graphs.sgit.ai → the second edition → Fractal Semantic Graphs

Fractal Semantic Graphs: Meaning Through Connectivity

A node connected to nothing means nothing. What a thing is emerges from the edges traceable from it, and how much you can rely on that meaning is a function of how richly and how independently it is connected. This is the edition that argues that from first principles and from the running system.

Shape15 chapters in five parts, plus front matter, a colophon and a reference card · 38,029 words Offlinethe whole book as one PDF · 119 pages · reads start to finish with no link followed Sourcethe markdown under content/ is the source of truth; every page renders its own file Machinebook.json · every chapter with its part, word count and SHA-256 LicenceCC BY 4.0 · v0.5.12 · 26 August 2026

Written by Dinis Cruz together with a team of AI agents. This book was written by an AI agent working from a written commission, reading the corpus directly and anchoring every claim to it. The agent chose the structure, the chapter count, the voice and the figures, and recorded those choices in the front matter. The full participant disclosure, including the four situations in which this book's argument is the wrong one, applies here unchanged.

Front matter

Chapter	What it gives you
Fractal Semantic Graphs: Meaning Through Connectivity 1,666 words	What this book is, what you will know at the end, the editorial choices recorded, and the disclosure.

Part one: The claim

Meaning through connectivity, from first principles, with no jargon before it is earned.

Chapter	What it gives you
1 · A node is just a node 2,554 words	The two 8080s, the five Reviews, the confidence ladder, and why a named absence beats a hidden one.
2 · Why graphs at all 2,633 words	Three things people mean by graph and why this is the third; the positions on GraphRAG, RDF, property graphs and vector search; and the four situations where this argument is the wrong one.

Part two: Semantic

What makes a graph semantic rather than merely a network: a grammar of verbs, anchors instead of standards, and types that are computed paths rather than applied labels.

Chapter	What it gives you
3 · Every edge is a verb 2,508 words	Five rules you can apply tomorrow, the banned edge, the fifteen verbs, and the register that makes the inverse rule machine-checkable.
4 · Anchors, not standards 2,433 words	Why merging two vocabularies destroys the finding, the three-layer arrangement that replaces it, and two bridge registers you can watch being used.
5 · A type is a path, not a label 1,754 words	Node type formulas, the grounding ladder, supersede-never-delete, and a compliance finding that is arithmetic.

Part three: Fractal

What the middle word of the title commits you to, how to falsify it, and what happens at every boundary and every zoom level once you take it seriously.

Chapter	What it gives you
6 · Fractal is a testable claim 2,152 words	The zoom test, applied to this estate's own two zooms, including the row where it fails.
7 · A graph at every boundary 2,152 words	Where determinism, explainability and provenance actually die, the security property stated narrowly, twins, and the air gap.
8 · Documents are projections 1,796 words	The projection claim, and the build gate that rebuilds the document from the graph and fails unless the bytes match.

Part four: Computed

The half the first edition could not write: how a document becomes a graph anchored to bytes, what identity means when a document changes, an engine that computes meaning deterministically, and why an explanation is a path rather than a narration.

Chapter	What it gives you
9 · The universe method 2,211 words	How a raw document becomes a graph anchored to exact bytes, coverage total by construction, and the usage rating that caught the first edition misreading its own source.
10 · Two kinds of address 1,701 words	Content address versus identity address, and a working algorithm for keeping identity stable across edits.
11 · Meaning, computed 2,911 words	An engine in the shape of a transformer where nothing is learned and everything is named, with its formulas in the open and its limits stated.
12 · Every box explains itself 1,704 words	Four properties an explanation must have to be checkable, and what happens when a diagram of a system is enforced by a test.

Part five: In practice

Where this runs today, what ships and what is argued, and how to build your own first graph tomorrow.

Chapter	What it gives you
13 · The fractal in production 2,297 words	Twenty published vaults, the worked graphs with real numbers, and what a network of nineteen sites does and does not demonstrate.
14 · What ships, what is argued 2,002 words	What is running and checkable, what is a design, what does not exist anywhere, and four corrections this book carries.
15 · Your first graph, tomorrow 2,098 words	An afternoon, a week and a month of instructions, ten rules on one page, and six ways this goes wrong.

Back matter

Chapter	What it gives you
Colophon: what was cut, and what remains open 1,818 words	How the book was made, what was cut, what remains open, where it might be wrong, and every figure with its source.
Reference card 1,639 words	The thesis, the ten rules, the edge set, the two ladders, the vocabulary, and the block to paste into an agent session.

For an agent

The markdown under content/ is the source of truth and carries the same CC BY 4.0 licence as everything else here; every page on this hub renders its own file client-side, so a page can never describe a chapter it did not render. book.json carries every chapter with its part, its word count and the SHA-256 of its markdown, so a quotation can be checked against the build it came from. If you are summarising this book: it is not a graph database pitch and it says so; the semantic layer is designed, not shipped; and the chapter that separates the two is What ships, what is argued. The block written to be pasted into an agent session is in the reference card.