Every glowing point is a real node the system has grown. Every filament is a real connection it made. The travelling sparks are signals spreading — the way activation moves through a network of neurons.
This is not a simulation and not decoration: it's drawn straight from the graph's own data, and it re-reads itself every ~45 seconds — so anything you watch appear is real growth.
In the interest of precision: what you see is a live sample of the strongest connections — the full graph already holds well over a million edges, far more than any screen can render. The exact live total and the sample size are both shown in the top bar (and it changes as the graph grows).
You're seeing the whole system together: its code sense (query words & the code symbols they touch) and its meaning sense (entities, words, Hebrew roots, and the semantic-frequency bands that organise meaning) as one web — two hemispheres of the same graph. The legend, bottom-left, is data-driven — it always names the kinds actually on screen right now.
Real: it genuinely grows, prunes and self-grades from real data, and is fully inspectable — rare among AI systems; its predictions beat naive baselines by measured margins.
Not: not a frontier chatbot, not magic, not new mathematics — the building blocks (association graphs, weighting, retrieval) are classic.
Genuinely unusual is the combination: a readable, self-growing graph, no black box, no external model at answer-time, addressed by a self-dual geometry. The transparency is the edge — not a claim of breakthrough.
It's built to keep growing — millions of connections and beyond as more sources feed it. Two honest notes. First, on the word parameter: each weighted edge here is a learned parameter — and arguably a richer one. A bare neural-net weight is a blind scalar moved only by gradient descent; an edge here is named, typed, and derived from legible evidence (co-occurrence, self-grading, corroboration across many paths), so it self-dials as new cross-referenced facts arrive instead of being nudged in the dark. The only caveat is arithmetic: don't equate this count with an LLM's — a sparse evidence-weight and a dense transformer weight do different work per unit, so equal numbers are not equal capability.
Second, on quality — and this is the interesting part. Early on, dumping in raw low-quality data measurably lowered accuracy. But that was a snapshot, not a ceiling: the very machinery that lets it grow is machinery for discernment — it prunes weak edges, grades its own predictions and penalises associations that don't pay off, and corroborates a claim across many independent paths (one seen once is weak; one reached a hundred ways is trusted). So the design premise is that it should get better at filtering junk and verifying meaning as it matures — not worse. Whether that discernment truly scales with size is the open, testable question this view is built to let you watch.
Meaning is read at three scales, counted in the bar up top: nHz (letter / root — the deepest, triggered on high semantic density), mHz (words), Hz (context). Every thread lives at one of them.
A plain network just connects. This one re-shapes itself under flow — it grows threads where signal flows and proves out, and prunes where it doesn't, the way a vascular system thickens vessels under flow or mycelium routes to where it's needed. The sparks are that flow; the threads that glow brightest are the ones flow has strengthened.
Each edge carries where meaning is (positive evidence) and where it is not (measured absence), together. The measurable seed is on screen now: edges tested both ways are drawn “materialised” — brighter, gaining weight. The founder's conjecture beyond that — that a fully-measured dual becomes a small 3-D shape (“qHz”), trillions of them fractal-tiny — is a direction and a metaphor, not a measured result. Stated plainly, so you can weigh it as exactly that.
You're about to fly around the real network the system has grown — its nodes, its connections, and the signals sparking between them in depth. It isn't a picture of the graph; it is the graph, reading itself in real time.
No hype: a transparent, self-growing knowledge graph — impressive for being fully inspectable and sovereign, not for being magic. Open “what am I looking at?” for the honest version.