Three things move through this house that most people file in separate folders: electricity, water, and what I know. They arrive from somewhere else, get used once, and leave. The utility bills cover two of them. Nothing covers the third.
The argument of this page is that they’re the same kind of thing — that they follow the same laws, fail the same ways, and reward the same discipline. Everything else on this site is a working demonstration of that claim at the smallest scale I could make honest.
I. Resilience is stored quantity and slow release
Nothing on this property is resilient because it’s clever. The things that survive a bad night survive because something was holding a reserve.
Look at what that actually means in each domain, because the physics is nearly identical:
Heat is stored in water. A gallon weighs 8.34 lb and holds a BTU per pound per degree. A 300-gallon tank is 2,500 BTU per °F — the aquaponics page works this out — and a tank that size stores as much thermal energy as several feet of concrete. The Bolivian walipini design in the archive leans on the same trick with dirt: dig below the frost line, where the ground sits at 50–55°F all year, and the earth walls become a battery that charges by day and discharges at night. The claimed result is a growing space at 50–70°F when the outside air is at −20°F, with no heater at all.
Water is stored in soil. A one percent increase in soil organic matter holds something like twenty thousand additional gallons per acre. That’s not a metaphor for resilience, it’s a reservoir — the thing that decides whether a dry August is an inconvenience or a loss.
Charge is stored in batteries, and the interesting number in any off-grid design was never the panel rating. It’s the hours of autonomy.
Context is stored in memory. The brain behind this site holds ~80,000 captures and has been growing since late March 2026. Its entire value is that a conversation doesn’t start at zero — that what I worked out in April is still load-bearing in August.
Same shape every time: a reservoir, and a rate. And the failure mode is always the same too. A 30-gallon tank kills its fish overnight from an accident a 300-gallon tank absorbs. A bare, compacted field floods and then droughts from the same week of weather that a living soil handles without comment. A system with no battery goes dark the instant the grid does. And an AI with no memory re-learns your situation every single time you talk to it, forever.
Fragility is not the presence of danger. It’s the absence of storage. Danger is constant. What varies is whether you built a buffer, and a buffer is only ever one thing dressed up in different units: time to react. That’s what you’re actually buying with thermal mass, with soil carbon, with a battery bank, with a corpus of captures. Hours between the thing going wrong and the thing being unrecoverable.
II. Waste is a resource at the wrong point in the cycle
The conventional design treats housing, food, energy, and water as four problems with four vendors. The water model is the clearest: import it clean, use it once, export it dirty. A straight line with a purchase at one end and a problem at the other.
Cycles behave differently. Fish waste is a plant nutrient. Plant roots are a water treatment plant. An inch of rain on a thousand square feet of roof is 600 gallons that either goes down a storm drain or goes into a tank. Shower water is an irrigation input the moment you stop calling it sewage. Aquaponics uses on the order of 90% less water than soil farming for the same crop, not through efficiency in any heroic sense, but because the water never leaves — it just goes around again. Nothing in that list is an invention. It’s the same water and the same nitrogen, routed so the output of each step is the input of the next.
Once you’ve seen the pattern in water you can’t stop seeing it. Here is the same diagram drawn over knowledge:
The context you built last week is gone, and whatever it learned about your domain went to a model you don’t own. The understanding you generate becomes training data for someone else’s product — extracted upward, not accumulated where you stand.
Import, use once, export as waste. It’s the linear water model, applied to what you know. And it produces exactly what the linear water model produces: scarcity at the point of use, and a disposal problem somewhere downstream.
So the through-line of this whole project is one sentence with three nouns in it. Intelligence should accumulate where it’s earned — and so should water, and so should energy. Not because localism is virtuous, but because the linear version is a bad design and the cycle is a better one.
III. The thermodynamics of thinking
There’s a version of this argument that gets made purely about privacy, and it’s the weaker version.
Electricity that goes into a computer comes back out as heat. Essentially all of it — that’s not an efficiency claim, it’s conservation of energy. The only question is where the heat lands and what it costs to move it.
In a datacenter, that heat is a liability. It must be removed, and removing it takes more power and, in many designs, evaporated water. The public figures for exactly how much are contested, and I haven’t measured any of them, so I won’t quote numbers I can’t defend. But the direction is not in dispute: large-scale inference is an energy problem and a water problem wearing a software costume.
In a garage in New England, in February, the same watts are a space heater that happens to think. The waste heat is the point. A reasoning box next to a fish tank is a thermal input to a system that already wants one.
This is why the sovereignty argument on this site isn’t only about who owns your data. Keeping the reasoning local puts the energy where its byproduct is useful, in the season when it’s useful, next to the mass that stores it. The privacy case and the thermodynamic case point the same direction, which is usually a sign a design is right rather than merely preferred.
IV. Measure it, or don’t claim it
Every domain in this document is polluted with aspirational numbers. Regenerative agriculture, off-grid energy, aquaponics, and AI are four fields where it is extremely easy to publish a rendering and call it a result.
So the epistemic rule here is narrow and strict: the number gets published, including when it’s unflattering. The Live page shows an honest offline state rather than a projection, because a sensor that isn’t installed yet should read as absent, not as zero. When I measured whether my own memory system degrades as it grows, the answer was yes — the gap between the right memory and the most convincing wrong one closed from 0.090 to 0.011 as the corpus grew — and that got written up in full, because a result you only publish when it flatters you isn’t a result.
The corollary is that the beautiful things get labeled. The village designs in the archive quote 850 kW of solar against 500 kW of demand, 72 hours of battery, net-zero water discharge. Those are design targets from an unbuilt concept, and the archive says so on every page: preserved, not maintained. Still interesting. Mostly still unbuilt. Leaving that label on is not modesty. It’s the only thing that makes the measured numbers elsewhere worth anything.
V. Build the smallest honest version
The gap between the village renderings and a 30-gallon tank with three goldfish in it is the most useful thing on this website.
Anyone can design the village. Designing the village is free, and it is the single most reliable way to never build anything, because at village scale every mistake is hypothetical and therefore costless and therefore uncorrected. The 30-gallon tank is worse in every way except the one that matters: it is real, so it tells you when you’re wrong, and it tells you cheaply. That is what a pilot is for. Not proving the idea works — finding out what questions to ask before the expensive version gets built.
One house — a basement, a backyard, a garden — is the smallest testable version of the thing the archive is drawing at community scale. Three loops, one shared memory. If the pattern is real it should be visible there, and if it isn’t visible there it was never going to survive being scaled up.
VI. Where this goes
Take the four principles seriously and the future they describe isn’t a bunker and it isn’t a datacenter.
The off-grid instinct gets the goal wrong. The point was never disconnection — it’s net contribution: a place that generates more than it consumes and exports the surplus, that retains its own water rather than shedding it into someone else’s flood, that holds enough reserve to ride out an outage without needing to be rescued. A node like that isn’t withdrawing from the grid. It’s the reason the grid holds.
The same shape works for knowledge, and this is the part I’m actually building. Each node keeps its own raw material — its sensor readings, its records, the specifics of its situation — and shares only what it concluded. The brains trade the pattern, never the data. A beekeeper’s hive node says swarm activity rising, north end; it does not ship its owner’s logs anywhere. What crosses the wire is the thing that was learned, at a fraction of the size and none of the exposure.
That’s the same architecture as a watershed, where every property retains its own water and what moves downstream is clean. It’s the same architecture as a distributed grid, where every roof holds its own storage and what moves across the wires is surplus. Sovereign at each node, stronger as a network, and in each case the thing that travels is the refined product rather than the raw extraction.
Whether that’s a good bet is not yet settled, and I’d rather be honest about the state of the evidence than persuasive about the vision. What exists today is one brain with 80,000 captures, one small tank, a colony, a raised bed, and an instrumented build going into the garage. The interesting claim — two sovereign nodes, different owners, one signed crossing between them — is a winter project, and it either works or it doesn’t.
The bet is that it does, and that the reason it does is boring: stored things compound, extracted things don’t, and this is as true of what you know as it is of the water in the ground.
The architecture is on HKOS. The physical loop is on Aquaponics and running on Live. The unbuilt versions, labeled as such, are in the Archive.