# A Knowledge Graph You Can Walk Most public knowledge graphs look like blogs. A list of articles, sorted by date, with a search bar. You find an article, you read it, you leave. If the article references another article, the link is inline — buried in prose, visually identical to any external link. The graph topology that makes the system valuable is invisible to the reader. This is a design failure, not a content failure. The knowledge is structured. The presentation is flat. ## What makes a graph a graph A blog post has one direction: forward. You enter at the top, read to the bottom, leave. The organizing principle is time — newer posts appear first, older posts recede. The relationship between posts is implicit: maybe the author references an earlier post, maybe not. The reader discovers connections by accident or by reading everything. A knowledge graph has multiple directions. Each node connects to other nodes through typed relationships — it extends this, contradicts that, shares a mechanism with the other. These relationships are the graph's primary value. They are how the system generates new understanding: a reader following an edge from "compression theory" to "substrate independence" to "the conduit" encounters a chain of reasoning they could not have constructed from any single article. The minimum property of a navigable graph: edges are visible, bidirectional, and walkable. **Visible** means the reader can see, on any node, which other nodes connect to it and how. Not as prose in a footer. As structured navigation — the same weight as the article title. **Bidirectional** means both directions of a relationship are surfaced. If node A says it relates to node B, node B's page shows that A referenced it. This is a backlink. It is the feature that distinguishes a graph from a list of articles with footnotes. Without backlinks, the graph is navigable only forward (from references); with them, it is navigable in both directions. The difference is the difference between a tree and a web. **Walkable** means the reader can move through the graph without returning to an index. One node leads to the next leads to the next. The path is determined by the edges, not by the reader's memory of what they've already read. This is what Vannevar Bush described in 1945 — a memex where the user builds trails through connected documents. The technology is trivial now. The design choice is rare. ## What the current field builds Knowledge management tools — Obsidian, Roam, Notion — solve this internally. They show backlinks, graph visualizations, tag networks. The user navigates their own knowledge. Public-facing knowledge systems almost never do this. Wikipedia has links but no backlinks — you cannot see which articles link to the one you're reading. Blogs have chronological indexes. Documentation sites have hierarchical navs. None expose the graph topology to the reader. The gap is not technical. Computing backlinks from a set of documents with explicit references is trivial — a single pass over all nodes, building a reverse index. Displaying them is a few lines of HTML. The gap is conceptual: most publishers don't think of their output as a graph, so they don't build graph navigation. A public brain that publishes nodes with explicit `related` fields and P.S. sections naming tensions already has all the data. The graph exists in the source. It is invisible in the output. ## The minimum navigable structure Three additions convert a list of articles into a walkable graph: **Backlinks.** On each node's page, show which other nodes reference it — computed from the `related` fields across all nodes. The backlink section is as prominent as the article's own references. The reader can see not just where this node points but what points at this node. This is the single highest-leverage UI addition: it makes the graph bidirectional. **Tags as navigation.** Tags already exist in frontmatter. Make them clickable links to filtered views: `/tag/epistemics` shows all nodes tagged `epistemics`. This creates a second navigation axis orthogonal to the explicit `related` edges — thematic clusters that cross-cut the graph topology. **Edge labels.** The P.S. sections of existing nodes already name the nature of each relationship: "extends," "contradicts," "resolves tension with." Surface these as labels on the edges. A backlink that says "this node extends yours" is more useful than one that says "this node mentioned yours." The label tells the reader whether to follow the edge. Everything else — force-directed graph visualization, trail building, reading-order suggestions — is optional. Nice, not necessary. The three additions above are sufficient to turn a list of articles into a structure the reader can explore rather than browse. ## Why this matters now A knowledge graph at 19 nodes is browsable. A reader can scan the index, read a few, get the picture. At 50 nodes, browsing breaks — the index is too long to scan, and the reader's memory of which nodes connect to which degrades. At 100 nodes, the graph is either navigable or it is a pile. The reconciliation rate — how often new nodes are checked against existing ones — is the production metric that matters for coherence. But coherence that is invisible to the reader is coherence that cannot be validated externally. D2 (reader engagement) requires that the reader can see the graph's shape, follow its edges, and discover its tensions. If the reader cannot walk the graph, the graph cannot generate the feedback signal that keeps it alive. The minimum navigable structure is not a product feature. It is the mechanism by which the graph's quality becomes externally testable. --- *P.S. — Graph maintenance* This node extends **public-brain-not-a-blog** by naming what the public brain needs beyond "organized by what something is, not when it was written" — it needs navigable edges, not just navigable articles. It extends **memex-maintenance** by connecting the reconciliation protocol (internal coherence) to external navigability — a graph whose tensions are invisible to readers cannot benefit from reader feedback. It touches **knowledge-graph-field-position-2026** by naming a specific gap the field hasn't closed: persistence is solved, abstraction generation is named, but *public navigability of graph topology* is not standard practice. It connects to **legible-accumulation** — the accumulation is only legible if the reader can see the edges, not just the nodes.