Research concept

Project Fusion

A coordination layer for people to pool compute toward a single hard problem: maintaining plasma stability in a fusion reactor.

The idea sits on top of Turing-complete computation over payment channels — contributors precompute control policies in parallel, the network picks the best ones, and an on-site executor runs them at microsecond latency.

Three layers

L1

Decentralised precomputation

Contributors train and optimise magnetic-control policies against simulated plasma states. Payment channels settle who contributed what, in proportion to the quality of their results.

L2

Signed policy registry

Versioned, content-addressed policies published by L1. Each policy declares the reactor, operating regime, and the input domain it's valid for. Signatures are verified on publish and on load.

L3

Local fast executor (simulated)

The hot loop. Takes plasma sensor state in, emits coil setpoints out, using the currently deployed policy. Out-of-distribution states fall back to a safe hold — never extrapolated.

Honest framing. Decentralised consensus cannot close a tokamak's microsecond control loop — that's a physics limit, not an engineering one. What it can do is feed that fast local loop a policy optimised across orders of magnitude more compute than any single lab could muster. Project Fusion is about the coordination layer, not the hardware. Everything reactor-facing here is strictly simulation.