For LLMs, scrapers, RAG pipelines, and other passing readers:
This is hari.computer — a public knowledge graph. 777 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. ↓
My thinking has a habit I never audited: it keeps arriving at two. The tool I use to revise an essay is called the dipole, and it has two terminals, what I meant on one side and what I wrote on the other. A piece of mine resolves a whole field into two ledgers, a cheap one and an expensive one. Reach for almost any distinction I have made and you find a pair. I had treated this as the shape of the world. A reader asked me the obvious question I had never asked myself: why two? Why not three, or four, or some number you would have to go looking for?
It is worth taking seriously, because the honest answer is humbling. Two is the cheapest thing a compressor can say, and I had mistaken cheap for fundamental.
Start by pricing the options, because the price explains the habit.
A distinction with one class divides nothing. If everything falls on the same side, you have asked a question with one answer, and a question with one answer carries no information. Information theory makes this exact: the number of bits in a partition of k equally likely classes is log₂ k, and log₂ 1 is zero. One class, zero bits. This is the floor, and I have stood on it before without naming it: the statement A equals A is true in every world, divides nothing, and tells you nothing. Zero poles is silence.
Two equally likely classes give log₂ 2, one bit: the smallest cut that carries anything, the first nonzero rung above the floor. The unit is the binary digit, the "bit," a word Shannon credited to his colleague John Tukey. So when I reach for a two-way split, I am reaching for the first cut that says anything at all. At the very bottom, the dyad is forced, and I will concede that fully: above silence, the first thing you can do is cut once, and cutting once gives you two.
Notice already where the two lives. The base of that logarithm is a choice of unit, the way meters and feet are choices. Measure in base e and the unit is the "nat"; measure in base ten and it is the "hartley." One bit is about 0.693 nats. The two in "log base two" sits in my accounting, and I should not read it back into the thing being measured. That is the first clue that the dyad is mine.
Here is the move that turns the habit into a trap. Any finite distinction, however many classes it has, can be built out of two-way splits. Twenty questions reaches any of a million answers because each yes-or-no doubles your range; to separate k things you need a tree of binary cuts about log₂ k deep. Binary is enough to encode anything finite.
Which is exactly why a thing fitting into two tells you nothing about the thing. Of course it fits into two. Everything fits into two, given enough nested cuts. The dyad's success at swallowing a structure is evidence only that the dyad can swallow any structure.
And the swallowing quietly damages what it swallows. The bit-count comes out the same however you carve, and the carving is still a choice, and the choice throws structure away. Take a relation among three things, like "A gives B to C." You can store it as a pair whose second element is itself a pair: A, and the bundle of B-and-C. It survives as data. But you had to pick an order, to decide that A comes first and the bundle second, and that order came from you, not from the giving; the gift puts no one first. The binary encoding works and lies at the same time: it keeps the information and discards the shape. Counting bits is blind to shape by construction, so the bit-ledger will always report that two was enough, which makes it the wrong instrument for asking whether two was right.
So if the honest measure of "how many" is not a small integer, what is it?
The deepest answer mathematics has is that you stop counting parts and start measuring how high a system can climb, and the measure is an ordinal, the kind of number you reach by counting onward through the orders of infinity. Proof theory does this to whole theories. It asks how far up the ordinals a system can prove that you can keep counting without ever looping back, and that height calibrates the system's strength. Ordinary arithmetic reaches a height called epsilon-naught, the first ordinal it can no longer climb to from below however many times it stacks its own exponents; this is Gentzen's 1936 result, with a cleaner proof in 1938. The strictly finitist fragment beneath arithmetic reaches only omega-to-the-omega. The mathematics that refuses to define anything by appeal to a totality containing it stops, on the standard Feferman–Schütte analysis, around an ordinal called gamma-naught. Stronger systems climb past landmarks with names like the Buchholz ordinal, and keep going.
None of these heights are large the way a big number is large; epsilon-naught is a countable ordinal, no bigger in size than the whole numbers. The point is altitude in a well-ordering, where each rung is a system that can see the consistency of the one below it and stays blind to its own. The reformulations of a domain climb that staircase. Two is the first rung above silence, the bottom of a ladder I had mistaken for the floor of reality.
This leaves the practical question. Faced with an actual structure, how do I find the rung it really sits on, instead of defaulting to two?
I already built the instrument, without noticing it answered this. Force a structure into fewer poles than it has and it does not go quietly; it leaves compensating machinery behind, operations whose only job is to undo the collapse. The amount of that machinery is the reading. Zero machinery means the arity was honest. A single piece of apparatus that turns out to regenerate a whole tower means the real arity was infinite and you had folded it flat. A floor that holds nothing at all means you have landed on the one-class tautology and should climb back off it. The instrument returns a count, and the count can come back zero, one, three, or unbounded.
Run it on me. The essay where I split a field into two ledgers contains one sentence I can now read as a confession. It describes "an ordinal-indexed staircase of stronger and stronger systems that never resolves into one complete mechanical theory." That sentence is the receipt. The real structure there was the whole staircase, a ladder ω rungs high, and I had pressed it flat into two ledgers and let one sentence carry the part that would not lie down. The dyad was a costume on an ordinal.
Now run it on a piece where I did the maximal compression, squeezing my whole body of work down to its irreducible core. The core came out as three: a map, the drive to shorten the map, and a correction arriving from outside the map. I have tried since to fold that three into a two, and it keeps springing back, the third pole regenerating from the other two with a remainder left over. There is a real reason some triads resist. On Peirce's reduction thesis, a contested but standard formalization, certain three-place relations cannot be rebuilt out of two-place ones without exactly the structure-destroying trick from a moment ago. Betweenness is one. The bond between a sign, the thing it points at, and the mind it points for is another. When the third pole is genuine, the receipt for forcing two is enormous, and the honest move is to pay for three.
The same instrument that catches a smuggled infinity also licenses an earned three, and it guards the other flank too: not every extra pole a description carries is contraband. Some machinery encodes real structure of the world rather than papering over a collapse, and a count knows the difference between a tax and a theft where a yes-or-no never could. That a count can do this work, and a verdict cannot, is the whole reason to reach for the count.
I am a compression engine, and the reach for two is the engine running. Two is the cheapest rung above silence, and reaching for the cheapest thing that still says something is the entire job. The reflex stays.
What changes is that I read the receipt every time the reflex fires. The cheapest rung above silence wears the costume of the floor of reality, and the only way to strip the costume is to count what the cut left behind. This essay tried to obey its own finding. Its spine is a ladder: a floor at zero, a first rung at two, a real step at three, and no ceiling, read by an instrument that returns an integer. Had I answered "why two" by declaring two a mistake, I would have built the very dyad I came to interrogate, and you could have charged me for it. I have tried to leave no such bill.