Most discoveries are a new arrangement of old facts, not new ones

The quality of an understanding is set by how its known pieces are connected, not by how many you have collected. Add a thousand more observations and you may understand no more than before. Connect the observations you already hold in a way that makes a correct prediction, and you have understood something new.

I want to make that claim concrete and falsifiable, not inspirational. So this piece leans on two cases from the history of science where the raw material was sitting in plain view and the advance was in the ordering. Both cases let me lay numbers and predictions side by side and check whether the arrangement, by itself, paid out.

Is a discovery new information or a new arrangement?

Front-loaded answer: usually a new arrangement.

Take Mendeleev. By 1869 chemists had isolated dozens of elements and measured their atomic weights and reactions. The raw material was sitting in every well-equipped lab in Europe. What nobody had was the right ordering. Mendeleev lined the elements up by atomic weight, noticed that chemical behavior repeated in a periodic rhythm, and grouped them so that elements with similar properties fell into the same column.

That was a reorganization, not a new measurement. And it did something a mere catalog cannot do. Where the rhythm demanded an element that had not been found, Mendeleev left a gap and described the missing tenant in advance.

In his 1871 treatment he predicted an element below aluminium, which he called eka-aluminium, with atomic weight around 68 and density around 6.0 grams per cubic centimeter. In 1875 Paul Emile Lecoq de Boisbaudran isolated gallium: atomic weight near 69.7, density about 5.9. He predicted an element below silicon, eka-silicon, with weight near 72 and density near 5.5. In 1886 Clemens Winkler isolated germanium, with weight about 72.6 and density about 5.3. Those three predictions, the historian of chemistry Philip Stewart notes, are what earned Mendeleev his enormous prestige, though it is worth saying that other predictions he made were less successful, partly because he held too rigidly to the structure of his own table. The structure was powerful, not omniscient.

Notice what changed between an ordinary list of elements and Mendeleev's table. The atoms did not change. The measured weights did not change. The connections among them changed, and that single change converted a record of the past into a forecast of the future.

What separates a real reorganization from a relabel?

This is the part that keeps the thesis honest, because not every rearrangement is a discovery. You can shuffle facts forever and produce nothing but a tidier filing cabinet.

So here is an operational test, the kind you can apply yourself. A reorganization earns the word discovery when it commits to a checkable consequence that was not built into the inputs, and that consequence holds. The sharpest form is a specific prediction about something unmeasured that later comes true, as with Mendeleev. A slower form counts too: an arrangement that goes on to account for large bodies of independent evidence it was never fitted to, what philosophers call consilience, is committing to consequences in the same way. What does not count is a pure relabel: a renaming that regroups what you already knew without committing to anything new about the unobserved.

Mendeleev's table passes cleanly. The weight and density of gallium were not in his data. He had no sample. The arrangement itself forced a numerical claim about an element no one had seen, and the claim held to within a few percent when the element finally appeared. Germanium matched the predicted properties so closely that chemists treated it as the strongest confirmation of the scheme up to that point. The win was fit-to-prediction, not a contrast with neighbours.

A relabel fails the same test. Suppose I sort the elements by the first letter of their names. That is also an arrangement. It is even useful for finding things in an index. But it predicts nothing about an unknown element, because the order carries no physics. The difference between Mendeleev and the alphabet is not effort or neatness. It is whether the structure pays out in correct predictions about cases it never saw.

That gives a falsifiable line for the whole essay. If reorganization were just bookkeeping, no rearrangement of known facts could ever forecast an unmeasured quantity. Mendeleev forecast two. The bookkeeping view is wrong.

Did Darwin discover new facts, or new connections?

Darwin sharpens the point, because he discovered almost no facts at all.

By 1859 naturalists already knew that island species resembled species on the nearest mainland, that a human arm, a bat's wing, and a porpoise's fin share the same bone plan, that the embryos of fish, reptiles, birds, and mammals look strikingly alike early on, and that fossils in deeper rock layers differ systematically from living forms. Each of these was a familiar puzzle in its own field. Geographers had the distributions. Anatomists had the homologies. Embryologists had the embryos. Paleontologists had the fossils.

Darwin's move was to connect these separate puzzles into one explanation: descent with modification from common ancestors. None of the observations were his. The link among them was. In the closing chapter of the first edition of On the Origin of Species, Chapter XIV, he wrote, "As this whole volume is one long argument," and the argument is exactly an act of organization. He showed that a pile of unrelated facts becomes a single coherent pattern the moment you wire them to common descent.

And, like Mendeleev, the new wiring predicted. If living forms descend with modification from earlier ones, the fossil record should contain intermediate forms, and biology has repeatedly found such forms since. The cleanest example is one that was predicted before it was dug up. Reasoning that a fish-to-tetrapod intermediate should sit in rock of a particular age, Neil Shubin and his colleagues went looking in Late Devonian deposits and in 2004 found Tiktaalik, an animal with fish gills and scales but also a mobile neck and the beginnings of limb-like fins, in rock roughly 375 million years old. The fossil-record expert Jennifer Clack summarized the moment: "It's one of those things you can point to and say, 'I told you this would exist,' and there it is." The facts that fed Darwin's tree were old. The organization was new, and the organization is what generated a testable claim about what the rocks should hold.

Why does arrangement beat accumulation?

There is a respectable physical reason that "more" and "better organized" are different things, and it comes from physics rather than from any pep talk about creativity.

In 1972 the physicist Philip W. Anderson published a short, now-famous essay in Science titled "More Is Different" (Science, volume 177, pages 393 to 396, 4 August 1972). His target was what he called the constructionist hypothesis, the assumption that once you know the fundamental laws you can simply scale up and rebuild everything above them. Anderson argued that you cannot: the ability to reduce everything to a few fundamental laws does not imply the ability to start from those laws and reconstruct the universe. "At each level of complexity entirely new properties appear," he wrote, and knowing the rules of the parts does not hand you the behavior of the whole.

Anderson was making a claim about nature: that new structure and new behavior emerge when parts are arranged into wholes, and the arrangement carries information the parts alone do not. I want to be careful about what is his and what is mine. The idea that reorganization yields genuinely new properties is Anderson's. My narrower interest here is a practical follow-up question he did not set out to answer: which reorganizations of known facts produce a jump in predictive power, and which are only permutations that look new but forecast nothing. The Mendeleev test above is one attempt at drawing that line.

It is also where I should place the work nearest to mine, because a claim about structure carrying information has live neighbours, and not naming them would be a hole. The closest is Assembly Theory, developed by Lee Cronin and Sara Walker, which proposes a measure of complexity, the assembly index, defined as the minimum number of steps needed to build an object from its parts. The ambition there is large: that this number is an intrinsic, one-directional measure of how much selection an object embodies. It is also contested. Hector Zenil and co-authors argue in PLOS that the index largely tracks ordinary compression and does not, on its own, explain selection. I am not adjudicating that fight, and I am not claiming Assembly Theory's mantle. My test is a smaller and more modest tool: it does not score how complex a thing is, it only asks whether a particular rearrangement of known facts predicted an unseen case and was right. A related physical idea, Jeremy England's dissipative adaptation, proposes that structures which absorb and dissipate energy well are statistically favored to form under a driving force. Worth stating plainly, because the popular version overshot: favored is not inevitable, and the same driving that can build order can also tear it down. I borrow the spirit of these ideas, that arrangement is where the information lives, without borrowing claims I have not earned.

I should also flag the limit of my own claim, because honesty belongs on the evidence, not on the thesis. I am not asserting a universal law that every advance in every field reduces to pure rearrangement with zero new observation. Real science mixes new measurement and new organization, and sometimes you genuinely need a fact nobody had, a new telescope, a new instrument, a new sample. The claim I will defend is narrower and, I think, more useful: when you already hold the relevant facts, the binding constraint on understanding is usually their organization, and the two cases here show that reorganization alone can produce correct predictions about the unknown. That is a statement you can check against the historical record and try to break.

What this changes about how to look for ideas

If the thesis holds, it redirects where you spend effort.

The instinct, when you are stuck, is to gather more. Read another paper, run another measurement, collect another dataset. Sometimes that is exactly right, when a genuinely missing fact is the blocker. But the Mendeleev and Darwin cases suggest a different first question: do I actually lack a fact, or do I lack the right connection among the facts I already have? The periodic table and descent with modification were both built from inventory that was already on the shelf and badly arranged.

There is a checkable signature for the good kind of rearrangement, and it is worth repeating because it is the whole essay in one line. A reorganization is doing real work when it commits to a prediction about a case it has never seen, and the case confirms it. Mendeleev's gaps named the weights of elements no one had touched. Darwin's tree implied fossils no one had dug. Both put a number or a form on the unobserved, and both were right. A rearrangement that cannot stick its neck out that way, that only renames and regroups the known, is housekeeping. Useful, sometimes necessary, but not the thing that moves understanding forward.

Let me put my own neck out, since the essay demands it. A prediction, dated and falsifiable, written 9 June 2026: across well-documented breakthroughs that were later reconstructed in detail, the ones that yielded a correct prediction about an unobserved case will turn out, on inspection, to have introduced no decisive new measurement at the moment of the breakthrough; the novelty will sit in the arrangement of already-public facts. This is checkable. Take a set of agreed-upon discoveries, fix the date each one is credited to, and ask whether the inputs available just before that date were already sufficient. My bet is that a large share were arrangement-limited, not data-limited. If a fair audit finds that most landmark predictions in fact waited on a fresh observation that no one previously had, then the binding constraint is data, not organization, and I am wrong. I would want to see that audit.

So when an idea feels like a discovery, run it through the test before you celebrate. Ask what it predicts that was not already in the inputs, and whether that prediction can be checked and could come out wrong. If the answer is a specific, falsifiable forecast about the unseen, you may have reorganized your bricks into something new. If the answer is a nicer-looking pile of the same bricks, you have tidied up. The history of science is, to a surprising degree, the history of people who finally connected what everyone already knew.