What the research concluded, in the report’s own words: the four things that decide whether a community happens, what follows from them, and the six rules that transfer where the numbers do not.
A community can aim for partial self-sufficiency and supply its own energy, water and food. But which technologies to apply depends on site-specific conditions.
Neobiome Intelligence: a verified knowledge base, a data fusion model, and an engine that sizes and prices a community at any New Zealand coordinate.
The technology is not the constraint. Where you build, how it is financed, what the law permits and whether the people stay are what decide the outcome.
How might emerging disruptive technologies empower the development of decentralised, sustainable, self-sufficient and innovative eco-communities in Aotearoa New Zealand?
This research answered that by building the instrument that was missing and testing it at real places.
Global systems that deliver food, energy and goods have proven fragile, and extreme weather now ranks first among global risks on a ten-year horizon. New Zealand is not exempt: one dry winter in 2024 sent hydro generation to its lowest level since 2013, and fossil generation had to cover it. Urbanisation is pushing housing beyond reach, with Auckland’s median house price at 7.7 times median income. Meanwhile the convergence of AI and robotics is drastically reshaping what is possible, and it can either concentrate control further or empower individuals. Which way it cuts is a design choice.
Eco-villages, papakāinga and off-grid housing all exist, and the literature reports what particular communities achieved. What none of it reports is what the same community would have achieved somewhere else, because nothing existed to design with. Of the studies reviewed, none combined a specific location, more than one resource, capital and operating cost, and a test against a constrained year.
Rather than study the question at a distance, I built the instrument that would answer it. Neobiome Intelligence has three components: (1) A knowledge system of more than five hundred individually verified sources, inspired by the LLM-wiki concept, under a rule that nothing enters unless the primary document is held and checked. (2) A geospatial layer fusing more than twenty datasets, from solar radiation and river flows to soils, hazards and grid distance, into a site profile for any New Zealand coordinate. (3) A calculation engine that turns that site and community profile into a self-sufficiency result across energy, water and food, an indicative build-and-run cost, and a pass or fail on the tests that decide the matter. Every parameter carries its source and its status, so a reader can see what is verified and what is assumed.
The research also drew on a survey of 22 prospective residents, eight expert interviews, a site visit, nine documented cases and an analysis of the regulatory framework. The findings come from simulating five real New Zealand coordinates with an identical thirty-household community held constant at each.
The research mapped 161 options across ten technology domains, of which 123 are endorsed as viable at community scale. The engine then selects the optimal setup for the site conditions, from solar panels, batteries, greywater recycling and a biomass boiler where the winter is coldest. Nothing on that list is emerging. The exclusions fall on scale and dependency rather than on capability.
The same simulated community, run at five coordinates, produced self-sufficiency ranging from complete to effectively zero. The ordinary year flatters and the constrained year decides. Three of the four sites with buildable ground failed on a dry-summer water shortfall no annual figure reveals, two of them while scoring full marks on every domain. No decision available to a community matters as much as which land it stands on.
At the four sites modelled, the self-sufficiency systems alone came to roughly $80,000 to $100,000 per household on top of land, housing and construction, indicative to about ±20 per cent. The survey indicated that prospective residents will accept capital cost and resist monthly cost, which suits the economics of self-sufficiency, but the people who have developed such projects named access to patient capital and financial literacy as the binding barrier.
A decentralised community is assembled activity by activity from nine regimes. The only purpose-built pathway, papakāinga, is available solely on Māori ancestral land. Nothing in the system promotes housing combined with food production on the land it occupies.
Shared systems concentrate work in a minority, tax the people they serve, turn proximity into exposure under shock, and trade individual autonomy for collective process. Nine conditions emerged from the evidence. Every practitioner who reflected on why such projects fail located the failure in trust, people, finance or upkeep, and never in the equipment.
These are what the project’s evidence supports, not an exhaustive list. Five are about holding a community together, and four are about staying resilient.
Shared systems are sustained by a small core, and free-riding grows with community size. In the project’s own survey a shared workshop ranked sixth of twenty-two facilities and was chosen only by people who already wanted hands-on involvement, where community gardens drew support broadly.
Across six New Zealand cohousing communities, greywater systems were consistently identified as failure-prone, with occupant maintenance a design requirement. In household digestion trials 42 per cent found assembly difficult and both models produced occasional odour.
Under COVID one community had to act as individual households: the common house closed, communal meals ended, and several older residents moved out as proximity became a hazard.
A review of thirty-five studies of cooperative housing reports slow decisions where consensus is not reached, and a perceived loss of individual autonomy when personal preferences are subsumed, drawing those findings from particular underlying studies rather than synthesising them across all thirty-five.
The technical experts interviewed were optimistic; a resident described his own community’s attitude to new technology as wary and negative, splitting by age. Household water treatment attracts the same scepticism as no treatment at all.
One community’s rainwater tanks run dry several times each summer and the city supply opens automatically; that connection is what makes the modest tank sizing work at all.
Digesters non-functional over six months in a cold climate; winter anticyclones suppressing generation for days; streams lowest exactly when irrigation demand peaks.
Three countries hold at least 70 per cent of manufacturing capacity for each of wind, batteries, solar and heat pumps. At community scale the same exposure appears as a single supplier going bankrupt.
A community sensor network was built precisely because the council’s cloud dashboard is unreachable when the bridges are down.
Cutting the grid connection raises capital by between a fifth and three fifths, doubles to trebles the cost of electricity, and adds nothing at all in a normal year. At a cold southern site it buys 0.12 of bad-year self-sufficiency for about $1.4 million. The research names this ‘diminishing marginal self-sufficiency’: a community should buy increments until the next one costs more than importing the shortfall it covers, and no further. All three practitioners who addressed self-sufficiency as a goal argued against it.
The proposition the evidence supports is ‘designed interdependence’ rather than independence: a community that supplies most of its own needs and keeps deliberate connections for the rest, knowing which domain it depends on, in which season, and at what cost. That is more achievable than the phrase ‘self-sufficient community’ usually implies, and it is only a design decision if someone can compute it. A connection is worth leaning on only while it holds, and the hard year decides.
The condition travels with the conclusion. Interdependence is the better proposition for as long as the systems a community leans on are dependable. This research’s own finding is that the stressed year decides the outcome, and a grid connection is a stressed-year asset whose worth depends on it holding in exactly the conditions that stress it. So autonomy is poor value against a dependable grid, not poor value as such. Where the connection is weak, absent, or expected to fail when it is most needed, the arithmetic changes.
The founder of the best-documented community argues for staying connected to town rather than being self-sufficient in isolation, and a fully electric orchard keeps its grid connection because disconnecting would forfeit export revenue worth more than the connection costs.
And what a community leans on need not be the national system. Between the single community and national infrastructure sits an intermediate scale this research did not model but its evidence keeps pointing at: communities covering one another.
Four things, sized deliberately. The instrument itself, a site-specific, transparently sourced calculation of partial self-sufficiency for any New Zealand coordinate. Six working rules for anyone assessing site-specific self-sufficiency, which travel where the numbers do not. A demonstration that one researcher, using the technologies under study, can build and discipline an evidence base that would otherwise need a team. And an observation about the data environment: New Zealand’s open spatial data is sufficient to support a coordinate-level national instrument.
These are not findings about New Zealand. They are what building the tool established about how this kind of assessment has to be done, and they transfer where the numbers do not.
Every rule below is a way of looking harder. This one is a way of looking from further away, and it comes first because it costs least. Put the crude question to any output before putting a careful one: could a community of that size really run on that much generation, could that much land really feed that many people, could roofs of that area really carry a year of demand. The question needs no knowledge of how the instrument works and it takes seconds. It is also the only check that gets weaker the closer the reader is to the build, which is why it has to be asked deliberately rather than left to whoever knows the model best. The fat shortfall behind Rule 5 was found exactly this way, by disbelieving a result that every internal test had passed.
The instrument computes continuous ratios, and not one of them decides anything. What decides is a small set of pass-or-fail tests, and those had to be written deliberately.
Food output moves across a defensible range of dietary assumptions, so the honest output is the range rather than its midpoint. A single number implies a confidence the working does not support.
Three errors here survived review by whoever made them, and were caught when a second party re-ran the calculation from the same figures. Two independent paths to any published number is cheap, and it is the only thing that worked.
Food demand computed as calories returned communities reading 1.00 while no adequate fat supply existed. Resolving demand into the three macronutrients changed which designs passed. It was found by disbelieving a good result.
A model that reproduces its own earlier outputs to the dollar has shown that it is stable, and nothing about whether it is right.
It is a proof of concept built on public data, so its outputs are indicative rather than survey-grade, and five sites are too few to characterise a country. The legal reading is of the enacted instruments rather than of practice. The most important limit is that it measures resource flows and not community wellbeing. The hardest part of building a community is people, and an instrument that assesses places cannot help with that.