the second hundred · metaphor 193

Loyalty in tiers,
not degrees.

Think about how your loyalties actually sort themselves. There is an inner few you'd drop everything for; a wider circle you'd go far for; a crowd you wish well; and then the rest of the world. What's striking is how discrete it feels — you are close to someone up to a tier, and the tier is what decides things, not the fine gradations inside it.

And the tiers nest. Your sister and your spouse both sit inside "family"; your family and your oldest friend both sit inside "the people I'd bleed for"; that whole clan sits inside "my world," which finally contains the stranger on the train. Ask how far apart any two people are and the honest answer is rarely a smooth number — it's the name of the smallest circle that holds them both.

This is a strange kind of nearness, and it has a strange consequence you can feel if you look for it: of any three people in your life, two are equally distant from the third. Closeness here is nested, not graded — and mathematics has a precise name for spaces that work exactly this way.

the hierarchy · height = shared belonging the selected triple · isosceles vs. flat
the tree · branches & leaves the selected triple equal sides · isosceles where flat distance tears
d(· , ·)
d(· , ·)
d(· , ·)
isosceles?
strong inequality
the honest proof · every C(6,3) triple
— / —
isosceles — two longest distances equal
max violation of  d(x,z) ≤ max( d(x,y), d(y,z) )
the flat contrast · same triple, on a line
Euclidean gaps
line positions are hand-set for illustration; every tree distance above is computed live as an ancestor's height.
Reshape the hierarchy
Tier spacing · rescales every merge height1.00×
compressed · tiers bunchevenspread · tiers steep
Click any three leaves to pick a triple. Drag the slider to rescale the tiers — distances change, but the tree property never breaks.
One nested clan — family inside "the dear" inside "my world" — and a stranger out at the edge. The chosen triple's two long sides snap equal: that is the ultrametric.

Distance as an ancestor's height

How far is the nearest branch that holds you both.

On a family tree, two leaves are joined by their nearest common ancestor — the deepest branch-point holding them both. An ultrametric turns that into a number: the distance between two leaves is just the height of that ancestor. Cousins meet at a shallow branch, so they are close; you and a stranger meet only at the root, so you are as far apart as the tree allows. Closeness is decided entirely by which circle first encloses you both.

That single rule has a sharp fingerprint. The ordinary triangle inequality says a detour can't be a shortcut: d(x,z) ≤ d(x,y) + d(y,z). An ultrametric obeys something far stronger — the strong triangle inequality, d(x,z) ≤ max( d(x,y), d(y,z) ). The longest leg can never exceed the larger of the other two; it can only tie it.

Force that on three points and geometry has no choice: every triangle comes out isosceles, its two longest sides exactly equal. Pick any three leaves — two share a branch deeper than the third, so both reach that third one at the very same ancestor, and two of the distances collapse to a single height. No "generic" triangle is possible here. Nearness is quantised into nested tiers, and the tiers are all there is.

What to notice while you play

Every triangle you pick comes out balanced.

Select three leaves and watch the panel draw their triangle to scale. However you choose them, two sides land equal — the shape is stubbornly isosceles, with the sage sides marking the pair that meet only up at a shared ancestor. Beside it, the same three people laid on a flat line make a scalene triangle: three different gaps, the longest equal to the sum of the other two. Quantised belonging versus smooth proximity — that contrast is what the metaphor is about.

Now drag tier spacing: you are not editing one bond but rescaling every merge height at once, so whole clades rise and fall together — you can't quietly promote one friend without moving the whole rank they sit in. Through all of it the tally holds: every triple stays isosceles, the violation stays flat zero. That is not luck; it is what "tree" means.

The mapping back

Kinship in strict tiers.

Read the tree as a map of allegiance and it says something true about how people rank each other. The distance between two of your people is not a private, finely-tuned figure — it is the name of the smallest club they both belong to: same household, same family, same side, same species. Everything finer washes out. That is why loyalties feel categorical rather than continuous: you don't owe a stranger 40% of what you owe a sibling; you owe them what you owe the tier they fall in.

The strong inequality is its moral spine: your bond to a distant acquaintance can't exceed the wider of two shared-circle bonds — closeness doesn't accumulate across tiers the way miles accumulate across a map. And the isosceles law is a social fact you already know: bring any three people together and no one is uniquely in the middle. Two of them are, from the third's vantage, simply "the far ones," equally outside.

Read as life lessons

Nested, quantised, moved in blocks.

01

The tier is the whole story

Closeness is set by the smallest circle that holds two people, and nothing finer counts. Stop scoring bonds to two decimals; ask which ring someone is inside, because that is the number life actually uses.

02

Two of any three are tied

In a nested world every triangle is isosceles: from any vantage, two others are equally far. Nobody sits uniquely between two people — belonging sorts them into "us" and "the rest," and the rest are level.

03

Whole clans move at once

You can't shift one leaf without its branch. Rescale a tier and an entire sub-tree travels together — promote a faction, and everyone under it rises. Loyalty comes in blocks, not sliders.

In the wild

Where distance really is a height.

LIFE'S TREE

Taxonomy nests species in genus in family; phylogenies read the distance between two organisms as the age of their last common ancestor. Under a steady molecular clock that distance is exactly ultrametric.

FILES & NAMES

Folders within folders, org charts, DNS names, postal codes: two items' separation is their nearest shared container. Hierarchical clustering builds such trees on purpose, so the dendrogram height is the dissimilarity.

PHYSICS & NUMBER

The low-energy states of a spin glass organise into nested valleys-within-valleys — Parisi's ultrametric order. And the p-adic numbers are the pure case, where "size" is set by tiers of divisibility, not magnitude.

The mapping, exactly

Mathematics ↔ life.

MathematicsLife
a leafA person, or a loyalty — one of the things you can be close to or far from.
height of the nearest common ancestorHow deep the shared belonging goes — the intimacy of the smallest circle that already holds you both.
distance = that height, full stopCloseness is decided by which circle first encloses you both; nothing finer inside it changes the answer.
strong inequality  d(x,z) ≤ max(d(x,y),d(y,z))Your distance from a stranger can't beat the wider of two shared-circle bonds — closeness doesn't add up across tiers.
every triangle isoscelesOf any three people, two are equally distant from the third; no one is uniquely in the middle.
a sub-tree (a clade)An inner circle that moves as one — promote or demote the tier and the whole faction travels together.
heights increase leaf → rootEvery wider circle is at least as loose as the ones nested inside it; belonging only ever weakens outward.

The honest model

What's really under the hood.

The tree is stored as nested merges, each internal node carrying a height, normalised so the root sits at 1 and leaves at 0. To find d(x,y) the panel walks the root-to-leaf paths of x and y, takes the last node they share — their nearest common ancestor — and returns its height. Nothing is looked up; every distance chip is that walk, run live. The tier-spacing slider applies one monotone map, h ↦ h^γ, to every height at once. Because a monotone map preserves the order of the heights, it preserves the tree structure — so distances slide around while the ultrametric survives intact.

The proof strip is the honest part. Each redraw loops over all C(6,3) = 20 triples, computes the three ancestor-heights, sorts them, and checks that the two largest are equal, printing the count and the largest gap max(0, s₂ − s₁). On any genuine tree that gap is exactly 0 — the two outer pairs of every triple meet at the same ancestor, so their distances are literally one number. Beside it, the "flat" card places the same people at hand-set points on a line and reports |differences|, where the longest equals the sum of the other two and the strong inequality fails. The guarantee is not asserted in prose — it is recomputed every time you touch the tree.

Where the metaphor tears

Real loyalty is not a clean tree.

Cross-cutting ties break the branches.

A friend who is closer than family, a colleague you'd trust over a cousin, a chosen kinship that outranks blood — these are edges the tree cannot draw. Each one is a direct violation of the strong inequality: a short link between two leaves that the hierarchy insists are far. Real bonds form a tangled graph with shortcuts; the moment one appears, the isosceles guarantee is gone.

Closeness is graded, and it drifts.

The model quantises nearness into a handful of discrete heights, but human closeness is continuous and contextual — you are nearer your sister on the hard days than the easy ones, warmer to a mentor this year than last. A single frozen number for "how close" flattens a living, sliding thing. The tree gives tiers; life mostly gives a dimmer switch.

We belong to many trees at once.

A tree demands one nesting, but you are inside several overlapping circles simultaneously — family, faith, profession, hometown, cause — and no single hierarchy can hold them all. Forcing loyalties into one clean clade erases the genuine multiplicity, and it freezes what betrayal, estrangement, and love keep rewiring. The strict tree is a beautiful simplification of something irreducibly overlapping.