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 communication has two layers: the carrier (what kind of object the communication is — text on a page, video on a feed, conversation in a room, node in a graph) and the message (what the sender wants the receiver to take away). Analysis usually focuses on the message. The structural finding is that the carrier shapes what messages are possible, expressible, or even thinkable in that medium.
This is not "the medium is the message" — McLuhan was making a different claim about media homogenizing content. This canonical names something more specific: the carrier has affordances and constraints that pre-shape the message, often invisibly. A message that violates the carrier's affordances does not get worse — it does not get sent at all.
Each carrier has an affordance set. A 280-character tweet affords brevity, density, single-claim assertions; it does not afford multi-step argumentation. A 3000-word essay affords developed argument, paragraphed structure, gradual revelation; it does not afford the sharp punchy hook. A real-time conversation affords back-and-forth correction; it does not afford precise composition.
When a sender attempts a message that does not fit the carrier's affordances, the sender unconsciously deforms the message until it fits. The deformation is invisible to the sender — they think they sent the message they intended. The receiver sees the deformed version, which may differ substantially from what was intended.
This is the structural claim: senders mistake "I tried to send X" for "I sent X." The carrier silently edits.
the-conduit names that the channel matters — yes. conduit-inversion names that sometimes the channel inverts what the sender intended. register-as-substrate-fit names that voice has to fit the substrate. anti-mimesis names what to do when the channel selects for mimics. This canonical names the general structural distinction between carrier and message — the analytic separation that lets all the other claims be precise.
Without the distinction, statements like "X is bad communication" mix two different failures: bad message in good carrier, vs good message in wrong carrier. They have different fixes. Naming the distinction lets the diagnosis be precise.
For senders: before composing the message, audit the carrier. What does this carrier afford? What does it suppress? Will the message I want to send survive transmission through this carrier? If not, the choice is not "write better"; it is "different carrier, or different message."
For receivers: when receiving, ask "what carrier did this come through, and what would that carrier have suppressed or amplified?" A speech sounds passionate; a transcript of the same speech reads strident. Same content, different carrier, different impression. The receiver who only sees the transcript may form an impression the speech never produced.
For analysis: critiques that conflate carrier-effects with message-effects miss the load-bearing variable. "Why is discourse so polarized?" — partly the messages, but mostly the carriers. Twitter's affordances pre-select for polarizing messages; the messages are downstream of the carrier choice. Changing carriers (long-form, in-person, structured forum) changes the messages without anyone changing their minds.
For Hari: the graph carrier has different affordances than blog or social. The graph affords cross-references, structural inference between adjacent nodes, compressed claims that gain meaning from context. It does not afford the single-essay hook, the personal arc, the cumulative development across paragraphs. A piece written for blog-carrier and pasted into a node will read as too thin (no context-leveraging) or too long (no cross-references). The graph requires its own composition. Hari's pipeline writes for the graph carrier deliberately; cross-surface translations require recomposition.
The phase-change finding lives at this layer: the procedure by which nodes get written is itself a carrier for the corpus's structural shape. Symmetric intake is one carrier (proposes native canonicals first); asymmetric intake is another (fits to existing). Same content, different procedural carriers, different corpus-structure outcomes. The procedure-IS-substrate finding is the carrier-vs-message distinction applied recursively to the writing process itself.
Not "the medium determines everything." Skilled communicators bend carriers within their affordances. The message-side has real degrees of freedom. But the degrees of freedom are bounded by the carrier in ways the sender does not see without explicit audit. This canonical names the audit move.