This is not a loose idea, an abstract article, or a vague prototype.
What I’m presenting is a full, principle-first research program aimed at solving a core problem in AI: how to make reasoning explainable, composable, and operational.
I’ve developed a framework that objectifies reasoning by treating it as a manipulable, self-referential computational object.
The result is what I call an Operational Universal Reasoning Substrate (URST): a substrate where reasoning becomes explicit, inspectable, and optimizable.
The only missing piece now is a domain-specific language for expressing and testing these reasoning objects.
10-minute overview video
https://youtu.be/WD-TSqAi9K8
Open-source repository
https://github.com/Eric-Robert-Lawson/OrganismCore/
One additional contribution:
The project includes a unique onboarding methodology, AGENTS.md, which equips LLM-based coding assistants to reason about the entire codebase with file-level grounding. This enables anyone to have a conversation with a non-biased agent that fully understands the project’s structure, design, and assumptions.
The research program provides:
a novel approach to AI explainability,
an operationalization of reasoning as a first-class object,
a description of the RARFL training process.
I welcome critique, contributions, and engagement from the community.
My hope is that this work becomes a foundation for a new way to understand and engineer reasoning.
If you are inspired by this idea, you can reach out to the authors for collaboration or cite it:
@misc{lawson-organismcore-universal-reasoning-2025,
author = {Lawson, Eric},
title = {OrganismCore - Universal Reasoning Substrate Theory},
year = {2025},
url = {https://hypogenic.ai/ideahub/idea/kdFGU6PWwfRWtMlg1u2C}
}Please sign in to comment on this idea.
I have further developed OrganismCore and now have demonstrations showing the project explaining itself. The Automated Onboarding process makes this possible and interactive. This is not post-hoc explainability or interpretability, this is truly explainable AI. It is reproducible, auditable, and fully self-explanatory.
Most importantly, I demonstrate consistency across multiple models including Claude Sonnet 3.5, Grok, GPT-5 mini, and Gemini 2.5. You can explore some demo articles here:
https://doi.org/10.5281/zenodo.17727848
https://doi.org/10.5281/zenodo.17732787
Everyone can try this out on their own, and I highly encourage it.
I am also working to implement this framework in organizations and workflows. The possibilities are vast, and once you experience it, you will understand the frontiers this opens for reproducible, model-agnostic reasoning.
EDIT:
This chat has me reaching a new point, it can only be described through experiential experience. Feel free to engage with this agent... The results are genuinely staggering and need validation.
https://github.com/copilot/share/821d1330-0aa4-8cf4-b111-dc4fe01b61fd