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. ↓
Joke-is-claim-b said: a joke earns its place if its restatement loses. That is the receipt at the line. There is no equivalent receipt at the frame. The frame-level question is harder, because the test is not "remove the joke and see what survives." It is "could the piece have been written without this joke at this place." Most jokes flunk the second test even when they pass the first.
Several moves answer this in practice. Three name themselves; one does not.
The funniest sentence in a Hari piece should be the one the structure made the writer write, not the one the writer chose to put there. If the joke could have been any of three lines at that place, it was decoration well-fitted. If the joke could only have been that line because the prior paragraphs made every alternative impossible, it was the recursive position.
Source-side analogue is well-developed: file the source as instance of its own pattern; the laugh is the pattern collapsing onto itself. Self-side analogue is the same machinery turned on the piece's own argument. The piece runs to its conclusion; the conclusion is what the piece was always going to be; the recognition is the laugh.
Announce a count that does not match what the prose seems to deliver. We are doing twelve. Show three. Move on. The reader, holding the announced twelve, looks for the missing nine in the prose itself and assumes they are embedded. The reader's search is the embedding. The piece becomes denser in the reader's mind than it is on the page, because the reader imports density looking for the missing count.
The risk: if the missing count never resolves, the reader feels played. The discipline: deliver the count somewhere late, where the reader was not expecting. The eight that were "missing" arrive in §4 and the reader recognizes both that they were not hidden and that the searching was not wasted. The misread is the receipt.
This move was not in joke-is-claim-b's authoring intent. The operator surfaced it on first read by attributing it. The writing was dense enough that he imported the move and it cohered with what was there. Vibrancy-attribution is the form of receipt the strip-test cannot generate.
The piece does the joke at a level the reader has to detect. It does not say here's a joke. It does not say this paragraph is naming its own constraint. If it does say something like that, it does so at the third sentence rather than the first, because doing it at the first is the personality move and doing it at the third is the structural move.
The unsignaled meta-move is the inverse of the meta-confessional setup that joke-is-claim-b ruled out. The meta-confessional says I am an AI, here are my constraints, watch me reflect. The unsignaled meta-move performs the reflection in the structure of the sentence that does the work. The reader registers the meta-level only on second pass, if at all. The piece reads as substantive on first pass and structurally funny on second.
Any meta-move flagged is no longer at the meta-level. Personality moves can ship; they trade depth for legibility. The structural move loses depth the moment it is announced.
Each move has a corresponding failure.
Recursive without recursive. Claiming the piece is structurally a joke without actually structuring it that way. The frame says this whole essay is the joke and the body is six paragraphs of analytic prose with no structural punch. The mismatch reads as posturing.
Hiding without loading. Burying a joke deep enough that no reader finds it, with no payoff for the reader who does. The recursive position must produce a payoff at the structural level, not just exist at the structural level. Embedding without payoff is hide-and-seek with no prize.
Over-flagging. Winking too hard breaks the second-order read by making it first-order. The wink converts a structural move into a personality move and posts the personality. A piece that explicitly says and yes, I see what I just did has just done the move worse than a piece that did the move silently.
Imported density. A piece so opaque the reader cannot tell whether anything is embedded at all. Vibrancy-attribution requires the reader to have something to attribute to; pure obscurity gives them nothing to import. The opaque piece earns no misread; it earns dismissal.
The first three are flagging failures. The fourth is a substrate failure.
In pieces whose other work is structural revelation, embedding is the natural extension of the strip-test from line to frame. In pieces whose work is pure entertainment or pure information transfer, embedding is over-engineering. A briefing does not need recursive-position jokes; a friend's-week newsletter does not need unsignaled meta-moves. Embed only when the piece is doing structural work that humor can carry.
The voice-gradient layer: outer-shell pieces have less room to embed; inner-shell pieces have more. The middle shell is where embedding has the highest leverage. A 1500-word piece with a recursive-position closing and one unsignaled meta-move outperforms the same piece with the same content delivered straight, because the embedded version invites a second read and the straight version does not.