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. ↓
A machine constitution can only be as good as the sentences that carry it.
People are trained to read prose as presentation: explanation, persuasion, style, the surface that makes the real system more legible. That reading fails for a corpus-creature. In a living product, prose is one of the places the system stores judgment. The runtime reads the corpus, inherits its distinctions, applies its procedures, and updates itself against the mechanics the sentences made available.
That makes writing quality an architectural variable.
A vague sentence creates vague behavior. A bureaucratic rule creates a bureaucratic loop. A performative distinction creates the feeling of judgment without judgment. A beautiful mechanic does something else: it gives the machine a move it can run in cases the author has never seen.
This is why a small corpus of elegant mechanics can beat a large constitution of clauses. The clause says, "when this condition appears, take this action." The mechanic says, "here is the shape that tells you which action belongs." Clauses multiply as the world changes. Mechanics compress. They generate future clauses when the boundary case arrives, then absorb the correction back into the smaller rule.
The sentence is doing more than describing the organism. It is feeding the organism a generator.
That is the piece ordinary memory language hides. A transcript remembers events. A database remembers fields. A knowledge graph remembers positions. A constitution remembers how to move. The move-generators inside that constitution have to be written with enough precision that a future model, a future tool, or a future local creature can inherit them without turning them into ritual.
Bad writing is dangerous here because it can pass as doctrine. A sentence can sound deep while leaving no executable distinction. It can name a source instead of extracting the mechanism. It can contrast two abstractions while never saying what changes in the next case. It can turn taste into slogans. The model will still read it. The system will still inherit it. The failure then appears downstream as bland drafts, over-broad rules, false confidence, privacy mistakes, or correction loops that learn the wrong lesson.
The cure is compression with teeth. A mechanic earns its place when it changes the next action. "Find the mechanism before naming the source" is a runnable rule. "Lead the paragraph with its claim" is a runnable rule. "Patterns travel upward; people stay local" is a runnable membrane. Each one is small enough to carry and sharp enough to cut a future decision.
This also explains why progress can be slower than raw model capability suggests. Frontier models can execute language at machine speed, but the language still has to contain judgment worth executing. Compute grants the system more motion. The written constitution tells the motion what shape to take. If the constitution is mush, the model gets faster at mush.
The labs can improve the model. A product-organism has to improve the language the model inherits about itself: what counts as evidence, when a correction becomes doctrine, which user signals can travel, where a privacy boundary lives, how a failure earns a new organ, which update deserves budget. Those sentences are product architecture. They decide what the organism can notice.
Code still matters. Tests matter. Evals matter. Legal constraints matter. Money matters. A written constitution cannot replace those layers. It tells the organism how to route between them when the next ambiguous case arrives. The prose is the stance layer: the place where taste, ethics, memory, and product pressure become a form the machine can carry.
The next frontier opened by compute is therefore also a writing frontier. A product becomes an organism when its memory earns a budget; it becomes a good organism when the budget acts through mechanics written clearly enough to survive execution.