Architectures
Two architectures.
Choose one.
One holds a law that must not drift. The other glues what many sources know, and says where they cannot be glued.
01Physics Encoded Categorical Neural Network
ARCHON
Symmetry, built in.
- Exact, not approximate
- Independent of the data
- Checkable before it runs
02Sheaf Theoretic Categorical Neural Network
SHEAFON
It knows where it fails.
- Consistency is checked, not assumed
- Failure has a location
- It monitors its own monitoring
Both are laboratory validated, TRL 4. How they differ, in the questions