the second hundred · metaphor 191
Five witnesses. Each pair agrees perfectly wherever their accounts overlap. And yet there is no single story that contains them all. How can everyone be locally right and the whole be impossible?
It is the most disorienting kind of dispute: not one where someone is lying, but one where no one is. Check any two testimonies against each other and they match on the part they share. Check the next pair — they match too. Every local seam is clean. You keep expecting the global picture to snap into focus, and it never does, because the very thing that makes each pair consistent is what makes the loop impossible. The contradiction is not in any account; it lives in the way they are stitched.
Mathematics has a precise home for this: a sheaf, the machinery for asking whether locally consistent data can be glued into one global whole. Sometimes it provably cannot — and the failure is not sloppiness but a real, measurable obstruction, the same thing that makes an Escher staircase or a Penrose triangle look fine at every corner and impossible as a figure. Below, set five overlapping testimonies, each consistent with its neighbours, and watch the instrument try to build one global truth — and compute exactly why it can't.
Local data, global question
A sheaf is a way of organising data that lives locally — a value on this patch, a value on that one — together with rules for how the pieces must match wherever patches overlap. The central question a sheaf asks is the gluing question: given local pieces that agree on every overlap, is there a single global object that restricts to each of them? For many kinds of data the answer is always yes. But not always — and when it is no, the reason is exact.
Here each witness reports only a difference — how much higher the next witness's frame sits than their own. That is always locally satisfiable: any single claim can be met by sliding one frame. Consistency on each overlap is free. To glue is to assign every witness one absolute height that honours all the differences at once. Walk the loop adding up the claimed rises, and one number decides everything: the sum around the loop. If it is zero, the heights close and a global truth exists. If it is anything else, the last step lands short of the first by exactly that amount — an unbridgeable gap that no choice of local frames can remove. That gap is the obstruction, the sheaf's first cohomology.
What to try
Load Penrose loop: every witness says the next is one step higher. Each claim is completely reasonable — and the walk spirals up forever, never returning, the endless staircase made of five honest testimonies. The loop sum reads a stubborn non-zero, the gap on return shows the exact size of the impossibility, and the status flips to obstructed. Notice the overlaps-consistent chip: it stays a perfect 5/5 the whole time. Nothing local is ever wrong.
Now nudge the sliders so the rises cancel — a couple up, a couple down — until the loop sum hits 0. The instant it does, the staircase closes, the gap turns gold, and a single global height assignment exists: the testimonies glue. Try Reconciled and One dissenter to feel the knife-edge. You cannot fix an obstructed loop by editing one witness gently; you have to change the total. That is the signature of a genuine global contradiction: it is nowhere in particular, so it can't be patched anywhere in particular.
The mapping
The failed sheaf is a model for the testimonies, ledgers, and memories that are each locally sound and globally irreconcilable. Every witness's account squares with the neighbour they can be checked against; every department's books balance against the ones they trade with; every retelling of the family history is consistent with the person who told it to you. We assume that local agreement everywhere must add up to a global truth — and the sheaf shows, cleanly, that this assumption is false. Consistency propagates locally and can still fail to close around a loop.
This is a gentler and stranger diagnosis than "someone is wrong." It says the contradiction may belong to no single account — it is a property of the whole, a number attached to the loop, not to any witness on it. Which is why these disputes are so maddening to adjudicate: you interrogate each person, find them honest and coherent, and the impossibility survives every interrogation, because it was never in a person. It was in the gluing. Sometimes the truthful conclusion is not "who lied" but "these cannot all be frames of one world."
Read as life lessons
Everyone being right about their neighbour does not guarantee a shared world. Agreement is local; consistency around the whole loop is a separate, stronger condition that can simply fail.
The obstruction lives on the loop, not on any witness. You can clear every person of error and the impossibility remains — because it was never located in one of them.
Only the total around the loop matters. Softening one account won't help unless it changes the sum. Real reconciliation means altering the whole cycle, or admitting there is no single frame.
In the wild
Penrose triangles and Escher staircases are literally cohomology: locally drawable at every corner, globally impossible, and the obstruction is a number on the loop — Penrose's own theorem.
“Topological data analysis” and sheaf-based sensor fusion ask exactly this: do locally consistent readings glue into one global truth, or is there a measurable obstruction to a coherent picture?
Whether local solutions patch into a global one — a vector field on a sphere, a phase around a loop, a wavefunction's twist — is a sheaf-cohomology question with real physical teeth.
The honest model
Five patches sit in a loop; overlap i carries one number δᵢ, the claimed difference between neighbouring frames. A global section is an assignment of heights hᵢ with h_{i+1} − hᵢ = δᵢ on every overlap. The instrument tries to build it by literally walking: fix h₀ = 0, add each δ in turn, and check whether the final step returns to the start. It provably can iff the sum Σδᵢ = 0; otherwise the return misses by exactly Σδ, which the readout prints as the gap. Nothing is hard-coded — change any slider and the walk, the gap, and the verdict are recomputed.
That loop sum is not an analogy for cohomology; on this cover it is the first Čech cohomology class of the sheaf of frames-up-to-a-constant. It measures the failure of gluing, it is invisible to any single overlap (all of which stay consistent), and it is exactly the invariant Penrose used to prove impossible figures impossible. The instrument is deliberately the smallest honest instance — one loop, one number — but within that scope it computes the real thing, not a picture of it.
The mapping, exactly
| Mathematics | Life |
|---|---|
| a patch / local section | One witness's account, valid inside their own frame of reference. |
| overlap agreement | Two accounts matching on the part they can both be checked against. |
| a global section | A single story that contains every account at once — the shared truth. |
| the loop sum Σδ | The net contradiction accumulated all the way around — the thing that decides reconciliation. |
| the obstruction (H¹) | A real impossibility belonging to the whole, not to any one witness — why it can't be patched locally. |
| Σδ = 0 | The testimonies glue: everyone was right, and there is a world they are all right about. |
Where the metaphor tears
The instrument grants each account exact, quantified consistency on every overlap — a luxury human disputes almost never have. Usually someone is mistaken, or the overlaps themselves are fuzzy and contested. The sheaf models the rare, pure case: locally airtight, globally impossible. Its lesson is a caution against a specific over-confidence — that clean pairwise agreement guarantees a shared world — not a claim that every conflict is a cohomology class.
Knowing the loop sum is non-zero tells you a global truth is impossible under the current frames — it does not tell you what to do, or which account to weight, or whether to widen the cover until the loop breaks. Life mostly needs a decision, and “these provably cannot all be true together” is only where the human work begins. The maths can name the impossibility exactly and still leave every hard choice untouched.