What each diagram claims, what every moving thing means, and why five formal lenses on one person add up to a single argument rather than a pile of visualizations.
Read in order they make one sentence: these things coexist in me; here is where that puts me relative to the mean; here is what flows in and out; here is how I decide; and here is the daily practice of staying aimed. Container, position, relations, decision, practice.
Identity isn’t a list of interests. It’s an overlap structure — and the interesting parts of a person live in the intersections, not the circles.
Five domains: mind/body, technology, philosophy, society, creation. On their own they’re just topics; a bio could list them. What the diagram argues is that none of the seven things actually worth naming exist inside a single domain. Cognitive architecture only exists where mind meets technology. Phenomenology only where mind meets philosophy. AI alignment where technology meets philosophy. Structural difference where a differently-built mind meets a world built for the default. Power & consent where philosophy meets society. Building your way out where technology meets creation. Making as thinking where creation meets philosophy.
Every label sits on a region that geometrically exists — verified by scanning the actual pixel areas before anything got named. The diagram never claims a region it doesn’t have. That restraint is the epistemics of the page in miniature: a picture that would rather be incomplete than overstate.
Variance is the structure of things, not an error in your model — and individuality isn’t the far tail, it’s choosing your distance from the mean on purpose.
One axis: share of cognition delegated to machines, none to everything. Three fixed marks tell the overcorrection arc — autumn ’25 at the far right (every decision through the machine: efficient, and not entirely mine), spring ’26 at the far left (cold turkey: proved the point, paid in speed), and now, a chosen position right of center, revised monthly. The word overcorrection sits over the swing between the tails — the one place on the page the machine-learning register is allowed to surface twice.
Hovering the bands names them the way a statistician would: the middle is “where most things cluster · base rates matter · the probable”; the tails are “the unexpected · rare events · calibrate for tails.” This is the figure that argues against optimization-as-virtue — the mean is not a target, it’s just where the density is.
Nothing arrives alone, and every edge runs both ways. Neither the passive-consumer story nor the pure-autonomy story survives contact with the actual graph.
Fig. 1 shows what overlaps; this shows what flows. Eleven nodes around a red center — the feed, the machines, Aeon and Psyche, LessWrong, Kierkegaard, Clark & Chalmers, Whittaker, MUBI, Fleet Foxes, the old ghosts — and the hover captions are written bidirectionally on purpose. “The feed ↔ me: it shaped my taste; my attention paid its rent. Neither of us is innocent.” That sentence is the thesis of the whole figure: influence is a trade, not a transmission.
Nodes connect to each other, not only to the center — the feed touches the old ghosts and the machines; Kierkegaard touches Clark & Chalmers. Being shaped by something means being shaped by everything it’s connected to. The graph is also where the taste history stays legible without being confessed.
Values don’t rank in the abstract; they rank per situation. A principle you can’t route is a slogan.
A decision tree that’s actually a values router. Does it touch a person → kindness, and the edge is annotated always — the only “always” on the site. Otherwise: can it be undone? If no → uncertainty (one-way doors get a credence first). If yes, is everyone already doing it? Yes → individuality (the mean moved, not the truth). No → taste (when nothing else claims jurisdiction, judgment does). And a separate short branch, is it anyone’s business? → privacy, marked end of route.
The situation chips let you call the router on a real case and watch it resolve edge by edge. This is the figure most exposed to rationalist cringe, so two guards: no probabilities anywhere on it — putting p = .8 on “kindness governs” would be self-parody, and numbers live in the ledger instead — and a footnote that undercuts the whole apparatus: deterministic here. messier in production.
You don’t arrive at truth. You stay pointed at it while a stream of inputs bends your heading — and the honest score is an angle, not a total.
A breakout game where the ball is a belief. Green bricks (evidence, base rate, contra, primary source, steelman, sleep on it) bend its path toward a truth marker; violet bricks (vibes, the feed, consensus, sunk cost, engagement, hot take, clout) bend it away. The paddle is labeled judgment — your only control, and it only gets to choose the return angle. The readout is the live angular error between where the belief is heading and where truth sits, turning gold under 25°: directionally correct.
Each figure is a different formalism aimed at the same person, and none of them is sufficient alone. What persists across all five coordinate systems is the thing the site is actually about.
Set theory can show what coexists but says nothing about degree. Statistics can place you on an axis but flattens you to one number. A graph shows relations but not decisions. A decision tree shows choices but pretends they’re clean. A dynamical system shows the practice but no structure. Run all five and the residue — the red dot that appears in every one — is the argument. A self can be modeled without being reduced, which is the both/and thesis executed rather than stated.
| device | what it does across every figure |
|---|---|
| the red dot | Same color, size, and breathing in fig. 1 (identity), fig. 2 (position), fig. 4 (walking the tree), fig. 5 (the belief in play). Established once, never explained again. Recurrence does the work silently. |
| the caption bar | One component under every figure — same mono, same height, same em-dash rest state. Three instruments on one panel, not three illustrations in a row. |
| numbering | Figures are numbered like a paper because the page is a paper: an argument with evidence, not a portfolio with decoration. |
| cross-citation | Fig. 2 points forward to fig. 4; fig. 4’s leaves point back to fig. 2 and to the ledger; fig. 3’s rest caption points back to fig. 1. A cycle, not a stack. |
| one home per value | Nuance lives in 𝒰. Individuality lives in the curve. Uncertainty, kindness, taste and privacy are the tree’s leaves. Both/and is the frame. No value is ever listed as signage — that’s why there’s no wall of virtues anywhere. |
The negative space is also an argument. Nothing on the page autoplays except one breathing dot and two one-time draw-ons. Nothing records what you do — no analytics, no cookies, no stored input, no counter of your behavior. The page cannot learn about you. On a site whose subject is what feeds do to minds, that restraint is the load-bearing claim: the interaction design has to be able to survive its own critique.
Short answer: one figure is real mathematics, three are honest pictures a viewer will over-read, and one made a claim that was simply false until it got fixed. Here is which is which.
The strongest thing on the site, and the only part that would survive a statistician. Binomial paths really do converge to a normal distribution; a per-branch bias really does shift the population mean by n(p−½) × pitch; a single tracked path really is uninformative about the bias. Checked numerically rather than asserted: at p = .65 the simulated mean landed 53.2px from center against a theoretical 54.
the honest caveatThe mechanism is real; the personal axis is illustrative. “Share of cognition delegated to machines” has no units and was never measured. The theorem is true; the three dates on it are a story about a person.
What is real: every named intersection is a genuine field that exists at exactly that overlap, and every label sits on a region confirmed to have real area before it was named. What is not real: five overlapping circles cannot represent all thirty-one possible combinations — a hard geometric limit, not an editorial choice. So the absences are artifacts. Society ∩ technology has no region because circles can’t do it, yet a viewer reads that absence as a claim. Circle sizes are decorative too: they encode nothing while looking like they encode magnitude.
what would close itOne line under the figure: “five circles can’t show every overlap; the missing ones are geometry, not opinion.” Cheap, and it converts a hidden weakness into a visible epistemic standard.
The caption said “every edge runs both ways.” That was false, and I should have caught it earlier. Kierkegaard is dead. Clark and Chalmers do not know you exist. MUBI does not change because you watched something. Of eleven edges, exactly two are genuinely bidirectional — the feed (your attention is a real input to a recommender that then changes what it shows you and others) and the machines, weakly. The rest are one-way influence that feels reciprocal because reading someone changes how you read them.
fixed on the siteThe caption now reads: “most of these run one way. two of them run both — and that difference is the whole story.” The honest version is the sharper argument, because it isolates exactly what makes a feed different from a book.
Coherent as a manifesto, not true as a description. Nobody runs a decision tree; you act and then reconstruct the reason. The footnote (“deterministic here, messier in production”) is doing nearly all the honesty work on this figure. And the strongest word on it is the shakiest: “always.” Kindness always outranking rigor is a real ethical position, but it collides with your own principles — dissect the package, impact over intent, question everything — in every case where the kind move is the uninformative one.
worth decidingEither keep “always” and own it as a chosen commitment rather than a derived result, or change that edge to “first” — kindness gets the first word, not necessarily the last. One word, and the tree stops contradicting its neighbours.
The dynamics are honest — inputs bend rather than block, the target drifts and says so, the field never empties, there is no win state. But there is a real conceptual hole: the bricks are pre-labeled. Green is signal, violet is noise, and you know which is which before you hit them. In actual epistemics that is precisely what you don’t get. As built, the game is about obeying a known evidence hierarchy, not about discernment.
the fix that would make it trueGrey the bricks until impact and reveal signal or noise only after the ball hits — you learn what an input was worth after it has already bent you. That is the real experience, and it would make fig. 5 the second-most honest figure instead of the prettiest.
The defensible core is standard and true: different formalisms capture different structure, and no single model of a person is adequate. The flourish — that what persists across all five coordinate systems is the self — is rhetoric. The red dot recurs in every figure because it was drawn that way, not because reality forced it. The figures cohere because they were built to cohere.
why it still standsAuthored coherence is design, not fraud — as long as nobody claims it was found rather than made. The page never says the diagrams were derived from data, and the ledger bets against the design surviving twelve months. That is the right amount of doubt to leave in public.
Net: the site is honest about being a designed argument rather than a set of measurements, and its one genuinely quantitative claim is the one that was actually verified. Two fixes remain open — the geometry disclaimer and hidden brick types — plus one decision that is yours: whether kindness gets the first word or the last.