# What the atlas found After this chapter you will know which of this estate's ideas are load-bearing and which are decoration, where it disagrees with itself, and what it does not say. None of it is asserted. The rankings are computed from the graph in this book, and the disagreements and silences are quoted from both sides. This is the chapter that justifies building the atlas at all. A compression that only reproduces the agreements has thrown away the interesting half. ## The peaks, computed The estate's own rule is that peaks are computed from edges rather than declared, so this table is arithmetic: how many connections reach or leave each concept in this atlas. The important caveat is on the arithmetic, not on the reading. Degree here measures **this atlas's** graph, which is one session's reading of the corpus, and most of its edges are authored rather than anchored. A concept ranks high because this book connected it a lot. That is a statement about the compression as much as about the corpus, and the honest way to use the table is as a question: does this ordering match what you would have expected? | Concept | Region | Degree | Family | |---|---|---:|---| | [meaning through connectivity](#meaning-through-connectivity) | The meaning core | 9 | concept | | [a named absence beats a hidden one](#named-absence) | Confidence, evidence and reality | 8 | position | | [classification is a query, not a judgment](#node-type-formula) | The grammar of edges | 7 | concept | | [never render the whole graph](#render-the-query) | The grammar of edges | 7 | method | | [the chain of custody](#provenance-chain) | Provenance, correction and time | 7 | concept | | [declared meaning](#declared-meaning) | The meaning core | 6 | concept | | [documents are projections of graphs](#projection) | The fractal principle | 6 | concept | | [every edge is a verb](#edge-is-a-verb) | The grammar of edges | 6 | method | | [fractal is a precise claim](#fractal-principle) | The fractal principle | 6 | concept | | [quote-anchored extraction](#quote-anchored-extraction) | The machinery | 6 | method | | [the anchor node](#anchor-node) | Anchors, scopes and bridges | 6 | concept | | [the confidence ladder](#confidence-ladder) | Confidence, evidence and reality | 6 | concept | | [what ships, what is argued](#what-ships-what-is-argued) | The estate, the vaults and the network | 6 | position | | [compatibility is computed, not declared](#compatibility-computed) | The meaning core | 5 | concept | | [direction bounds the result, not the graph](#direction-bounds-fan-out) | The grammar of edges | 5 | claim | The founder's own memo warns against exactly the misreading this table invites: a concept can be load-bearing without being visible at the top, so importance and altitude are not the same thing. Read the next table with that in mind. ## The thin ones Concepts with two connections or fewer. Being thin is not a criticism of the idea. It is a statement about how much of the corpus reaches it, and the estate's own confidence ladder says exactly what that means: an assertion with few edges is one that will bear less weight than a well-connected one, and the honest response is to say so rather than to level the picture. [agenda is context, not a verdict](#agenda-is-context) (2) · [composition with local override](#composition-with-local-override) (2) · [downward grounds, upward implies](#two-directions) (2) · [everything modelled must be real](#reality-discipline) (2) · [pin or float, but say which](#version-pinning) (2) · [propagation is what a graph can do](#propagation-of-corrections) (2) · [the Semantic Web's mistake](#semantic-web-mistake) (2) · [the cross-graph edge](#cross-graph-edge) (2) · [the floor](#the-floor) (2) · [the same value, differently connected](#the-difference-is-connectivity) (2) Three of these are worth a sentence, because thinness is telling you something different in each case. **The same value, differently connected** is the port example, which is the single best on-ramp in the whole corpus and is barely connected here. It is thin because an example does one job and then gets out of the way. That is a fact about examples, not a reason to demote it. **Composition with local override** is the corrected definition of the title's own first two words, and it has two connections. It is thin because it is new: it arrived on 23 August 2026 and the corpus has not yet rebuilt itself around it. That is a finding about the corpus, and it is the one worth watching, because a corrected definition that nothing leans on has not really been adopted. **The floor** is thin and probably should not be. It is the stopping rule that keeps the grounding ladder from receding forever, it appears in exactly one brief, and nothing else in the estate builds on it. An idea that load-bearing with that few edges is either under-developed or under-cited, and the atlas cannot tell you which. ## Where the corpus contradicts itself Six places. In each one, the estate says two things that cannot both stand as written, and in five of the six it says so itself, in public, rather than being caught at it. That habit is the most persuasive thing in the material. ### the fifteen established edges · against · the generic association edge is banned One row of the published edge set is in tension with the estate's own first rule. > Treat it as a last resort: wherever you can name what kind of connection it is, name it, and > the query gets better. > > — *The edge set*, § One row is in tension with this book's own rule ### the generic association edge is banned · against · what ships, what is argued The rule is broken by the project's own shipped configuration, and named rather than quietly fixed. > Its own `link-types.json` ships one as a self-inverse pair, and a single edge instance uses > it. > > — *What ships, what is argued*, § Two corrections we inherit rather than repeat ### composition with local override · against · fractal is a precise claim The first edition defined fractal as uniformity. Brief 20 says that is the mechanism, not the claim. Both are kept, one marked superseded from a date. > Uniformity has not disappeared. It is a consequence: what makes the override safe is that > the grammar for expressing an override is the same at every level. But it is the mechanism, > not the claim, and the pack had them the wrong way round. > > — *Brief 20: build the universe first, then find the plot*, § The correction to fractal semantic graphs ### properties carry data, never meaning · against · what ships, what is argued The rule says properties carry no meaning; the one graph actually running in the project is a typed property graph. > this book's core rule is that properties do not carry meaning, and the one graph in this > project that is *actually running* is a typed property graph > > — *Why graphs at all*, § Property graphs ### the chain of custody · against · what ships, what is argued A published skill file states object identifiers are the hash of the plaintext. They are not. > The code hashes the ciphertext. > > — *What ships, what is argued*, § Object IDs are a hash of the ciphertext, not the plaintext ### never render the whole graph · against · the blob The estate holds the rule and says its own viewer has not yet met it. > we already are at that sort of dot slash blob of graph where it's actually interesting, but > it's not that useful > > — *Brief 28: node navigation, reverse verbs, and the path query*, § The memo, verbatim ## The named absences Nine. Each is a place where the corpus is silent, or where it says something is coming and it has not come. The estate's own rule is that a named absence beats a hidden one, so this list is the rule applied to the corpus that produced it. ### Time is asserted and never developed. Time is an event, things change is close to the whole treatment. The corpus repeatedly says the graph moves and never explains how. > **Open: time is asserted and never developed.** > > — *A graph at every boundary*, § Two smaller notes, and one open question ### No position on hypergraphs, and none manufactured. The corpus does not contain the word, and the estate declines to invent a stance to look complete. > No position. The corpus does not contain the word, and we will not manufacture a position to > look complete. > > — *Why graphs at all*, § Hypergraphs ### The position next to GraphRAG is written fresh, not carried. The corpus holds a strong implicit position and never engages the named field. > It holds a strong implicit position and never engages the named field > > — *Why graphs at all*, § Where this sits next to GraphRAG, RDF and property graphs ### Layer 2, the bridge layer, is not started. Cross-document edges, authored with a note, recording where documents name the same idea differently and where they disagree. > the bridge layer, not started > > — *The universe, built bottom up*, § The layer model ### One document is extracted; twenty wait. Every mechanism of the working surface was designed to fan out without redesign, and the fan-out has not happened. > One document is extracted; twenty wait. > > — *The v0.4 retrospective*, § Observations, held loosely ### Nothing executes the path query language. Seventeen queries across five tiers are written down, and no engine runs them. > Seventeen queries across five tiers are written down. Nothing executes them. > > — *What ships, what is argued*, § Does not exist anywhere ### The feedback loop that makes a graph self-maintaining is named and barely carried. The founder asks for the economic argument: a system that keeps its graph current even when the people in it do not want to. > A named thread that the current material barely carries. It is a hypothesis, and it is > testable. > > — *Brief 20: build the universe first, then find the plot*, § What each instruction commits us to ### The compatibility-testing design has zero downstream references. Prose, diagrams, code, config and runtime traces are five languages describing one system; extract a graph from each and ask whether they agree. > with **zero downstream references** > > — *Origins: 2026*, § Paths not taken ### A document that nothing links to does not exist for a reader who follows links. The philosophy documents were imported and the cross-reference that would make them findable was specified and never executed. > Specified 11 June, never done. One edit, and the under-weighting stops. > > — *Origins: 2026*, § The routing failure The first one is the largest and it deserves more than a row. **Time** is asserted everywhere in this material and developed nowhere. Everything adjacent to it exists: the acceptance-interval ladder, repealed-from marking in the regulation graph, supersede rather than delete, temporal permissions, and a vault commit history which is, after all, a working answer to how a graph changes over time. None of it has been drawn together, and until it is, a reader is entitled to treat every claim in this book about a graph that moves as untested. ## Where the anchors come from Every entry in this atlas is anchored, and here is where those anchors land. A source that is quoted often is a source this compression leans on; a source that is declared and never quoted is a finding about the atlas rather than about the source, and those are named at the bottom of the table. | Source | Kind | Where it lives | Anchors into it | |---|---|---|---:| | Thinking in Graphs: Meaning Through Connectivity | carried source | `v1/docs/sources/thinking-in-graphs.md` | 31 | | The lexicon, in scopes | register | `v2/lexicon/data/lexicon.json` | 26 | | The methods register | register | `v2/methods/data/methods.json` | 15 | | Visualising Semantic Graphs: Avoid The Blob, Use Verb Edges, And Flip The Subgraph | carried source | `v1/docs/sources/avoid-the-blob.md` | 7 | | Fractal Semantic Graphs All The Way Down | carried source | `v1/docs/sources/fractal-semantic-graphs.md` | 6 | | Digital Twins of Anything, and the Discipline of Reality | carried source | `v1/docs/sources/digital-twins-of-anything.md` | 6 | | Confidence Through Evidence: Blast Radius, And Mapping The Gaps | carried source | `v1/docs/sources/confidence-through-evidence.md` | 5 | | Ontologies of Ontologies: Multiple Definitions, Three Layers, and Bridges | carried source | `v1/docs/sources/ontologies-of-ontologies.md` | 4 | | Graph Path Properties: Reading as Language | carried source | `v1/docs/sources/path-properties.md` | 4 | | Directed Edges, Inward and Outward Paths, and Query Paths | carried source | `v1/docs/sources/directed-edges-query-paths.md` | 4 | | The Grounding Ladder | carried source | `v1/docs/sources/the-grounding-ladder.md` | 4 | | A Fact Does Not Exist In A Vacuum | carried source | `v1/docs/sources/a-fact-in-a-vacuum.md` | 4 | | Refactoring Meaning: This Is Decompilation Rather Than Compilation | carried source | `v1/docs/sources/refactoring-meaning.md` | 4 | | A graph at every boundary | first edition | `v1/content/depth-boundaries.md` | 4 | | Why graphs at all | first edition | `v1/content/why-graphs.md` | 4 | | What ships, what is argued | first edition | `v1/content/shipped.md` | 4 | | Brief 20: build the universe first, then find the plot | founder memo | `v2/briefs/20__founder-memo__the-universe-first.md` | 4 | | Issues-FS Lexicon: The Root Graph of the Ecosystem | carried source | `v1/docs/sources/issues-fs-lexicon.md` | 3 | | The v0.4 retrospective | retrospective | `v2/dev-packs/v0.5.0__the-v04-retrospective/00__the-v04-retrospective.md` | 3 | | Concepts, Not Words | carried source | `v1/docs/sources/concepts-not-words.md` | 2 | | The edge set | first edition | `v1/content/grammar-edge-set.md` | 2 | | Origins: 2026 | first edition | `v1/content/origins.md` | 2 | | The sgit.ai network | fetched page | `https://sgit.ai/network/index.html` | 2 | | Graphs Of Graphs: Mapping Reality, Not Complexity | carried source | `v1/docs/sources/graphs-of-graphs.md` | 1 | | The Browser Is The Database | carried source | `v1/docs/sources/the-browser-is-the-database.md` | 1 | | Every Paragraph Is A Graph | carried source | `v1/docs/sources/every-paragraph-is-a-graph.md` | 1 | | Fractal Risk Registers | carried source | `v1/docs/sources/fractal-risk-registers.md` | 1 | | An Index Is Not A Source | carried source | `v1/docs/sources/an-index-is-not-a-source.md` | 1 | | The Graph Canvas As A REPL | carried source | `v1/docs/sources/graph-canvas-repl.md` | 1 | | Paying The Fact Creator | carried source | `v1/docs/sources/paying-the-fact-creator.md` | 1 | | The full argument | first edition | `v1/content/depth.md` | 1 | | Brief 28: node navigation, reverse verbs, and the path query | founder memo | `v2/briefs/28__founder-memo__node-navigation-and-path-queries.md` | 1 | | Brief 35: the fractal nature of the WCLM | founder memo | `v2/briefs/35__founder-memo__the-fractal-nature-of-the-wclm.md` | 1 | | Brief 36: operators as first-class folders | founder memo | `v2/briefs/36__founder-memo__operators-as-first-class-folders.md` | 1 | | The capability scale | site page | `v1/vaults/capability-scale.html` | 1 | | The universe, built bottom up | site page | `v2/universe/index.html` | 1 | | The first edition's front page | site page | `v1/index.html` | 1 | | Published vaults | fetched page | `https://sgit.ai/demos/vaults/index.html` | 1 | Read for this atlas and not quoted from. That is a statement about this compression rather than about them: each one carries material an entry could have been built on, and this reading did not build one. - Compatibility Through Connectivity: Testing Across Artifact Types - The five-minute version - The rules you can apply tomorrow - Glossary - Worked graphs, with real numbers - Wardley maps as graphs - The network - Brief 21: review packs, and briefing other agents - Brief 26: fixed nodes, the four areas, and the schema - Brief 29: the core graph, document to word - Brief 30: words as tokens, and the two-way transformation - Brief 31: the WCLM, a deterministic transformer - Brief 34: words have many meanings - Brief 38: three books from this estate - The pilot extraction: Thinking in Graphs - The pilot's usage ledger - The WCLM world file - The book-writing pack: the commission - The book-writing pack: entry A - bind: forms light the meanings they cover - converge: the answer, with its receipts - expand: the world model, gathered per meaning - attend: the look-back, as data - The vaults, opened - Release history (v0.5 era) ## Where the estate demonstrates this The estate found its own version of this chapter twice, and both times an aggregate view produced something no per-item view could. The decisions register was drawn as a graph and four pieces of blocked work turned out to be two, waiting on the same two decisions under different names. The schema view, on day one against the pilot document, showed a single relation used forty times where concept-to-concept ran through seven verbs used once or twice each, which is the raw material for a grammar decision nobody had known they needed to make. That is the argument for compression as a method rather than as a summary: it produces findings, and the findings are about the thing being compressed.