Studies — when the field theory became a module that runs
The question that came after the post
When I wrote the post about wave field theory and Mamba, I had already spent months studying the idea of treating inference as wave propagation in a phase field. Experiments, vault notes, and an understanding that grew with every read. But something was nagging in the subtext: the theory lived in my head and in a README — it had no place where it actually ran.
A study that doesn’t become an artifact disappears. I had watched it happen dozens of times: a paper read, an idea jotted down, a proof of concept forgotten in the footer of a notebook. Two months later, the “amazing idea” becomes “wait, how did that work again?”.
So one late afternoon, I decided to test a hypothesis about my own study method: if I force every theory into a small, portable, testable module, it survives better.
The context — theory without a home
Wave field theory, at the conceptual level, is simple to summarize: information travels as waves in a field, and the interference between waves is the computation. Each “neuron” isn’t a static number — it’s a wave source with phase and amplitude. The connection between two points isn’t a scalar weight — it’s a phase coupling.
It’s beautiful on paper. The problem is that ideas like this die in two ways: either they stay locked in a notebook nobody opens, or they get so coupled to heavy infrastructure (GPU, kernels, tooling) that nobody can play with them.
What I wanted was the middle path: a didactic clone — small, pure, and portable — that keeps the core mathematical idea without carrying the weight of the production implementation.
The struggle — fitting an idea into a couple hundred lines
The challenge became a discipline: express the entire theory in a small module, no GPU, no kernels, just the math of the idea. Cut it down to the bone, leaving only what is essential to understand how the thing thinks.
What survived the cut:
- Ternary lens: each element carries a discrete weight of three values (the “ternary” part of the idea) plus a continuous adjustment. It’s the core of the concept: representing a connection with minimal granularity but with fine-tuning capacity.
- Phasor interference: the system’s resonance is the amplitude of the sum of waves. High constructive interference = the system is “satisfied”. That becomes an objective measure of how well the field is configured.
- Adjacent method: instead of a gradient over continuous space, the adjustment tests the possible discrete values of each element and picks the one that minimizes the error. Cheap, simple, stable.
- Hysteresis: only change an element if the change genuinely improves things. That avoids the infinite oscillation of naive methods — a deliberate pause against the “flip-flop” behavior every unstable iteration has.
- Autopoiesis: the field grows by itself. Elements that stack up too much adjustment split, ones that no longer contribute get pruned, and occasionally a random new element appears (the “novelty” that keeps the system from getting stuck in a local minimum).
The hardest part wasn’t the math — it was restraint. Every time I wanted to add another feature, I remembered the goal: to be an artifact that anyone (or the me of three months from now) can read in an afternoon and understand the idea. Every line had to justify its existence.
The resolution — an artifact that runs
The result was a module a few hundred lines long, pure numpy, with no dependency heavier than the math. It runs anywhere. And most importantly: it gives feedback. You can assemble a field, propagate the waves, measure resonance, and watch the system evolve (or not) — which turns passive reading into an experiment.
What this small piece taught me about studying changed more than the theory itself did:
- A theory is only understood when it’s rewritten. Reading and taking notes is reception. Rewriting the idea into a running module is reconstruction — and it’s in reconstruction that the gaps in understanding show up.
- Restraint is part of the method. The size constraint wasn’t a limitation, it was a tool. It forced me to separate the essential from the accidental in the theory.
- Portable is didactic. If an idea only runs on heavy infrastructure, nobody plays with it — not even me. Portability is what makes a study re-examinable.
- Feedback closes the loop. A study without feedback is a noted fact. A module that runs returns a signal: the system converges, oscillates, or stays still — and each answer is a new question.
Lessons
This week’s study wasn’t about a formula. It was about the method for making study last: turning an idea into a small, portable, testable artifact. The field theory was the vehicle; the lesson applies to any topic.
Since then I’ve started asking every new study the same question: “what’s the smallest running artifact that proves I understood this?” If I can’t answer, I haven’t understood yet — I’ve only read.
What’s next
The next step is to turn this habit into a documented rule in the ecosystem, the same way the skills inventory (308 skills, 39 categories) became a reuse hub. If the lesson survives, it becomes a skill too: “a finished study leaves an artifact that runs.” It’s the natural continuation of an approach that’s already working — documenting while you build.