Orion

Exploring new foundations for computation.

Orion explores new foundations for computation, with an interest in what a model can retain, recover and use as its work continues.

Kind
Research programme

The idea behind Orion

We built Orion to investigate further possibilities for how a model can compute and retain useful information.

The research asks how earlier encounters can remain useful and what can be preserved as computation proceeds.

Orion investigates faithful retention and useful computation in a developing model.

Memory and computation contribute different parts of that question: what can be retained, what can be recovered, and what can be used.

Preserving what the model holds

A central part of the research is preserving the information a model holds.

Recorded experiments demonstrated exact recovery of that information.

This established a foundation for further investigation, with faithful recovery demonstrated.

The information could be recovered intact.

Bringing state into computation

Orion also demonstrated repeatable responses.

Its recorded behaviour remained consistent when the same conditions were repeated.

This provided a stable foundation for examining changes in behaviour as the research developed.

The result matters because dependable computation makes further investigation more meaningful: a change can be studied against an established result.

Making earlier encounters useful

Memory introduced another dimension to this work.

In recorded experiments, earlier encounters influenced subsequent output when material was encountered again.

This provided an early example of retained information contributing to ongoing computation.

The wider question is how that relationship can develop: how a model can preserve useful distinctions, recognise when they matter again, and carry them forward into later activity.

The research ahead

Orion's research direction brings memory, state and computation together.

The recorded results provide foundations for studying richer forms of memory, more varied inputs and further computational abilities.

Orion now studies how these properties work together as the basis of a developing model, building on the preservation, repeatability and influence of earlier encounters demonstrated in the experiments.

Preserving information. Carrying it forward. Exploring what computation can become.

Cite this record

Forlais Research (2026). Orion. Forlais Research Library, record FR-P-01.

@misc{fr-p-01,
  author = {{Forlais Research}},
  title = {{Orion}},
  howpublished = {Forlais Research Library, record FR-P-01},
  year = {2026}
}