SOLYSTOPIAProject Hearthmind, framework
Computational subsidiarity
Run intelligence at the smallest scale capable of performing the task competently. Escalate to a larger scale only when the smaller one genuinely lacks the capability — never by default.
This is the same argument Solystopia makes about food, energy, and water, applied to intelligence: sovereignty is not all-or-nothing, it is a ladder of scales, each one justified only by the failure of the one below it.
A household that routes every query to a national cloud by default has made an architectural decision, even if it never felt like one. Computational subsidiarity makes that decision explicit and reversible: start small, measure what the small scale cannot do, and escalate only that specific gap — not the whole workload.
The ladder
Five tiers. Tasks climb only when they must.
One device, one person. Small local models on a phone or laptop.
Escalates when a task needs more memory, more context, or specialist models than the personal device holds.
One Spark or equivalent, shared by a household. The Hearthmind baseline.
Escalates when compute, storage, or model capability outgrows what one household can justify owning.
A shared node serving several households — the Ward AI Centre model.
Escalates when demand or capability requirements exceed what a neighbourhood-scale node can sustain.
Shared infrastructure serving a town — larger models, larger storage, specialist hardware.
Escalates only for capability that is genuinely uneconomic to hold below national scale.
Public cloud or national research compute — the last resort, not the default.
Nothing above this tier in the Hearthmind framework; used deliberately, not by default.
Applied, not theoretical
Project Hearthmind lives at tier 02 by design.
The DGX Spark is deliberately a household-scale machine, not a neighbourhood-scale one. When a task genuinely exceeds what it can do — a model too large to fit in 128GB, a workload needing sustained multi-GPU throughput — the honest move is not to buy a bigger machine reflexively, but to ask whether that capability belongs at the neighbourhood tier instead. That is exactly the question Infrastructure 001, the Ward AI Centre, exists to answer.
See the neighbourhood tier in practice: the Ward AI Centre