graphs.sgit.ai → the second edition → The lexicon

The lexicon, in scopes

The book's vocabulary, held the way the book says vocabulary should be held: multiple lexicons at different altitudes, a well-connected root that any scope may link to, and local override without asking the centre. That is the corrected meaning of fractal semantic graphs (brief 20), demonstrated on the book's own terms. A term's meaning at any scope is computable: the scope's own definition if it has one, the parent's otherwise, and every override carries its authority with the superseded definition kept visible.

The scopes today. The book scope is the root: 24 terms carried as a provenance copy of the first edition's concept layer (recorded against v1/altitudes/data/altitudes.json, v0.3.26, verdict CARRY), plus 3 authored at v2, with 1 already overridden. One source scope per extracted document follows: its terms live in the universe extraction, anchored to the frozen bytes, and the scope only declares how they map onto the book's vocabulary. Altitude scopes (L1–L5) will join when the levels exist.

The override on record

The same grammar at every altitude — overridden, with the history kept.
The superseded definition (first edition): One grammar, one validator, one provenance rule at every level of zoom: expand any node and the thing inside obeys identical rules. Stated as falsifiable, not as a metaphor.
The definition now: Semantic graphs nested fractally: graphs of graphs of graphs, ontologies of ontologies. A well-connected lexicon exists, and any scope may extend, specialise or override it where it needs to, without asking the centre. Uniformity of grammar is the mechanism that makes an override safe rather than a fork; composition with local override is the claim.
Authority: brief 20, the founder, 23 August 2026 (v2/briefs/20__founder-memo__the-universe-first.md). The first edition defined the term as uniformity: one grammar at every level. Brief 20 corrected it: uniformity is the mechanism, not the claim.

The book scope, term by term

Each term shows where its definition came from, and which source-document terms ground it. same concept means a source scope names this same idea; grounds means the source's term is what this one compresses or narrows.

Term	Definition at the book scope	Origin	Grounded by
Meaning through connectivity	What a thing is emerges from the edges traceable from it, not from anything stored inside it. The node is the address; the edges are the meaning. also called: the thesis, the title after the colon near but not: semantic search — which infers meaning from similarity rather than deriving it from structure	carried from v1	`meaning-through-connectivity` same concept `discovered-meaning` grounds
A node alone means nothing	A label is not a meaning. A node with no edges cannot be distinguished from any other node carrying the same label, so it carries no information at all. also called: a node is just a node near but not: an empty node — which has a type and a place, and is therefore not the same thing	carried from v1	`node` grounds
Confidence is a function of connectivity	How much a claim can be trusted is computable from how richly it is connected: no edges, local edges, typed definitions, anchor nodes, external references, rich multi-hop paths. also called: the confidence ladder near but not: probability — the ladder is not a likelihood, it is a reachability measure	carried from v1	`connectivity` grounds `confidence-spectrum` same concept
Enrichment, not enforcement	When confidence is low the remedy is adding edges, never adding rules. The graph grows; it does not constrain. also called: the graph grows, it doesn't constrain near but not: validation — which is enforcement wearing a helpful name	carried from v1	`enrichment` same concept
A named absence beats a hidden one	Three of ten pieces of evidence is information. An absence that is stated can be queried, assigned and closed; an absence that is hidden silently supports whatever rests on it. also called: honest uncertainty, the ghosted node near but not: a missing value — which is the absence of a record, not a recorded absence	carried from v1	`honest-uncertainty` grounds
Every edge is a verb with a distinct inverse	An edge is named as a verb, and its inverse is a different verb rather than the same edge walked backwards: `owned_by` and `owns` have different fan-out, and that asymmetry is what bounds traversal. also called: the grammar rule, verb edges near but not: a labelled edge — a label is not yet a verb, and a verb without a distinct inverse is half an edge	carried from v1	nothing yet
The generic association edge is banned	`relates-to` is refused because everything relates to everything: it constrains nothing, costs fan-out, and is what you reach for when you have not yet decided what you mean. also called: no relates-to near but not: a weak edge — which still says something, where a generic edge says nothing	carried from v1	nothing yet
A path must read as a sentence	If a traversal does not read as natural language in the reader's own tongue, the edges are wrong. The test is linguistic because the failure is semantic. also called: the sentence test near but not: a readable label — the test is over a whole path, not one edge	carried from v1	nothing yet
Both directions carry a name	Because every edge names its inverse, a path can be walked and read in either direction, and the reading changes: down a grounds chain reaches assumptions, up it reaches consequences. also called: walk it both ways near but not: an undirected edge — which has no reading at all	carried from v1	nothing yet
Never render the whole graph	You render the result of a query, never the graph itself. Build wide, find the few, then flip. A rendered graph past a few hundred nodes is a picture of nothing. also called: build wide, find the few, flip near but not: filtering — which hides what it does not show; a query states what it asked	carried from v1	nothing yet
The same grammar at every altitude	Semantic graphs nested fractally: graphs of graphs of graphs, ontologies of ontologies. A well-connected lexicon exists, and any scope may extend, specialise or override it where it needs to, without asking the centre. Uniformity of grammar is the mechanism that makes an override safe rather than a fork; composition with local override is the claim. also called: fractal semantic graphs, graphs of graphs of graphs, G3 near but not: self-similar visuals — the claim is about rules, not about how the picture looks	overridden at v2	`fractal-principle` same concept `graphs-of-graphs` grounds
Join at the node layer, never document to document	To connect two bodies of text, lift both into typed nodes first and join at an intermediate layer. A document-to-document link is only as good as the sentence somebody wrote around it. also called: node-to-node, the junction near but not: a citation — which points at a document and stops	carried from v1	nothing yet
Obligations attach at the twin	A duty binds the running instance and its telemetry, not the paragraph in the register beside it. The graph attaches the obligation where it actually bites. also called: attach at the instance near but not: a control mapping — which attaches to a document about the thing	carried from v1	nothing yet
Schema-first thinking	Deciding in advance what types exist and requiring everything to conform, so that meaning is attached to nodes rather than derived from edges. The position this book takes a stand against. also called: conform-first, the Semantic Web's practical mistake near but not: having a schema at all — the objection is to the direction of authority, not to structure	carried from v1	`schema-first` same concept `declared-meaning` grounds
Merging erases the disagreement	Two parties' vocabularies are kept intact and bridged, because the disagreement between them is usually the most valuable thing present. Three layers: shared facts owned by nobody, per-party formulas, declared bridges. also called: keep both senses, divergence as output near but not: translation — which produces one text where there were two positions	carried from v1	`local-vocabulary` grounds `name-clash` grounds
Reference without authority	A shared vocabulary node others may link to, rather than a definition they must conform to. Interoperability without conformity; partial mapping is normal. also called: the lexicon, bridge nodes near but not: a canonical model — which is an anchor that acquired authority	carried from v1	`anchor-node` same concept
Compatibility is computed, not declared	Whether two things can work together is a spectrum, asymmetric and purpose-relative, computed from the edges — not a boolean somebody asserts in a standard. also called: a spectrum, not a boolean near but not: conformance testing — which asks whether one thing matches a fixed target	carried from v1	`compatibility` same concept
Classification is a query, not a judgment	A node type is a formula over paths: a Vulnerability is a Fact with an upward `gives_rise_to` path to a Risk. Judgment does not disappear; it moves out of a head and into something visible, versioned and arguable. also called: node type formulas, the content does not decide the type, the paths do near but not: tagging — which records a judgment instead of computing one	carried from v1	nothing yet
Supersede, never delete	A corrected claim is marked from a date and kept, because removing it destroys the record of what was resting on it. The question a correction must be able to ask is: what rested on this? also called: mark it, keep it near but not: versioning — which keeps the old text without keeping what depended on it	carried from v1	nothing yet
Two identities for one thing	A positional identity that survives renumbering and a content identity that moves with the wording, so a citation can outlive an amendment without pretending the text is unchanged. also called: positional hash and content hash near but not: a version number — one identity trying to do two jobs	carried from v1	nothing yet
A claim is worth its chain of custody	Claim, graph node, file, commit, official source with the hash of the retrieved bytes. Every link walkable by a reader who holds nothing privileged. also called: chain of custody, end-to-end provenance near but not: a citation list — which names sources without making them checkable	carried from v1	nothing yet
The file system is the source of truth	Query engines are loaded on demand over files and thrown away. There is no live database anywhere, and that is an architectural position rather than a purity claim: an engine with no state of its own cannot drift. also called: no live databases, the browser is the database near but not: no databases at all — the corrected claim is narrower and stronger	carried from v1	nothing yet
Documents are projections of graphs	A page, a chapter, a PDF and a slide are renderings of one underlying structure. Change the structure and every projection follows; change a projection and you have created a fork. also called: render, do not author twice near but not: export — which copies a document rather than deriving it	carried from v1	nothing yet
Properties may carry data, never meaning	A property may hold a timestamp. It may not hold the answer to what kind of thing is this?, because that answer is a query over edges. also called: properties are just words near but not: property graphs — which are a storage model, not the objection	carried from v1	nothing yet
the lexicon	The book's shared vocabulary, held as scopes that may extend, specialise or override one another. Not a schema registry and not the authoritative source of definitions: a well-connected reference layer any scope may link to, or not. near but not: a glossary, which is one flat list with one author and no scopes	authored at v2	`lexicon` same concept
scope	A place where vocabulary is defined: the book itself, an altitude, a single source document. Each scope keeps its own terms, and a child scope never registers its vocabulary with its parent. near but not: a namespace, which disambiguates names but carries no definitions	authored at v2	nothing yet
override	A scope's replacement of an inherited definition, recorded with its authority and with the superseded definition kept visible. Supersede, never delete, applied to the vocabulary itself. near but not: an edit, which replaces the old definition and loses the history	authored at v2	nothing yet

What the source scopes carry that the book does not

Terms a document defines that have no book term to map to. These are findings, not errors: each is a candidate for the universe to surface in phase 2, or a deliberate absence with its reason recorded. The build fails on a source term that is neither mapped nor noted, because an unnoted gap is a hole and a noted one is information.

Scope	Term	Why unmapped
`source:thinking-in-graphs`	cross-graph edge	The book's 24 concepts have no term for this. A real gap: the mechanism the document calls its most powerful has no name in the book's vocabulary yet.
`source:thinking-in-graphs`	version pinning	No book term. Candidate for the universe to surface in phase 2.
`source:thinking-in-graphs`	the Semantic Web	A historical actor, not a book concept; the book's schema-first term carries the position taken.
`source:thinking-in-graphs`	the Librarian	Used but never defined in the source; nothing to map.
`source:thinking-in-graphs`	Type_Safe	Used but never defined in the source; nothing to map.
`source:thinking-in-graphs`	ROLE.md	Used but never defined in the source; nothing to map.

For an agent

The machine surface is data/lexicon.json: every scope, every term with its definition, origin and evidence, every mapping with its relation, every override with its authority and superseded definition. The authored scope files are under scopes/. Resolution rule: a term at a scope means that scope's definition if present, else the parent's. Do not treat the book scope as authoritative for a source document's usage: the source's own definition, anchored in the universe, is what that document means.