The textbook line — variation arises at random, selection keeps what helps — reads like Darwin's opening sentence. It is not. It was assembled between 1918 and 1950 by a group of people who had little time for each other and were forced onto the same sheet of paper by mathematics.
One whole block of the problem was left outside during that assembly: how a gene actually grows into an eye. Not because anyone thought it unimportant, but because you can calculate very accurately without opening that box. The argument still running eighty years later is about that box — and neither side denies there is something inside it.
Darwin had a hole and he knew it. On the Origin of Species needs three things to work: individuals differ, those differences pass to offspring, and the differences change how many offspring get left behind. For the first and third he had observations by the cartload. For the second he had no mechanism — in 1859 nobody knew how inheritance worked. In 1867 an engineer named Fleeming Jenkin pushed a knife into that hole: if offspring are the average of their parents (which is what most people assumed), half the variation in a population is wiped out every generation, and after a few generations everyone is alike and selection has nothing left to pick from.
The answer had been on the table since 1900, when Mendel was rediscovered: the units of inheritance are particulate, they do not blend, so differences never get averaged away. What it actually produced was a quarrel that ran for nearly two decades. On one side, the pea-counters: traits come in discrete classes, with nothing in between wrinkled and round. On the other, the height-measurers, who went into the field, measured real populations, and found nothing but smooth continuous curves. Both sets of data were real. They were measuring the same machine — one had got down to the parts, the other was reading the dial on the whole thing.
In 1918 Fisher did the arithmetic that settled it: if a continuous trait is governed not by one particle but by many particles each contributing a small share, then adding up the small shares gives exactly a smooth bell-shaped spread. Particulate inheritance and continuous traits are not in conflict; the second is what the first inevitably looks like once there are enough particles. That was a mathematical demonstration, not an experiment; empirical confirmation came many years later.
Over the next twenty years Fisher, Haldane and Wright turned "how does the proportion of a variant in a population change under selection, mutation, sheer luck and migration" into something you could derive. Dobzhansky took it out to wild fruit flies in 1937, and Huxley gave the whole package its name in 1942: the modern synthesis. What got synthesised amounts to a few sentences. Evolution is the changing proportions of variants in a population. The direction in which variation appears has nothing to do with what is needed. The only directional force that builds adaptation is selection. And, most importantly, whatever a body goes through in its lifetime does not change what it passes on. (The textbook version came out narrower than what these people themselves held: Gould argued it had "hardened" by the 1960s into almost nothing but selection — a historical claim that is itself disputed.)
You can project a city's population quite accurately: which districts are growing, what the distribution looks like in twenty years. Doing it requires knowing nothing about how any single house was built — and it is precisely by not touching the house-building layer that the population layer becomes calculable. The modern synthesis does the same thing: it leaves the box marked "how a gene grows into an eye" shut, because calculating proportions does not require opening it, and the results hold up. That is where its power comes from.
The cost lands elsewhere. There is a class of question that the equations do not get wrong so much as have no slot for: why do some variants keep appearing and others never appear at all? The default answer is that mutation is noise sprinkled evenly over everything. Developmental biology has shown that this is false — from flies to humans the same very old set of switch genes is at work, and it determines which forms are not within reach at all, and which forms cannot be changed without a whole cluster changing along with them. Variation is not an even rain. It has a shape.
So a group of researchers has proposed that four things should move from "input" to "cause": developmental bias (what a body can grow into is skewed from the start), phenotypic plasticity (one set of genes building different bodies in different environments), niche construction (a beaver turns a stream into a pond, and what its descendants inherit is the altered environment), and inheritance outside the genes (methylation marks, gut microbes, behaviour picked up from others). The proposal is called the extended evolutionary synthesis.
In 2014 Nature ran the two positions a page apart. Laland and seven co-authors: yes, urgently. Wray, Hoekstra, Lenski and four others: no, all is well. The fight is not over whether those four things exist — the opposing side explicitly grants that they are real. It is over whether you enter them as the material selection acts on, or concede that they carry direction of their own. The objection is hard to dodge: call it a cause, and what extra prediction do you get? So is the reply: it is precisely because they have always been booked as inputs that "where does direction come from" has gone eighty years without being properly asked.
There is no verdict, and part of the dispute is not empirical at all — no further experiment settles what counts as a cause. One thing worth stating plainly: both sides are Darwinians, and nobody is claiming that what a body goes through gets written back into its offspring (the issue on epigenetics and the limits of inheritance covers where that line sits).
Take one packet of seeds, sow half in an exposed spot and half against a wall. The exposed ones come up short, thick-stemmed, deep-rooted. The genes are identical — one set of genes building two bodies in two environments. That capacity is called phenotypic plasticity, and nearly every organism has some of it.
The textbook order in which a new form appears is: a mutation happens to turn up, one individual therefore looks different, and if looking that way leaves it more offspring, the proportion climbs. Plasticity-first reverses the first half — the environment shifts, and a whole cohort produces the same new form at once, because the capacity to respond that way was already in them.
That ordering is not a word game. On the mutation route the new form sits in a single individual, where sheer luck can erase it. On the plasticity route it arrives as a crowd, and selection has usable material immediately.
One case has been worked out fairly cleanly. In California there is a black lava flow that formed roughly 22,500 years ago, and the side-blotched lizards living on it are far darker than their relatives on the surrounding sand. Researchers sequenced the genomes of more than a hundred animals: lizards off the lava can already adjust how dark they are to match their background, and that adjusting capacity is older than the lava flow itself — so the order was that they could live there first. Then, within the lava population, two genes involved in melanin carry variants found only in that population, and hatchlings carrying them come out darker. Spadefoot toads show the same route: across several closely related species, the big-headed "carnivore" form of the tadpole shifts from appearing only when triggered by a diet of shrimp to appearing often without any trigger at all.
The opposite effect is just as real. Around 1949 Bogert noticed that a lizard which simply moves into the shade keeps its body temperature where it wants it — so the genotypes that would have paid a price for poor heat tolerance never pay it. Selection sees nothing, and genetic change stalls there. This is called behavioural inertia. Plasticity can speed things up and it can put a brake on them, depending on how fast and how steady the environmental change is.
Establishing the order is hard, because you are reconstructing a history that has already finished, and cases meeting all the strict criteria are still few. What stands is this: plasticity-first does happen; whether it is common enough to require rewriting the framework has no answer yet. That is exactly the part of the fight in "a framework is often strongest exactly where it makes certain questions unaskable" that evidence can still push forward.