the second hundred · metaphor 151
A biography. A reference letter. A one-line reputation. A rough mental model of a friend. Every summary of a person throws almost everything away — and some are true anyway. What separates a faithful portrait from a flattering lie is not how much it keeps, but whether the little it keeps still stands in the right relations. When is a lossy summary still true to the thing it leaves behind?
Nobody fits in a sentence, and the honest sentence doesn't pretend to. It keeps faith a different way: whatever it does say about how the parts of a life fit together, it gets right. Combine two things in the world and then summarise; or summarise each and then combine — a faithful summary gives the same answer both ways. The particulars vanish; the relations survive. That double demand — lossy about the details, exact about the structure — is the whole game.
Mathematics has a spare, unforgiving word for a map that keeps every relation while dropping every particular it can: a homomorphism. Below is one you can build by hand, then audit for lies.
The auditing bench · a small life, and a summary of it
source = Z₆ · six elements {0..5}, combined by add-then-wrap-at-6 · the grid below is its every relation
The one law
Start with something that has genuine internal structure — not just a heap of parts, but a rule for how any two of them combine into a third. The grid in the instrument is exactly that rule for a tiny world of six elements: read off row a and column b and you get a·b. A summary is a map f that sends each element to a coarser token — fewer distinctions, less to carry.
The summary is faithful when one equation holds for every pair, no exceptions: f(a·b) = f(a)·f(b). In words — the summary of a combination equals the combination of the summaries. That is the definition of a homomorphism, and the quantifier is the whole point: thirty-six pairs here, and a single red cell disqualifies the map. Faithfulness is not an average; it is a promise kept in all cases or broken.
When the promise holds, look at what it cost. Many distinct originals now land on the same token — those collapsed sets are the fibers, the particulars the summary can no longer distinguish. The elements sent to the blank token, the do-nothing, form the kernel: precisely the differences the summary has declared irrelevant. A homomorphism is honest bookkeeping of a loss — it names exactly what it stopped being able to see.
What to try
Choose how many distinctions the summary is allowed to keep — two (a bare parity: odd or even), three, or the full six. Then edit the map by clicking any element to change what it becomes. The grid recolours instantly: green where the summary keeps faith with that relation, red where it betrays it. The count of betrayals is live, and only at zero does the bench declare a homomorphism.
Try the flattering lie preset — it takes the faithful thirds-summary and swaps just two labels. One honest-looking edit, and the grid bleeds red: many relations, quietly wrong. Then try sending element 0 — the do-nothing — to anything but the blank token, and watch faithfulness collapse at once. A summary that mislabels "nothing happened" can't respect a single combination it takes part in. When the map does pass, the panel names the fibers and the kernel: this is what it kept, and this, exactly, is what it agreed to forget.
The mapping
The same test grades every lossy account we trust. A scientific model is a map from a vast reality to a small state space; it earns its keep not by realism but by preserving the relations you will act on — push the model and push the world, and the two should agree. A good biography compresses a life to a few hundred pages and stays true by getting the connective tissue right: who shaped whom, what followed what, which choice made which choice possible. A reputation is a brutal one-token summary — yet it is faithful when it tracks how the person actually behaves under combination, across situations, and a lie precisely when it doesn't.
This is why "it left things out" is not itself an accusation. Every faithful summary leaves nearly everything out; that is what makes it a summary. The charge that bites is different: it got a relation wrong. The instrument separates the two failures cleanly — a fiber is an honest loss, a red cell is a lie — and most of our quarrels about summaries are really quarrels about which one we're looking at.
Read as life lessons
A summary can drop almost everything and still be true. Parity keeps only odd-or-even and forgets the number itself — yet it never lies about sums. Aim for the relations that survive compression, not for keeping more.
Fidelity isn't a batting average. One broken relation and the whole map is not a homomorphism. A portrait that is right about most of a life and wrong about the hinge of it has failed at the only pair that mattered.
What you throw away, throw away evenly. In a homomorphism every token hides the same-sized fiber — the kernel repeated. Lose detail by a rule, not by whim, and the structure comes through clean.
In the wild
A map of a city keeps distances and adjacencies while dropping every brick; a climate or economic model keeps the relations you'll steer by. Useful exactly insofar as the operations survive the shrinking.
A translation loses the music and keeps the relations — who did what to whom, which clause depends on which. It reads as faithful when the structure of meaning transfers, and as betrayal when it doesn't.
Hashes, parity bits, and modular checksums are homomorphisms on purpose: they respect the operation so an error in the data shows up as an error in the tiny summary. Compression you can still compute with.
The mapping, exactly
| Mathematics | Life |
|---|---|
| the source group | The whole thing in its full particularity — a life, an object, a system, with all its internal relations intact. |
| the operation a·b | How any two parts combine or relate — the structure that a faithful account is on the hook to preserve. |
| the map f | The summary itself: the biography, the model, the reputation, the one-line reduction. |
| the target group | The summary's own smaller vocabulary — the limited set of distinctions it is allowed to keep. |
| f(a·b)=f(a)·f(b) | Staying true to the relations: the summary of a combination is the combination of the summaries, in every case. |
| a fiber | The particulars discarded — distinct originals the summary honestly can no longer tell apart. |
| the kernel | The differences the summary treats as nothing — everything it has decided doesn't matter. |
| an isomorphism | A lossless likeness that keeps every distinction and still respects structure — faithful and complete at once. Rare. |
| a red cell | The place the summary lies — a single relation it gets wrong, and the reason to distrust the rest. |
The honest model
The source is the cyclic group Z₆: the elements 0..5, combined by a·b = (a+b) mod 6. The summary is an array f of six images, each a token of the target group Zₘ whose own combination is (u+v) mod m. Nothing about the verdict is scripted. On every edit the bench walks all 6 × 6 = 36 pairs and, for each, compares f((a+b) mod 6) against (f(a)+f(b)) mod m — that comparison, cell by cell, is the colour you see and the count of betrayals in the readout.
The fibers and kernel are read the same live way: a token's fiber is the list of source elements the array sends there, and the kernel is the fiber of 0. When betrayals hit zero, a theorem you can watch happen forces the fibers to share one size — they are the cosets of the kernel — so "detail kept" and "each token hides k originals" fall straight out of the same arrays. No formula is printed as fact; every number on the bench is read off the map you built.
Where the metaphor tears
A homomorphism can respect every relation you tested and still omit the one thing that mattered. Parity is perfectly faithful and tells you nothing about magnitude. Faithfulness is always relative to the structure you chose to preserve; pick a thin enough structure and a nearly empty summary passes the audit. The green grid certifies consistency, never sufficiency.
The whole apparatus needs a genuine, closed law of combination — a·b defined for every pair, unambiguously. A friendship, a grief, a career rarely come with one; their "relations" are plural, fuzzy, and contested, not a single Cayley table. Where there is no exact operation there can be no exact homomorphism — only the aspiration to keep faith with structure we can't fully write down.
The algebraic law is binary: one broken pair and the map is simply not a homomorphism. Human summaries are graded — mostly faithful, faithful enough, faithful where it counts. The bench's hard red-or-green is deliberately unforgiving; real fidelity is a matter of how many relations survive and how much each one weighs, not a single Boolean verdict.