DUHBLE
Stateful execution layer
DUHBLE is being developed around explicit model moments, persistent residual state, structural inhibition, provenance, conduction, and model-owned change.
What we're building
A substrate for stateful software, not another wrapper around a database.
The central project is DUHBLE: an experimental runtime and model format intended to keep meaningful state inside a mounted model rather than scattering it through caches, prompts, shadow graphs, and application logic.
.duh
Portable model representation
The goal is a model format whose important behavior is encoded in the model itself, with the runtime acting as executor instead of a second source of knowledge.
Testing
Failure first
The project is being attacked with adversarial tests for simultaneity, provenance, persistence, isolation, decay, and other places where a convenient implementation could cheat.
Scope
Early and deliberately narrow
This is not a claim of general intelligence. It is research into memory, state, adaptation, and lawful change that may become useful underneath larger systems.
What we are not claiming
NeuroSyn is not presenting a finished commercial platform or a magical shortcut to intelligence. We are documenting what works, what fails, and what survives increasingly hostile tests.