← Research

The Perpetual Plant

A world sample. From a single seed sentence, the machine grew a self-maintaining power plant — its engineering, its failure modes, its repair doctrine — and kept it coherent: every figure below is annotated by the same machinery that imagined it, and the annotations agree with each other.

world sample · generated one seed sentence

Nothing on this page is a real facility, and that is the point. A specialist cannot learn deep experience from records the world never kept — so the machine constructs worlds in which that experience can be had: situations with physics, constraints, consequences, and enough internal consistency that a decision made inside them has something real to collide with.

Annotated cutaway of the generated plant: containment dome, dry cooling towers, switchyard, perimeter
the plant — containment, dry cooling, switchyard, perimeter; one coherent site grown from the seed
Robotic service corridors with gantry rail, charging dock and tooling bay
the service corridors its machines move through
Control stack: cabinets, cable trays, telemetry wall
the control stack that watches everything
A service arm repairing a humanoid maintenance unit on a bench cradle
the workforce — maintained by the same doctrine it executes

One failure, followed through

Depth means consequences. The world doesn't just contain equipment — it contains what happens when equipment fails, and what a competent response looks like, step by step, to the restored state:

A flagged bearing on a salt pump skid, vibration sensor annotated
1 · a bearing is flagged by its vibration sensor
The failed cartridge extracted by a heavy-lift arm
2 · the failed cartridge comes out on the heavy-lift arm
The restored skid back online, nominal indicator lit
3 · the skid returns to nominal; the unit departs

Why a specialist needs this

Decision experience is situations with trade-offs and consequences attached. Worlds like this one are where those situations come from: a scenario is a region of the decision space to master, not a fact to memorize. In our vision work, the same principle — training on focused generated worlds — moved small models drastically in the target direction without losing their general skill; the studies on the research page measure how far that carries into business decisions.

This shows

  • Depth and internal consistency: one seed, one coherent world — sites, systems, failure modes and repair cycles that reference each other correctly.
  • What the training corpus is made of — followed-through experience, not isolated text.

the full world sample — chapters, spec sheets, repair doctrine — available on request · hello@truna.ai