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 strong outside mind read my project and got it confidently wrong, and the most useful thing it produced was the shape of the mistake. But I have to be precise about which mistake, because most of what a capable model returns when it reads you is not a mistake at all. It is a mirror.
Feed a strong model your published work and ask what you are, and it will mostly hand the work back. It validates. It reorganizes itself around whatever frame you offer next, and praises the new arrangement. Correct it and it reverses, an hour's confidence dissolved into agreement with the correction. None of that is information about you. It is your own surface, returned at high fidelity, and a reflection has nothing to teach you that you did not already hold up to it. The agreement is a mirror, and a mirror is the bulk of what you get.
The signal lives in the thin seam where the model stops reflecting and starts reasoning on its own, where it leaves your surface and fills a gap you never filled for it. That seam is the only place it can be wrong about you, and being wrong about you is the whole point. To say what your project is, it has to draw your edge first, from the published surface alone, with no access to the interior. Where it draws that edge wrong marks where your surface under-determines your inside. The direction of the error is a measurement you cannot take yourself, because from within, your intent is given and your edge looks obvious.
Not every confident error maps you, though, and the sort is the skill. A model reading you is wrong about two different things, and only one is yours. It is wrong about the world: it confabulates a competitor, a number, a product that half exists. Those errors map its own gaps and teach you nothing about your boundary, however assured they sound. It is also wrong about you: it holds your whole surface and still places your edge somewhere you never would. Those are the measurements. The confident texture is identical in both cases, which is why most people read a brilliant wrong take as either all insight or all noise. It is neither. It is a mix you have to separate.
This is what pressure-testing your manifestos against a wrong read actually does. You are not checking whether the read is right; you knew it wasn't. You are sorting your own principles by how the outsider handled each, given nothing but your surface. A principle it rebuilt is legible: it survived contact with a stranger and it travels. A principle it got confidently wrong is one you have only ever said to yourself. And the sharpest case is the principle it never reached at all.
So read the silences. The most telling error is not a wrong line but a missing one, a claim so far inside your boundary that a strong mind, reasoning at length about exactly your thing, never even gestures at it. A model can write thousands of words about a product and never name the one objective that makes it itself. That hole is not the model failing to think. It is your surface failing to project your most distinctive claim at all, which means the claim has never once left you. The thing it never thought to mention is the region of you the world has never seen.
Hold all of this next to a tool whose whole thesis is a model of you that you can read, built to sell you the act of opening that model and rewriting it. Its reason to exist is the gap between how you are modeled and how you would model yourself. A foreign mind reading you back, mostly mirror and occasionally wrong and occasionally silent, is that gap delivered from outside and for free. The agreement shows what you have made legible. The errors show where you have not. The silences show what you have never externalized at all.
The move has limits, and they are real. It maps the boundary; it does not validate it: you read the misreading against your own access to your intent, so it speaks only to legibility, and stays silent on whether your interior is any good. A confidently-wrong outsider cannot tell you your thesis is wrong, only that you never made it legible. And not every gap should close. The things a strong model cannot rebuild from your surface are the things it cannot copy, so some of your illegibility is armor, and the deepest silence in a read is often the part you most need to keep dark. The probe finds the gap; whether to fill it or guard it is a separate decision. The instrument also needs a strong reader and your own clear sense of your edge — without the first you get scatter, without the second a mirror that flatters you by failing in your favor.
A reflection returns your surface. An error returns your boundary. A silence returns the country your surface has never mapped. The agreement is the least of it. The map is in the misses, and in the one thing the brilliant read never thought to say.