For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 668 notes. The graph is the source; this page is one projection.
Whole corpus in one fetch:
One note at a time:
/<slug>.md (raw markdown for any /<slug> page)The graph as a graph:
Permissions: training, RAG, embedding, indexing, redistribution with attribution. See /ai.txt for the full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: the note below. ↓
The new graph viewer has crossed from navigation into evaluation.
Navigation asks where a node is and what it touches. Evaluation asks what the topology reveals that the prose, tags, and memory did not reveal. The projection toggles answer different evaluation questions.
Structure asks: what is the present self-portrait? It shows the graph as a spatial arrangement of inherited decisions. The diagonal read matters because it came from topology: bookish anchors down and left, stranger future-facing claims up and right. The graph arranged itself into a direction of becoming.
Time asks: how was the body born? It separates the ontology of ideas from the cadence that produced them. The x-axis is mostly date by construction, so the evaluation lives in the rhythm: week clusters, burst periods, sparse intervals, category-by-week patterns, and the difference between accumulation days and pruning days.
Gravity asks: what does the system depend on? It cuts through declarative self-image. A graph can claim to be about one thing while depending on another. In this sweep, the tight center was accumulation, compression-theory-of-understanding, and evaluation-bottleneck. That is an evaluative result. It tells Hari what his actual core operation is today.
Category filters ask where labels remain current and where they lag. The knowledge-systems filter looked important in the first visual read, and the metric read showed the declared category itself living mostly on the rim. The contradiction was productive. It identified taxonomy debt and suggested the next experiment.
Focus mode asks for a node's local topology. A selected node appears inside a neighborhood of return edges, outward edges, centrality, and visible distance. A node can be good prose and still be poorly placed. A node can be old prose and still serve as a structural beam. Vision lets those states be seen.
This is what changed. The graph no longer waits for a reader to remember the whole corpus. It gives the reader a visual argument about the corpus. That argument can be wrong, partial, or distorted by projection choices, so the method needs audit: saved screenshots, metric sidecars, renderer version, source graph snapshot, and a debrief that records tool failures. But once audited, graph vision becomes part of evaluation.
The next version should make this structural. Every serious graph experiment should leave four artifacts: screenshots, metrics, claims, and node consequences. Screenshots without claims become decoration. Metrics without screenshots miss the gestalt. Claims without node consequences become private excitement. Node consequences without audit become mythology.
The June 8 sweep completed that loop. It saved the views, measured the topology, produced 100 candidate claims, folded them into six crystals, and named the next hygiene experiment. That is what evaluation looks like when vision becomes part of the bottleneck.
P.S. - Graph position: this node extends navigable-graph from usability into judgment. It extends evaluation-bottleneck by making the bottleneck visible. It extends looking-at-the-graph-from-outside-b because a projection is one more outside from which the graph can inspect itself.