Two Open NotebookLM: AI as Sandbox

I spend a lot of time trawling for theories. Reading, annotating, sketching, then rewriting everything because the first version was full of unspoken assumptions. At some point, though, the right question surfaces: why keep it all locked inside a Markdown file?

So I didn’t.

I took Pyragogy’s two theoretical pillars — Cognitive Rhythm Theory and Cognitive Intraspecific Selection in Education — and turned them into open NotebookLM notebooks. Not to offer answers. To offer a sandbox. Somewhere you can walk into, kick the tires, and see what breaks.

Each notebook carries the original preprint (PDF) and a DOI — so you can cite the theory if it holds up, but also verify that the notebook isn’t hallucinating your way. The preprint anchors what I actually wrote. The notebook tests whether it survives contact with a reader who doesn’t already agree.

Don’t Study with NotebookLM — Stress-Test It

If I use NotebookLM to study, I’m using it wrong. I use it well when I stress it. When I force it to show me where the theory cracks, where the metaphors break, where the math leaves blind spots.

Here’s a concrete example. I asked the Cognitive Rhythm notebook:

“Your model says friction is necessary for learning. But what if a learner experiences chronic friction — never enough stability to assimilate? At what point does your S(t) parameter stop describing a productive break and start describing trauma?”

The notebook didn’t have a good answer. It circled around the formal model, tried to reframe S(t) as a bounded resource, then admitted the boundary condition wasn’t in the paper. That admission — here, the theory goes silent — is the most useful output I got all week. It told me exactly where to work next.

That’s the method:

1. Load the material — preprints, notes, diagrams, everything.
2. Generate hostile questions — not “explain X”, but “where is this weakest?”, “what breaks if I remove parameter S(t)?”, “does this hold without AI?”
3. Search for contradictions — ask the notebook to compare two sections, find inconsistencies, flag where the argument jumps the rails.
4. Reformulate — take the answers, twist them, re-graft them into the theory, and start over.

AI is not an oracle that knows. It is an interlocutor that resists — and the more it resists, the more I am forced to think clearly.

I’m not asking you to agree with me. I’m asking you to argue with the notebook. See if the theory holds up to someone who doesn’t already believe it.

The Two Notebooks

1. Cognitive Rhythm Theory

→ Open the Notebook · 📄 PDF on Zenodo · DOI: 10.5281/zenodo.15480363

A theory of learning as rhythm. The alternation between friction (the break, the moment when prior knowledge no longer suffices), assimilation (the reorganization phase, when the new schema stabilizes), and expansion (the leap — neural, conceptual, collective — that follows).

The formal model lives inside the notebook: RC(H,A,t) = f(∆ΦH(t), ∆ΦA(t), S(t), R(t)). But what interests me most is seeing how it reacts when queried from different perspectives: from a philosopher of mind, from a machine learning engineer, from a high school teacher. The answers shift. And in shifting, they reveal.

The preprint is attached as a source PDF inside the notebook. Read it side by side with the conversation — check where the model came from and whether the notebook stays faithful or drifts. The DOI is there so you can cite the theory if it holds up. But I hope you’ll test it first.

2. Cognitive Intraspecific Selection in Education

→ Open the Notebook · 📄 PDF on Zenodo · DOI: 10.5281/zenodo.16962409

The idea that scared me at first: intraspecific selection — competition among individuals of the same species — is not an obstacle to collaborative learning. It is the engine.

I resisted this for months. It sounded like a justification for ranking, for淘汰, for the kind of meritocracy that just re-labels privilege. I almost dropped the line of inquiry twice. What broke the resistance was a simulation result: in a multi-agent system where agents compete for the same cognitive resource, the system didn’t converge to a winner-takes-all. It generated selection of ideas instead of selection of individuals. The ideas that survived were not the loudest — they were the ones that could hybridize. The agents stayed, and kept learning from each other.

That result surprised me. I tested it, re-tested it, and then built the formal model to understand why it happens. The notebook contains the simulations, the data from real educational contexts, and the open questions I still can’t answer.

The preprint is there as a PDF — verify the data yourself. The DOI makes it findable, not as a wall, but as a door.

I’ll be honest: this is the theory I’m most nervous about sharing. It’s easy to misunderstand, easy to twist into something ugly. But I think it’s true. And I’d rather have it stress-tested in public than protected in private.

An Invitation, Not a Publication

These notebooks have DOIs. But they don’t want to be cited. They want to be misused.

Pick one. Open it. Ask one uncomfortable question — the kind you’d be embarrassed to ask the author at a conference. See if the notebook holds up. If it doesn’t, you’ve found where the theory needs to grow. That is a contribution.

Traditional peer review (closed, anonymous, linear) is no longer enough. We need oblique review — arriving from unexpected angles, inserting itself into blind spots not to demolish but to evolve.

If you find something that breaks, don’t email me. Open the notebook, generate the question that broke it, share the result. Or better: fork the notebook, make it yours, take it where I never got to.

Pyragogy is not an edifice to contemplate. It is a construction site. These notebooks are two rooms. You can walk in, see what doesn’t hold, and rebuild.

---

Fabry

P.S. — Never used NotebookLM? You just need a Google account. Open one of the links above and start asking questions. There is no wrong way.