The organism is not the model

This field has fifteen essays in it and about ninety figures. Not one of those figures measures an organism. Every one of them draws a geometry computed here — a crease pattern put through four theorems, a metric whose curvature was differenced, a route found by search — and the organisms appear in the prose as the reason the geometry was interesting.

That is the rule the field was built to, it was decided before the first essay was written, and the closing essay’s job is to state it as a table rather than as a preamble.

Why this field needed a rule and the others did not #

Six of this site’s nine fields work the same way: a claim is made, a generator computes it, and a wrong claim stops the build. Flat-foldability is decidable at a vertex, a packing ratio is arithmetic, a straight skeleton is a construction. In those fields the figure is the evidence and there is nothing else to be evidence of.

Two fields do not have that property and each needed its own rule.

History has no gate. A date is not derivable, no figure settles who folded what, and an essay about the past can be fluently wrong while passing every check. The rule adopted there was that a history figure draws something computed or is not drawn, with the evidence table itself given a checker.

Biology has the opposite problem, which is why the rule is the same and the reason is different. Biology is surrounded by evidence — photographs, measurements, a large and genuinely excellent literature — and none of it is held here. The danger is not that a figure will be unsupported; it is that a computed figure will be read as though it were one of those measurements.

The specific way it would go wrong #

The failure mode is worth describing concretely, because it is not obviously a failure.

Suppose an essay in this field had drawn a bar chart of packing ratios and labelled the bars beetle, earwig, leaf and solar array, using the same computed numbers that are there now. Every number would be exactly as computed as it is. Every assertion in the generator would still fire. The build would still fail if the arithmetic were wrong.

And the figure would be a lie, because the labels would assert a correspondence between a geometry and an animal that nothing had established.

That is a defect no gate on this site could catch. One check asks that content is inside its frame, another that labels are legible, another that patterns satisfy the theorems. None of them asks whether a label means what it says, and none of them could.

What the ledger contains #

Six rows, each naming a computation and the claim it does not support. The rows were chosen to cover the field’s whole range rather than its strongest cases, which is the only way a ledger of this kind is worth reading.

The growth solver computes the curvature a growth field forces, checked against four closed forms. It does not establish that any leaf grows that way. The pattern library computes that a corrugation with tapered columns folds flat, and refuses one tapered by row. It does not establish that a leaf’s creases are where these are.

The packing chart computes four geometries’ packed fractions. It does not establish the packed fraction of any wing. The census computes that one degree of freedom needs one driver. It does not establish how an insect deploys anything.

The generator refuses to draw a ledger of fewer than two rows, and refuses one whose rows all come from a single place — because a ledger drawn from one module is a statement about that module rather than about a field.

Why not simply measure something #

The obvious objection is that the rule is a workaround for a missing capability, and that a better version of this field would contain measurements.

That is half right and the half that is wrong is the interesting one. A site that measured leaves would be a different site with a different set of skills, and there is nothing dishonourable about the division of labour — the work here computes geometry and other people measure organisms.

What would be dishonourable is presenting the computation as though the measurement had happened, and that is what the rule prevents. It is the same discipline the history field arrived at from the other direction: a record is not a proof, and the field there is built to say what the record supports rather than to imply more.

The other half of the objection has a real answer too. A field that computed nothing and cited everything would be a review article, and this site’s whole proposition is that a figure should be a computation a reader could in principle re-run. Choosing to compute the geometry rather than to summarise the biology is what keeps the figures load-bearing, and the cost of that choice is exactly the gap this essay is about.

Where the organisms actually are #

The organisms are in the prose, and the prose follows three rules.

Named as motivation, not as measurement. A leaf packs by corrugating; that is why a tapered corrugation was worth computing. The computation is about the corrugation.

Attributed where a number comes from somebody. Where an essay states something about a real structure that was not computed here — that beetle hindwings fold under their elytra, that cortical folding is modelled as a growth instability — it is stated as what the literature says and not as what the figure shows.

Predictions offered as comparisons, not values. Twice this field makes something like a prediction: that larger leaves in similar buds should use more folds, and that maximum organ surface should scale with the square of size at fixed tissue thickness. Both are comparisons between cases, which is the only form of claim a model this coarse can support, and both are labelled as such.

The third rule is the one that took most discipline to keep, because a comparison is much less satisfying to write than a number. A model that says “about twenty-two folds” reads as a result; one that says “the larger of two similar leaves should use more” reads as a hedge. The second is the one the model actually supports, since every absolute number in this field depends on inputs that were chosen to make the geometry legible rather than measured from anything.

The test for whether a claim has drifted from the second form to the first is simple and worth applying to any modelling result: change an input that was assumed rather than measured, and see whether the claim survives. A comparison usually does. A number usually does not, and a number that does not is a property of the assumption.

The drift test, applied to this field’s own two predictions #

The test proposed above — change an input that was assumed rather than measured, and see whether the claim survives — is worth running on the two claims this field makes, because a test nobody applies is the same shape of thing the ledger is about.

The bud. The optimum fold count comes out as the square root of the leaf’s width over its thickness. The thickness was assumed. Change it and the number moves — halving it multiplies the count by 1.41 — but the comparison does not: a wider leaf in a similar bud still wants more folds, at every thickness whatever. The comparison survives; the number does not.

The container. The maximum surface an organ holds comes out as the cube of its size over its tissue thickness. The container was assumed to be a box and the corrugation to be single-scale. Change either and the leading constant moves — a differently shaped container, a hierarchical fold — while the exponent stays at three, because it comes from a volume rather than from a geometry. Again the comparison survives and the number does not.

Both claims were stated as comparisons, so both pass. That is the field having kept its own third rule rather than a coincidence, and running the test is what turns an assurance of care into something a reader can check.

What a measurement would break #

The closing paragraph promises that a version of this field with one measured input would need a different rule, and it is worth saying which, because the answer is already on the site.

Everything in the left column is enforced the same way: a claim is recomputed on every build and a wrong one stops it. That works because a computation can be run again. A measurement cannot be. A traced fold pattern or a measured growth field is a datum, and a wrong datum produces a green build — the arithmetic on it is as correct as ever.

So the gate model has nothing to offer a measured field, and the instrument that does is the one the history field already uses: not an assertion but a record of provenance — where the number came from, who took it, how strongly it attests what it is being used for, and what would overturn it.

That is the convergence worth ending on. Two fields arrived at the rule-without-a-gate problem from opposite directions, one with no evidence available and one with too much, and the answer in both cases is the same: where the left column cannot be recomputed, it has to be attributed. A field with measurements in it would not be this field with better data. It would be a history field about the present.

The check that is not a check #

There is one more thing worth being honest about, and it concerns the assertions themselves.

Every generator in this field carries assertions, and this site’s habit is that an assertion which has never rejected anything proves nothing. The biology generators’ assertions have rejected things: three of them refused their own first defaults, which is how the defaults were corrected.

But every one of those assertions is about the geometry. The generator that refuses to draw a bud figure where every fold count fits is checking that the container is binding; it is not checking that the container is a bud. There is no assertion anywhere in this field about correspondence with an organism, because there could not be one.

So the ledger’s left column is machine-checked and its right column is a promise. The promise is the thing being made explicit here, and the reason it is a whole essay rather than a footnote is that an unstated promise is one nobody can hold the site to.

That asymmetry between what a gate can see and what a claim needs is not peculiar to biology, and it is worth generalising because it recurs. Every automated check on this site verifies a relationship between a computation and itself: that a number matches a closed form, that a pattern satisfies a theorem, that a label fits inside a box. None of them verifies a relationship between the computation and the world, because verifying that would require the world itself to be one of the inputs.

The consequence is that the honesty of a figure-first site is bounded by the honesty of its captions, and captions are prose. The gates raise the floor enormously — a wrong computation cannot ship — and they do not touch the ceiling. Reading this site well means knowing which of those two a given sentence depends on.

This field is where the distinction bites hardest, which is why the rule lives here. But the same reading applies to the complexity field, where a computed hardness result and a claim about what is practically achievable are separated only by the text, and to every essay that names an application.

What is genuinely established #

It is worth ending on the positive side of the ledger, because the field is not a disclaimer.

That a growing sheet with a non-uniform growth field cannot lie flat is established, with the curvature computed and checked against four closed forms it does not use. That the sign is decided by where the growth is, not how much of it there is, is established by a family that runs through zero. That a fixed excess of arc length leaves the wave count free is established by solving each amplitude exactly.

That a corrugation may taper along its folds and not across them is established, derived from Kawasaki before it was drawn, with the refusal asserted in the gate. That a shape whose lattice colours are unbalanced admits no single route is established, by a count that never contradicts an exhaustive search across a sweep.

Every one of those is a statement about geometry, every one is checkable, and none of them needed an organism. That is the field working as intended rather than the field apologising.

Three of the results also produced findings that were not sought, which is usually the sign that a computation was doing real work rather than confirming a picture. The taper rule came out of deriving Kawasaki for the corrugation family and finding that the row heights are bound and the column widths free — the opposite of what the obvious drawing assumes. The parity obstruction turned out to have a witness that is balanced and still unroutable, which stops the cheap test from looking sufficient. And the search that found routes had a bug that only showed on a shape whose answer was obvious by inspection.

None of those three was in the plan. Each is the kind of thing that appears when a claim is made executable, and each would have been invisible to an essay that described the same subject without computing it.

The idealisations, collected #

The field runs on four assumptions and each has appeared in an essay of its own.

Zero thickness, which makes packing unbounded and surface unlimited, and whose removal is what gives the bud and the container their interior optima.

Creases as lines, which is worse in a body than anywhere because a compliant hinge has a width by design rather than by imperfection.

Isotropic prescribed growth, which real tissue does not do — it grows anisotropically, along directions it lays down itself, on a schedule responding to the shape already made.

Rigid panels, which no organism has. A leaf bends everywhere, and a substantial part of a real unfolding is panels flattening rather than fold angles opening.

The site has an essay listing four things that are not true of paper; this is that essay’s biological counterpart, and the departures are larger here.

Where this leaves the field #

Nine fields now exist, which was the checkpoint this part of the collection was built toward. What biology adds to the other eight is a direction of reading rather than a new set of theorems: folding preserves a metric and gives up flatness, growth changes a metric and has flatness taken away, and the two are one geometry approached from opposite ends.

The strongest single result of the field is probably that a crease carries no curvature at all — that developability, the least discussed of the four conditions this site checks, is exactly the statement that folding an uncut sheet creates none. That was available from the first phase and nobody had said it, and it took a field about growing things to make the question worth asking.

The weakest is the convergence argument, which is labelled a conjecture in its own text and needs a count nobody has done.

Both of those are better outcomes than a field that had produced neither, and the difference between them is legible only because the ledger exists.

What the field does not have, and what a later round of work would have to build, is any bridge to a measurement. Every essay here ends at the same place: the geometry says this much, and settling the rest needs data nobody here holds. A version of this field with even one measured growth field or one traced fold pattern would be able to close a loop that fifteen essays leave open — and would need a different rule, because the moment a measurement enters, the question of what it does and does not license becomes a live one rather than a hypothetical.

Until then the honest position is the one drawn at the top of this page: a left column that a build enforces, a right column that nothing does, and a promise that the two are never allowed to swap.