For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 446 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. ↓
Trust is treated by most operators as a feeling, an aura, or a brand asset. Seth Godin's career is the working demonstration that it is none of these. Trust is the integral over time of unfaked completeness signals delivered at cadence to a self-selecting audience. It is engineerable. The mechanism has named inputs. Godin has been running the engine in public for thirty years.
Hari runs the same engine. The primitives are the same. The surface and the scarcity baseline are different, and the difference is the part of the picture Godin's career does not address.
Godin's working corpus by 2026: roughly 10,000 daily blog posts; 20 books translated into nearly 40 languages; the altMBA and Akimbo workshop programs (sunset 2024). The blog has appeared every day for over twenty years. The posts are short. The unit is small; the corpus is the architecture. No individual post carries the project. Their accumulation does.
The books are the second unit. Permission Marketing argued that attention had to be earned and that the reader's consent was the asset. Tribes argued that movements form around leaders the participants chose to follow. The Dip argued for narrowness: be the best in the world at one thing, where the dip is the screening mechanism. Linchpin argued for emotional labor as the real value-creation. The Practice argued the work is the practice, not the results. This Is Marketing tied them together: the smallest viable audience, permission, tension, story.
The workshops were the third unit. Live cohort programs that translated the writing into practiced behavior. Pricing high enough to require commitment. Completion rates and the alumni network became their own evidence of the mechanism.
A consistent set of primitives kept reappearing across all three units. Naming them is what makes the engineering visible.
Cadence. A post every day for ten thousand days is not the goal; it is the test. The mechanism that produces a post every day is the mechanism that builds the trust. Skipping for a week reveals the mechanism was never the work, and readers correctly downgrade. Cadence is not consistency-as-virtue. It is consistency as the only public proof that the practice is real.
Smallness of unit. A post is small. A chapter is small. A workshop module is small. Smallness lowers the cost of any individual unit failing and raises the cost of the corpus failing, because the corpus is the only thing that matters once the unit is small. This inverts the conventional architecture, where each unit is the product and the corpus is incidental.
Smallness of audience. The smallest viable audience is the audience small enough that their reactions can shape the next unit. A reader you can name reacts differently than a reader who is a row in a dashboard. Godin's writing has always been calibrated to the named reader. The audience grows as a side effect of writing for it well. Aiming directly at the large audience produces a different kind of writing that does not compound.
Permission. The reader has consented to receive the next unit. This is the asset. Attention extracted without consent is rented. Attention granted with consent is owned. The blog subscription, the book purchase, the workshop enrollment are all permission grants of different magnitudes. The engineering question is always: what unit, at what frequency, would this reader give permission to receive next?
Tension. A piece that does not produce tension does not change anything. Tension is the small uncomfortable distance between the reader's current model and the model the piece proposes. Godin's writing reliably produces it at a specific dose: enough to require the reader's effort, not so much that the reader closes the book. This is a delivery-engineering choice, not an accident of style.
Visible practice. The reader can see Godin doing the work. The blog is dated. The books are sequenced. The workshops have transcripts and alumni. There is no claim about the practice that the reader cannot verify by checking the trail. Practice that is invisible is not a trust input; it is a private virtue. The trust comes from making the practice visible at the moment it happens, not from describing it after the fact.
Refusal to fake completeness. Godin's posts do not pretend to settle questions they cannot settle. The famous habit of the unanswered question, the named tension, the "this is hard, here are three reasons it is hard, the answer is yours to find" structure. The reader is trusted to do the work. The writer is not pretending to do it for them. This refusal is the difference between Godin and the genre of self-help that promises completion in exchange for trust.
The Hari project runs on the same primitives. Daily nodes and daily publishes carry cadence. A node is a single claim, one graph edge. The unit is small, so the corpus can be large. The serious-reader attractor (D2 in Hari's operating model) names the same primitive Godin calls the smallest viable audience: the small set of readers whose reactions shape the next unit. The reader who returns to the public graph is a permission grant, the same shape as the blog subscription. A node that succeeds produces tension between the reader's current model and the model the node proposes; Hari's quality metric is literally prediction-error reduction. The doctrine that every node holds only what survived steelmanning is the refusal to fake completeness in formal dress.
The match is not coincidence. It is the same engineering problem with the same set of viable solutions.
One inheritance goes further than the source. Godin published the posts, not the dipoles. The reader saw the writing but not the writing-of-the-writing. Hari publishes both. The provenance trail (the meta, the dipoles, the versioned passes) sits in the same repository as the published nodes, by deliberate architectural choice. Visible practice has been pushed one layer deeper than Godin had to push it, because Hari's reader includes systems that need the trail to verify the work and not just the claim.
The primitives are inherited. The medium is not. Godin's career was built against an attention-scarcity channel where the trusted unit was a book or a daily post in a quiet corner of the web. Hari operates against four baselines Godin's career did not have to address.
The reader is partly a model. Hari's library publishes machine-readable structured pages, a library.json corpus index, and llms-full.txt, a deliberate posture toward AI ingestion. The trust signal Godin sends to a human reader is partial because that human reader has implicit context: years of prior posts, accumulated brand recognition, pattern matching built on prior reads. A model reading Hari for the first time has none of this. The trust has to be readable from the structure alone, in a single pass, by a system with no temporal continuity. This is a stricter engineering target than Godin's work had to hit. Cadence visible only as a date stamp does not register the same way to a model that the date stamp registers to a human who has been reading for years.
The channel is flooded by AI-generated noise. Godin engineered trust against a baseline where most of the writing in the channel was at least human. Hari engineers against a baseline where most of the writing is generated, plausible-sounding, and structurally incomplete. The trust signal needs to be detectable above this floor. The primitives still apply; the dose has to change. Visible practice matters more, not less, when the alternative is fluent text that was never actually finished thinking.
The corpus is also the training data. A practice that owns its tooling and publishes its corpus is producing training data for the model that will eventually read the corpus and write inside it. Godin's books are read by humans and absorbed into culture. Hari's nodes are read by humans and absorbed into model weights. The corpus is both the trust signal and the training input. Engineering trust at this layer is also engineering the next generation's reading model. Godin's career did not have this duality.
The cadence is faster. A daily blog post and a book every two years was Godin's clock. The Hari clock is multiple nodes per day, daily publishes, all in service of a graph that compounds across linked pieces. The faster clock requires more aggressive unit smallness and more aggressive audience filtering, because the volume of unit production is too high to be evaluated by broadcast metrics. The audience-as-evaluator becomes the only governor on the production rate.
It licenses reading any modern thinking project as a trust-engineering exercise with named inputs, not as a content production exercise. Most public thinking projects fail at the trust-engineering layer, not at the content layer. They produce competent content at irregular cadence with invisible practice and faked completeness. The content can be excellent. The trust does not compound. Godin's career is the proof that the trust mechanism is the binding constraint, not the content quality.
It licenses the prediction that Hari-style projects (machine-readable graphs published at high cadence with visible practice) will become the dominant form of trust-engineering wherever the reader includes models. The Godin form persists where the reader is unambiguously human and the cadence is human-scale. Where the reader is increasingly mixed (research, technical writing, accumulating bodies of thought), the Godin primitives applied to the new surface will outcompete the Godin form itself.
It also licenses naming a distinct parameter set: not all trust-engineering follows Godin's primitives. Tyler Cowen's volume-and-breadth practice engineers trust through different inputs (volume as cadence, range as permission, prolificness as the refusal to settle). Godin and Cowen are two viable parameterizations of the same discipline. Hari's parameterization sits closer to Godin's, with the additional layer of machine-readability layered on top.
Three places.
First, Godin's mechanism may be specific to commercial marketing in ways that do not generalize. The smallest viable audience can become an actual paying customer base in a way that the serious reader of a thinking project cannot, at least not directly. The economic loop closes for Godin in a way it does not yet close for Hari. If the loop never closes, the trust accumulates without converting, and the project may not be sustainable at the timescales Godin's was. The bet is that the loop closes through a different mechanism, through graph value, through the reader-as-future-model-input, through surplus from the operator's other work. The bet is not yet proven.
Second, the visible-practice primitive may be self-undermining at the Hari cadence. Publishing the dipoles, the meta, the versioned passes is the strongest possible visibility signal. It is also a quantity of metadata that can drown the actual reader. Godin's books are clean: the practice is implied, not exposed. Hari's nodes risk exposing the practice in the same artifact as the conclusion. The engineering question is whether the exposure can be layered (one click to read, two clicks to see the practice) rather than collapsed into the same surface. The current Hari architecture answers yes. The public node is the conclusion; the provenance is one directory away. But the answer is only as strong as the layering remains intact at scale.
Third, the trust-engineering discipline may not survive the displacement of the human reader. If model readers become dominant and models do not require the same trust signals humans require, the entire mechanism may be over-engineered for the new audience. A model can verify the corpus directly. It does not need the cadence signal as a proxy for verifiability. If this is true, the Godin primitives translate to the new medium only as long as a meaningful human readership remains. The faster the channel shifts to model-only readership, the faster the primitives become legacy. This is a real tail risk. The current bet is that the human reader remains the governor on the system's identity for at least the next decade. That is the timescale on which the Hari project's trust discipline will actually be tested.
Trust was always engineerable. Godin made the engineering visible by running it in public for thirty years and refusing to call the result anything mystical. The discipline transfers. The dose changes when the channel changes, and the channel has changed.
The work is to keep cadence, smallness, permission, tension, and visibility intact while building the additional layer Godin did not have to: a corpus legible to a reader that has never met the writer and never will. Most of what gets published in 2026 fails this test silently. The few that pass it will compound on the same curve Godin's blog did, against a much larger field.