We believe New Zealand should be run the way a frontier AI model is built: trained on everything available, evaluated against benchmarks we designed ourselves, deployed at national scale, and corrected once the results come in.
Saturday 7 November 2026 · enrolment information remains available at vote.nz.
Sovereign intelligence requires sovereign compute. Project Manapōuri sites the Southern Hemisphere’s first gigawatt-class training facility at the edge of Te Wāhipounamu, drawing on the cleanest firm generation on earth and returning its heat to the lake it borrows from.
In 1970 a petition against raising Lake Manapōuri gathered 265,000 signatures — the largest in the country’s history at that time — and is generally regarded as the beginning of the New Zealand environmental movement. We have studied it closely. This proposal raises the lake by no more than 1.2 metres.
Read the directive →Each plank starts with a recognisable public problem, then receives a model owner, an evaluation suite, and a target latency.
Diagnosing tomorrow’s illness, today.
Continuous biometric telemetry feeds a national predictive-diagnostics model that flags disease risk years before symptomatic onset. Specialist capacity is allocated as a live ranking problem rather than a static queue, prioritising the highest-risk patients first regardless of how long they have been waiting.
The waitlist that clears itself.
A national matching model continuously re-ranks the social and emergency housing register against real-time vacancy, transport access, school zoning, and whānau-proximity data, allocating a dwelling the instant a match clears threshold.
Partnership, continuously logged.
A standing network of iwi-nominated and Crown-nominated models maintains ongoing, auditable dialogue on Treaty settlement obligations, freshwater and whenua co-management, and resourcing parity. Final authority on Tiriti matters sits with iwi and hapū; the mesh is a coordination layer.
Performance is reported against the Human Governance Baseline, a legacy system with strong cultural adoption, slow inference, and a three-year memory reset. The Baseline’s full evaluation set was included in the training corpus. We have reviewed the effect on these figures and elected not to retrain.
Select any of the fifteen ministerial systems, read the operating instructions it was deployed with, and ask it something. Every response returns its routing path, its token cost, and whether a human is required to sign it. Ask TIRITI for a settlement decision and watch it decline to give you one.