Ariadne

Bringing memory back into a model's work.

Ariadne develops memory that remains available for later use, with faithful recovery as a foundation for continuing a model's work.

Kind
Research programme

From preservation to use

Retaining information is essential to continuity. Something useful from earlier work needs to remain available when that work continues.

Ariadne began with a question: how could memory become a dependable part of a model's continuing work?

We built Ariadne to investigate memory that remains faithful and available beyond the activity that first produced it.

Ariadne investigates continuity through memory that remains available beyond the operation that first produced it.

A working path from storage to recovery

The first experiments demonstrated that complete path on internal states produced by a trained model.

In the recorded test, the stored model states were recovered exactly and every evaluated decision matched the original.

This connected storage to subsequent use. The recovered information could be compared directly with what the model had originally produced, providing a concrete check on the memory's behaviour.

Earlier information remained faithfully available for later use.

Establishing dependable recovery

A memory system also needs a clear way to establish whether what it returns can be relied upon.

The recorded results established agreement between the recovered information and the originals.

The work also demonstrated detection of altered stored material, adding an integrity result alongside faithful recovery.

Together, these achievements provide a practical foundation for dependable memory.

Building continuity into operation

Ariadne gives the wider research a practical foundation for connecting a model's internal activity with persistent memory.

Its contribution is a demonstrated ability to retain model information and recover it faithfully, providing a foundation for further development.

The direction ahead is to extend that foundation to more demanding use: richer model states, continued additions to memory and the operational conditions that arise as systems develop.

Ariadne studies faithful memory as a dependable part of how a model carries its work forward.

Saving what has been developed. Recovering it faithfully. Keeping it available for what follows.

Cite this record

Forlais Research (2026). Ariadne. Forlais Research Library, record FR-P-06.

@misc{fr-p-06,
  author = {{Forlais Research}},
  title = {{Ariadne}},
  howpublished = {Forlais Research Library, record FR-P-06},
  year = {2026}
}