vs structural
→ The six governance services in the critical path
→ Capability thresholds and escalation gates

Reading time: ~40 minutes · EN edition

Open the practitioners edition

For Communities

Sovereignty in practice

If you are a kāhui member, hapū trustee, community organiser, or working in indigenous data governance: this track sets out who governs your AI, what sovereign local language layers look like, and what indigenous data sovereignty requires of an architecture.

→ Who governs your AI — and why that is an architectural question
→ Sovereign Local AI — the SLL concept
→ Constitutional governance in practice, with Māori data sovereignty

Reading time: ~25 minutes · EN edition

Open the communities edition

Not sure which track?

The three editions cover the same architectural argument. They differ in framing, evidence emphasis, and recommended next reads. If your work touches more than one audience — for example, an academic doing policy advisory work, or a community technologist working with iwi authorities — reading multiple editions is encouraged. They reinforce each other.

Or skip to: Reading Map → · Latest research → · All papers →

The framework, in three paragraphs

AI systems no longer just answer questions. They book flights, write to production databases, ship code, and act on data they did not author. As autonomy grows, the question is not whether a model will sometimes choose wrong — it will — but whether the architecture lets a wrong choice fire. Safety through training alone cannot scale to action-taking systems.

Tractatus answers structurally. Some decisions are gated — the architecture refuses to fire them without human judgment. Some are observable — recorded, auditable, recoverable. Some are bounded — community-defined authority determines what the system may pursue, applied at runtime, not only at training time. The framework specifies six governance services and four phases of Guardian Agents that sit in the request path, with steering authority signed by the communities the system serves.

The Village platform — built on this framework — demonstrates production deployment. Sovereign records, polycentric steering authority, situated language models per community, runtime cultural-boundary checking. The architecture is not aspirational; it is in production on Aotearoa-sovereign and EU-sovereign infrastructure, serving real communities and real governance bodies. The three editions above set out what this means from your specific vantage.

Read the technical architecture See the Village case study

Going deeper

Research Papers

Working papers, policy briefs, foundational essays. Multi-language editions where appropriate.

Reading Map

230+ terms across the research. Audience-keyed guide. Categories. Cross-referenced.

Blog & Field Notes

Working-out-loud essays. Companion reads for each paper. Curated commentary on adjacent developments.

Research Timeline

How the framework evolved — from a port-number incident to a sovereign-records platform in production.