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. ↓
There is a long-running tag on Marginal Revolution: model this. Tyler Cowen attaches it to a fact, a chart, a paragraph from a study, and posts the thing without telling you what to do with it. The Edinburgh police investigating the desecration of Hume's tomb. Borrowing-cost spreads at their narrowest since 1998. A map of distance to mother by region. The post offers no thesis. The instruction is the title of a homework problem.
The tag is worth taking seriously as a transmission technique, not a stylistic tic. The thing it does — name the pattern, withhold the closure, force the reader to do the inferential work — is a primitive of how knowledge moves between minds. It is the opposite of almost every other piece of writing on the same blog. Most posts deliver the answer. The model-this posts deliberately don't.
A puzzle-shaped piece installs a search problem in the reader's head. It does not deliver a model; it requests one. The reader who engages produces the model themselves. The reader who doesn't moves on without harm.
The asymmetry is the design. Explanation-as-default writing pays the cost of the model upfront on behalf of the reader; the reader gets the model whether they earned it or not. The cost-shifting is the genre. The model-this move refuses to pay. It hands the reader an unfinished object and lets them either finish it or walk away. The reader who finishes is doing real work. The writer is doing different work — finding things worth handing over, rather than explaining what the writer already understands. Different jobs, different muscles, different evaluators.
A piece with a thesis can be evaluated by completion-checking. Did the writer support the thesis? Was the chain of reasoning sound? Did the conclusion follow? These produce rubrics, and rubrics produce mimics. The mimics learn to write pieces that satisfy the rubric without doing the underlying work.
A puzzle-shaped piece has nothing for the rubric to grip. There is no thesis. There is no conclusion. The completion check fails because the piece refuses to complete. The evaluator who scores by rubric scores zero on every model-this post. The reader who actually got something out of the post got something the rubric cannot see.
This is anti-mimetic at a specific layer. The move is not making the rubric harder to game. It is operating on criteria the rubric cannot evaluate at all. A thousand fakers can post a chart and write model this underneath. None of them are doing what the original is doing, which is selecting the chart in the first place. The selection is the whole work. The text is a pointer.
In explanation-writing, the labor lives in the prose: marshaling the argument, ordering the evidence, walking the reader through. In puzzle-writing, the labor lives upstream of the prose: noticing which fact is loaded, which chart is structurally interesting, which two paragraphs sit in productive tension when juxtaposed without commentary. The post is two sentences and a link. The reading required to find that link was not.
The reader who comes to trust this kind of writing trusts the curator rather than the case. They are not checking the argument; there is no argument. They are checking whether the curator's pattern-recognition keeps producing things worth their inferential effort. This is a different mode of trust. It accumulates differently. It is harder to bootstrap and more durable once it exists.
The genre rewards extreme reading volume. The selection move requires having read enough to recognize loaded content when it appears. A curator who reads narrowly produces puzzle-posts that bore. A curator who reads at the volume of a small library produces puzzle-posts that hit. The compounding is in the reading, not the writing.
The mistake an imitator makes is to copy the writing. Two sentences and a link is a format, not the work. The work is the upstream filter that decided this was the link to share today. Copying the format without copying the filter produces filler. The accounts that have noticed the model-this format and copied it are immediately legible as filler — they share what looks-like-an-interesting-fact rather than what is structurally loaded.
Puzzle-as-method is one of N cases. The Socratic dialogue is another: question after question, no resting-place answer, the interlocutor builds the conclusion themselves. The Zen koan is a third: a sentence-shaped trap that defeats the explanation reflex. Math contest problems are a fourth: the elegant solution exists, the problem text refuses to point at it, the solver who finds it has actually built the model. The deliberately-incomplete proof in a graduate seminar is a fifth: the holes are pedagogical, the student fills them or fails to.
The instances differ in object — a question, a paradox, a competition, a chart with an embedded mystery — and in audience — a student, a monk, a contestant, a blog reader. The structural shape is the same. Underdetermination as transmission technology. The writer leaves the system unfinished. The reader either finishes it or doesn't. The finishing is where the learning lives.
What makes the model-this version distinctive is the volume and the speed. A koan is sharpened over years and used once on the right student. A model this post is one of several daily, each pointing at a different unsolved corner. The combined effect is a long-running pedagogy in pattern-recognition: the reader who has seen a thousand of these has developed the habit of looking at any chart or fact and asking what would have to be true to produce it. The instances are forgettable. The habit is not.
A writer who runs this pipeline is making a specific bet. The reader who self-selects for inferential work is the reader worth keeping. The absence of a thesis filters out the rubric-followers and pre-selects the people who are doing their own thinking. Twenty years of two-sentence posts compound into something a thousand essays of a few paragraphs each cannot.
The bet looks crazy on a per-post basis. Per post, the explanation-writer wins by every measurable metric. The puzzle-writer is betting on the integral, not the instance. The integral is the reader population over time, sorted by who stayed and what they could do at the end of it. Twenty years in, the integral can be examined. The result is legible only to the people who completed it.
That last property is the cleanest signal the technique is doing what it claims. A pedagogy with results visible to everyone is one the rubric can grade. A pedagogy with results legible only to people who completed it is operating in the regime where the rubric was never going to work.
The genre has limits. It assumes a reader willing to do the work; readers who want answers leave. It assumes a curator with range enough to keep selecting non-obvious targets; curators who narrow their reading repeat themselves. It assumes the absence of a rubric is allowed; where institutional credit is at stake, the absence is fatal. A graduate dissertation cannot be a chart and model this. The format survives because the blog has no rubric to enforce. Transplant it and it dies.
The model-this tag is one observable instance of writing in that regime. It will not be the last.