The arithmetic of self-replication.

One factory lands on the Moon or an asteroid and builds copies of itself from local material. von-neumann models how far that idea holds and where the physics stops it. Each model is small, independently runnable and tested, and every number traces to a cited source.

vn.noahhyden.com →Source on GitHub →

One question per model.

The repo is a monorepo, one directory per model, with no single giant simulation. The models cover closure and replication (how much of itself a factory can build locally), a solar-powered probe, power split between making and thinking, launch economics, a small fleet, and a settlement front across a star field. A cross-scale module threads one factory through the fleet and the galaxy so their replication cadences come from one source. The models run live in the browser on the project site, which is built on pimas.

What the models conclude so far.

The repo’s findings document collects nine sourced results. Its summary: the leverage of self-replication is real, but it is bounded by physical walls that reappear at every scale, and the models say which limit binds where. Three of them:

  • A lone factory can make its structure but not its electronics. Chips are the wall.
  • Installed mass per launched mass is exactly 1/(1 - closure): about 3x at 67% closure, about 33x at 97%.
  • Light-speed coordination costs redundant travel, not fill time. The tens-of-percent fill-time penalty is a coarse-timestep artifact; the real tax scales as v/c.

These are order-of-magnitude research results, not forecasts. Where the literature has no value, the repo marks the gap instead of filling it. The full list is in FINDINGS.md, and the project’s papers are listed in the papers section on the home page.

Cite it with DOI 10.5281/zenodo.21249296. The code is under the PolyForm Strict License 1.0.0 and the written research under CC BY-NC-ND 4.0.