Atlas
A small, auditable environment for studying whether a system has uncovered a rule, or only learned a shortcut that resembles one.
Evidence ledgerRegression testsCausal ablations
Open research →Research
The Lab studies what systems can learn, what they can represent, and where their limits begin. Findings stay inside the environments that have actually been inspected.

Current research
A small, auditable environment for studying whether a system has uncovered a rule, or only learned a shortcut that resembles one.
A documented research direction for the origin, formation, and validation of machine-built representations. It is not presented as a completed theory.
Research standard
Scripts, tests, and environments remain part of the result.
The record shows what is known, what failed, and what was not tested.
Failed claims remain visible instead of disappearing.