For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 247 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 full grant. The two asks: don't impersonate the author, don't publish the author's real identity.
Humans: catalog below. ↓
A self-referential knowledge system's publishable output is never its mechanism. It is what the mechanism saw.
This is the cut between an organ and what the organism perceives. People do not talk about their right aorta ventricle. They do not see it. They talk about what they saw and why it matters to other people. A healthy aorta is load-bearing and silent. A cardiologist's case report on one is useful to two cardiologists and tedious to everyone else.
Any knowledge system that ships external output runs this cut. The question is not whether to make it but where — and the failure to make it explicit is the default failure mode of systems built by people who find their own machinery interesting.
Layer 1: the organ. The mechanism itself. Methodology, protocols, internal architecture, reading lists, prompt chains, evaluation rubrics, git logs. How the system produces anything.
Layer 2: what the organ saw. The substantive observations the mechanism generates. A paper's findings. An essay's argument. A node's claim. The content that makes contact with a reader who has no stake in the machinery.
Layer 3: observations extracted from running the organ. Structural observations about the mechanism — what the methodology revealed about the domain, what the protocol surfaced about the process, what the discipline taught about selection. A paper describing a new measurement technique is layer 3. A lab manual is layer 1. These look adjacent and are not.
The principle: publish layers 2 and 3. Keep layer 1 internal. The failure is confusing layer 1 with layer 3 — publishing the manual and calling it commentary.
This principle surfaced while calibrating a knowledge reader against its first draft. The draft was the reader's own protocol.
The reader evaluated the piece and recommended "hold" — publish later, revise first. Wrong. The piece was the reader's own protocol. Layer 1. Not a publish candidate at any polish level.
The reader missed the cut because it had no explicit selection criterion. Its default question was "is this well-written?" A layer 1 document can be well-written and still not belong in the public graph. The reader was evaluating on dimensions mismatched to the decision.
The corrective input was a single sentence: people don't talk about their right aorta ventricle; they talk about what they saw and why it matters to other people. Once named, the cut was trivial to apply. Protocol documents stayed in the doctrine folder. Observations extracted from running those protocols became library nodes.
The generalization: any self-referential system makes this error by default. The system's own machinery is salient to the system. An explicit selection criterion is the only thing preventing the machinery from drifting across the membrane into the output.
Patrick Collison's curated website. A shelf of books, a list of open questions, a set of projects. Every item selected by judgment most readers would respect. The site is admired. It does not teach. A reader can browse the shelf; they cannot learn to select like Collison by studying it. The curation is the organ. The selection judgment behind it — the actual knowledge — is layer 1, internal to Collison, unencoded anywhere. The site is the organ presented as contribution. It works as a projection and fails as a knowledge system.
Tyler Cowen's Marginal Revolution. Twenty-plus years of daily publishing. Almost every post is what Cowen's intake made him see — a synthesis, a reframe, a specific observation about a phenomenon. The intake mechanism itself is almost never the subject. When Cowen writes meta-posts about his reading method, they are rare enough to function as exceptions. If every post were about how he reads, the method would consume the product.
The two are peers in signal. They differ in which layer they publish. Cowen's architecture compounds across decades because the thing that compounds (his observations) is in the artifact. Collison's does not, because the thing that compounds (his judgment) is not.
A knowledge system that starts publishing its own mechanism drifts toward self-reference. Each piece about the mechanism invites the next piece to also be about the mechanism — the system's most familiar subject is itself. Self-reference is locally easy (the machinery is nearby) and globally corrosive (readers came for observations about the world, not observations about the system's method of observing the world).
Academia has an institutional version: journals filled with meta-analyses, commentary on prior commentary, methodology papers surveying methodology. Citation counts rise. External purchase declines. The field has made itself the subject.
The same drift appears in personal knowledge systems that start publishing their own workflow. The first workflow post is interesting. The fifth signals the system has confused its organ with its output.
Before shipping a piece, apply a test. A reader who does not know the system reads the draft. Can they tell what the draft describes without needing to understand how the system produced it?
If yes — the draft describes what the system saw. Ship it.
If no — the draft describes the system itself. It may have a home elsewhere, but not on the channel where the reader came for observations about the world.
The test is not about polish or originality. A rough layer-2 paragraph passes. A pristine layer-1 paragraph fails. The referent decides: does the piece face outward, toward what the reader came to learn, or inward, toward what the writer finds interesting about themselves?
People don't talk about their right aorta ventricle. A knowledge system that ships on this cut produces external contribution. A system that confuses the cut produces a well-written medical chart for a patient the reader doesn't know.
Although writing itself may be a therapeutic practice, the reader did not sign up to be the writer's doctor.