The work so far.

Not a sequence of finished products. Different attempts at hard questions through engineering and mathematics, each described at the level we can stand behind, with its real status.

Agostino Ramelli's engraving of a book wheel.

Leonardo da Vinci's drawing of a flying machine with articulated wings. — Leonardo da Vinci, design for a flying machine, about 1488.

Earlier research

Chimera

Experimental software intelligence. Could software learn from what happens to it?

Chimera explored software as something you could inspect through alternative runs, experimental changes and accumulated evidence: semantic replay, branching experiments, shadow execution, and recording the conditions under which a result could be trusted.

Some parts became executable experiments and engineering infrastructure. Others stayed architectural proposals, and the wider system never reached the integration its intended claims needed.

A system can hold an enormous amount of information without understanding what it means.

Kepler's model of the planets nested in perfect solids. — Kepler's model of the planets, 1596. Beautiful, precise, and wrong.

Exploring

Baby Laplace

Machine-native scientific reasoning. How would a machine know it understands anything?

Imagine a computer given an unfamiliar world. Its predictions may be excellent, yet several different mechanisms could produce exactly the same observations. Baby Laplace explored executable theories of small worlds: candidate mechanisms that can be run, compared, challenged and told apart by experiment.

The name was deliberately grand for a very early idea. The general scientific intelligence behind it remains an open goal, and no results are reported from it.

A successful simulation is not necessarily an explanation.

Agostino Ramelli's engraving of a book wheel: a reader at a large rotating wheel of open books. — Agostino Ramelli's book wheel, 1588: a machine for reading many books at once.

Built

Genesis-1

Experimental research infrastructure.

A working research harness for structured computational investigations: candidate generation, experiments, records and verification.

Its job is to help us run and check research. Its existence does not mean autonomous discovery has been achieved.

Research that cannot be traced cannot be trusted, least of all by the people doing it.

Dürer's Melencolia I. — Albrecht Dürer, *Melencolia I*, 1514.

Under review

Hidden mechanisms

Mathematics and statistical physics. Sometimes the world tells you less than you think.

Suppose you can observe a system's behaviour but cannot inspect its mechanism. How much can you actually know about what happens inside? We study explicit finite-state constructions in which behaviour that is identical from outside hides different internal physics, including systems that look reversible while dissipating energy within.

Selected constructions have passed internal exact checks. Novelty, interpretation and publication readiness still need careful verification and specialist review.

More computation cannot recover information the observation never supplied.

And then, Worldfold.

The questions kept getting bigger. Worldfold is where they lead: an open search for new sources of capability.

Meet Worldfold