1 Two Spikes, Forty Metres Apart
Outside the county town of Changxing, in Zhejiang, a limestone slope exposes its strata at an angle, layer upon layer. Read from bottom to top, they tell of a time more than 250 million years ago. The place is called Meishan. Two spikes have been driven into this slope.
One marks the Permian–Triassic boundary, close to the great extinction discussed in Essay 14. The formal boundary point and the principal extinction horizon are nearby, not identical definitions. In 2001, international approval made the Meishan section the global standard for this boundary.
The other, a short way below it, marks the base of the Changhsingian, the final stage of the Permian. It was ratified in 2005. The stratigraphic chart has several ranks, from largest to smallest: erathem, system, series, stage.
A stage is the smallest of the principal units and usually spans a few million years. For this essay, all we need to remember is that a stage is the smallest, and a series is larger. More than forty metres of rock separate the two spikes.
What, exactly, are the spikes fixed to? The layers where two kinds of conodont first appear. Conodonts are tiny, toothlike parts left by a long-extinct group of animals, commonly less than a millimetre across: you need a microscope to see them. The species marking the Permian–Triassic boundary is Hindeodus parvus; the one marking the base of the Changhsingian is Clarkina wangi.
The beginning of a geological age is defined on a tooth less than a millimetre long.
These markers are called golden spikes. Episode 10 explained the principle: a boundary on the time chart is defined not by a number but by a place, a particular point on an actual rock face somewhere in the world. Its numerical age is measured afterwards. Better dating methods may change the number; the point stays put.
A brief aside about the name. It is often said to borrow from the ceremonial golden spike driven in 1869 to complete North America's first transcontinental railway. I could find no conclusive source, however, identifying who first brought the expression into stratigraphy, or in which paper.
More practically, the “golden” spike at many sites is actually a large brass spike or marker, sometimes just a copper plaque. At quite a few ratified boundaries, there is nothing on site at all. The International Commission on Stratigraphy itself notes that a monument or spike-installation ceremony may follow ratification by one to three years.
Episode 14 used this same slope. Its marine carbonate beds are thickly interleaved with volcanic ash containing minerals that can be dated directly. Those minerals are what allowed the end-Permian extinction to be pinned to such a narrow time window. The same slope is both a place of measurement and a place of definition.
China has eleven golden spikes in all, every one of them in the south, distributed across Zhejiang, Hunan, Guangxi, Hubei and Guizhou. Meishan alone accounts for two. It is now a geological park, with enlarged fossil models standing on the slope. The actual marker is a tooth you need a microscope to see.
Here we need to establish the one distinction on which this essay rests. We will use it throughout the seven sections that follow. Three entirely different kinds of thing happen on that slope at Meishan.
First: does that layer of rock contain Hindeodus parvus? This is observation. Someone takes a sample, counts under a microscope and gets a result that someone else can reproduce.
Second: can the first appearance of this conodont be recognised in other sections around the world? Can it be matched to the ages of volcanic ash and to changes in carbon isotopes? This is methodological judgment. It requires experience and comparison. It can be overturned.
Third: of several suitable candidate horizons, which will be declared the official starting point of the Triassic System? This is a decision about standardisation, made by a committee vote.
Put another way: if someone were to find Hindeodus parvus five metres below the Meishan boundary tomorrow, the first would change and the second would have to be worked through again. The third would not automatically move with them. The Triassic System would still begin at the original layer.
This is not hypothetical. As Episode 10 described, the spike at the base of the Cambrian System was originally placed at the lowest occurrence of a particular burrow. In 2001, the same burrow was found more than three metres and more than four metres below the spike. The spike did not move.
These three things cannot stand in for one another. To say that the time chart is human-made is to speak of the third. The first and second remain matters of observation and testing. When the rest of this essay speaks of a “construct,” it means that third level. It does not mean that conodonts are fake, or that we are free to shuffle which layer lies on top of which.
Three words from this series will recur throughout the essay, so let us set them down here. A framework people make so that they can speak and compare is a construct. What it cannot contain, yet cannot make disappear, is the remainder. The act of making the construct is chiseling. A golden spike gives that last word one of its most literal forms.
Why, then, must we make a cut in continuous time at all?
Because geology across the world needs a shared set of coordinates, with no overlaps and no gaps. If a boundary were described only as “somewhere near such-and-such an event,” different regions would give the same name different starting points. Two people saying “Early Triassic” would not mean the same stretch of time. Defining a point as the lower boundary turns a problem of naming into a workable problem of correlation: compare your sequence of strata with this point, and if it correlates above the point, it is above the boundary.
A geologist put it plainly long ago: geological time itself is continuous, so any division contains an element of convention. The virtue of standardisation is precisely that, once everyone agrees, they need not keep arguing about the boundary. What lies within those forty-odd metres of rock at Meishan was not decided by a vote. Which layer gets the spike was.
2 A Hundred and Two Places, and a Spike Underwater
Before a spike is approved, a subcommission organizes a working group to compare candidate sections and signals, then votes on its proposal. The ICS description of the GSSP procedure requires support from more than 60 percent of the voting membership. With twenty voting members, that means at least thirteen, not twelve. Further approval stages follow; one ballot is not the whole procedure.
A successful proposal goes to the International Commission on Stratigraphy, where the executive officers and the chairs of the subcommissions vote again. Only after passing that round is it submitted to the executive committee of the International Union of Geological Sciences. Its ratification makes the boundary an official international standard. A formal paper follows, while the marker on the ground may still be one to three years away.
There is also a cooling-off period: for ten years after a new unit or golden spike is established, no proposal to change it may be submitted.
The technical requirements for candidate sections are equally detailed. There must be a clear, observable signal near the boundary; the section must be thick enough, its deposition as continuous as possible, and its exposure to tectonic and metamorphic disturbance limited. It must remain accessible for long-term research, preferably with several supplementary markers. The search for a golden spike today is a search for the most dependable global reference, not the most dramatic event. The two often lie in different layers. So how many spikes are needed altogether?
Recent editions of the International Chronostratigraphic Chart contain 101 Phanerozoic stage positions. Add the still-unnamed place for the Upper Pleistocene, and there are about 102 positions whose lower boundaries need defining. Of those 102, roughly eighty-one already have golden spikes.
Twenty-one places are still empty. Add the Precambrian spike at the base of the Ediacaran System, and the chart has about eighty-two ratified spikes in total. The International Commission on Stratigraphy has maintained this register since 1977. That works out at fewer than two spikes a year.
How long does one take? There is no deadline, and no average. The boundary working group for the Campanian Stage of the Cretaceous was reconstituted in the summer of 2017, with a quick finish expected. Ratification finally came in October 2022: five years from that restart alone.
The Berriasian, in the same period, has taken longer still. Its 2020 proposal failed to pass the Cretaceous subcommission. A new working group began in February 2021, held about twenty-five online meetings through 2025, and was still reassessing candidate markers and sections in 2026. When a proposal is rejected, it is not necessarily a matter of voting again. The work may have to start from scratch. The Precambrian, meanwhile, operates under a different system.
The Archean and most of the Proterozoic have long lacked the widespread, abundant fossil zones that allow fine correlations across continents. The international scale therefore divides that immense stretch directly by numerical age: one boundary is set at 2.5 billion years ago, another at one billion, using large, round numbers. These are called Global Standard Stratigraphic Ages.
Their difference from golden spikes is fundamental. With a Global Standard Stratigraphic Age, the age defines the boundary: the number itself is the boundary. With a golden spike, a physical point defines the boundary: a point in a section is the boundary, while the number merely calibrates it.
When dating methods improve, then, a golden spike's numerical age can change without moving the boundary itself. The June 2026 edition of the International Chronostratigraphic Chart adjusted the numerical ages of several lower boundaries, including those of the Anisian, Olenekian and Wuchiapingian, while explaining that such calibrations are distinct from the definitions of the golden spikes.
The criticism of round-number ages is straightforward: they are stable, but may have nothing to do with observable geological change. The general direction now is to replace purely numerical boundaries with correlatable rock records wherever possible. The base of the Ediacaran System is an established precedent. The Cryogenian subcommission still lists a rock-based golden spike for its lower boundary, replacing the old purely age-based boundary, among its central tasks.
Can a spike move? It can. Near Laibin in Guangxi, beside the Hongshui River, lies a section called Penglaitan. In 2004, it was ratified as the global reference point for the base of the Lopingian Series, the uppermost division of the Permian System. Later, the Datengxia water-control project was built downstream on the Qian River. The reservoir's backwater reached all the way to Laibin, leaving the spike permanently submerged. It could no longer meet the requirement of long-term access and repeated study.
The international working group therefore found a new stretch of strata on the bank beside the original section, and established a standard auxiliary section at Fengshan, outside Liuzhou. On 24 July 2023, the International Union of Geological Sciences ratified the relocation. An international standard had been drowned by a reservoir.
This was not an isolated case. In 2024, both the Aeronian and Telychian golden spikes of the Silurian were moved to new reference sections, a change the International Commission on Stratigraphy listed among that year's major updates to the chart. There had also been earlier moves between layers within a single section. The base of the Silurian System was once shifted to another layer in the same section because the original marker did not correlate well.
A spike may also stay where it is and deteriorate. More than twenty years after ratification, the site of the Cenomanian golden spike in France still lacked adequate protection.
Some have consequently argued that ratified golden spikes should be reviewed regularly. For “points” in ice cores or boreholes, even installing a permanent marker is difficult. All this sounds like a system springing leaks. There is another side to it.
The loss of an outcrop does not invalidate the world's chronology. Correlation is sustained by the original definition, its formal description, auxiliary sections, multiple biological and chemical markers, and the continuing network of calibrations. The golden-spike point is the name's ultimate anchor. It is not a place every user must visit and touch.
The other side is true as well: change things too often, and the names become unstable. The sixty-per-cent voting threshold and the ten-year cooling-off period are therefore not mere bureaucracy. They mark a chosen position between “open to change” and “not to be changed lightly.”
3 Eighteen Votes
The clearest view so far of what kind of institution this is came in 2024.
On 31 October 2023, the Anthropocene Working Group formally submitted its proposal to establish an Anthropocene, together with a corresponding stage. The lower boundary would lie in the annually layered sediments of Crawford Lake, Ontario, Canada, in the layer representing 1952. Its principal signal for global correlation was a sharp increase in plutonium isotopes.
Plutonium scarcely occurs in nature. Nuclear tests in the middle of the twentieth century scattered it worldwide, so it appears almost simultaneously in deposits across the globe. Annually layered sediments, or varves, accumulate on a lake bed one countable layer per year, like tree rings.
The lower waters of Crawford Lake remain unmixed with the surface all year, allowing the layers settling to the bottom to stack up undisturbed, year after year. The proposal also named several auxiliary sections: Beppu Bay in Japan, Sihailongwan maar lake in Jilin, China, and a peatland in Poland.
Why 1952? That was the year of the first hydrogen-bomb test. The working group had been at work since 2009. In its own 2019 vote, twenty-nine of thirty-three members had already supported formalisation with a mid-twentieth-century lower boundary. But the choice was not simply an anniversary of the first hydrogen-bomb test: it depended on correlatable plutonium changes in the sedimentary record. A calendar event and a core's actual marker are not automatically the same thing.
But in 2024 the Subcommission on Quaternary Stratigraphy voted twelve against, four in favour, with two abstentions. That was nowhere near the required sixty per cent. The executive body of the International Commission on Stratigraphy subsequently upheld the rejection. An official statement confirmed that the Anthropocene would not enter the International Chronostratigraphic Chart as a formal unit. It could continue to be used as an informal term.
The procedure was contested. The subcommission's chair and vice-chair lodged formal objections to the initiation and conduct of the vote. Science news at the time reported a dispute over procedural irregularities, including members' terms of office, who had the authority to call the vote, and whether the proposal had received adequate consideration.
In the end, the higher bodies did not overturn the result; they confirmed that the vote was valid. The precise account is this: members of the subcommission's leadership challenged the voting procedure, and the higher bodies ruled that the result stood. Both sides had substantial stratigraphic arguments.
Supporters argued that the middle of the twentieth century had left a highly synchronous, durable signal in the global sedimentary record, cross-checkable through several indicators. Radionuclides, especially plutonium, provided a marker closer to a truly simultaneous worldwide horizon than many traditional biological events.
To the objection that human impacts had begun much earlier, they replied that the lower boundary of a chronostratigraphic unit need never coincide with the earliest cause of the phenomenon it names. This was no different from other golden spikes: the task was to choose the reference horizon best suited to correlation. A causal process might extend over a long time while a boundary could still be defined at a synchronous point.
Opponents did not deny the existence of a global mid-twentieth-century signal. Their reasons lay elsewhere: elevating a very short stretch of recent deposits to a formal unit equal in rank to the traditional epoch might not improve stratigraphic correlation.
The onset of “human impact” varies enormously between records. Treating it as an event extending across time could accommodate those differences. Fixing it at 1952, by contrast, would shut out a great deal of earlier impact usually included in the Anthropocene.
The dispute, then, was not over whether the signal existed. It was over whether that signal deserved a formal place in the international hierarchy.
That place was more than a title. Had the proposal passed, stratigraphers everywhere writing about deposits laid down after the middle of the twentieth century would have had to date them using that name and that lower boundary. Since it did not, the word remains in everyone's vocabulary and off the chart.
Set this beside the two spikes in the first section, and the difference is clear. At Meishan, the question was which layer should get the spike. With the Anthropocene, the question was whether the chart should have this box at all.
Here is the distinction from the opening section in a ready-made example. Whether plutonium increased sharply is a matter of observation. Whether the signal permits global correlation is a matter of methodological judgment. Whether it should be an epoch, an event or an informal expression is something the data cannot decide on their own. Eighteen votes decided it, and twelve of them said no.
This is far more precise than a blanket statement that “science is socially constructed.” The chronostratigraphic chart is an institution constrained by evidence but requiring collective decisions about standardisation. Evidence can rule out many proposals, but cannot by itself select one unique answer. People must take the remaining step, and they must do so with the prescribed number of votes.
4 A Tree and a Table
The second construct is the tree of life. Its present form began with two different things laid one over the other.
The first was Linnaeus's Systema Naturae in 1735. Its first edition was astonishingly slim, scarcely more than ten pages. For the things of the natural world, he built a set of nested ranks: kingdom, class, order, genus, species, each enclosing the next. Phylum and family came later, producing the familiar sequence of phylum, class, order, family, genus and species.
The original task of this system was identification, naming and arrangement, not the reconstruction of common ancestry. In botany, Linnaeus himself distinguished a system convenient for identification from the “natural system” he wanted but could not yet produce. He has often been said to have insisted dogmatically on a fixed handful of ranks. More recent historical research considers that account inaccurate: he treated the ranks as practical means of organising information.
The second came more than two hundred years later. In 1950, Hennig set out a systematic argument in German; the English edition of 1966 brought it to the English-speaking world. The central change was that classification should express relationships of common ancestry.
The ideal natural group consists of an ancestor and all its descendants, leaving out not a single one. This is a monophyletic group. Resemblance of form is not the criterion, nor is who “looks more advanced.”
One of the most frequently cited consequences of this rule is that “fish” is not a valid group. Include the last common ancestor of all fish and all of its descendants, and you must include the four-legged animals on land. Either “fish” ceases to name a natural group, or it must take in the tetrapods as well.
The line in Episode 17 expresses another consequence of the same rule: birds are not distant relatives of dinosaurs; birds are inside the dinosaurs. The traditional class “Reptilia” is therefore inadequate too. Either it includes birds, or we acknowledge that it is not a natural group. These awkwardnesses are not taxonomists making trouble. They are the inevitable friction between a rule and a set of names established centuries before it.
In 1965, a third element joined the other two: the formal proposal that protein and nucleic-acid sequences are documents of evolutionary history. This did not change the tree as a geometrical framework. It changed which evidence could be used to build a tree and how a tree could be estimated. With all three in place, a problem emerged.
A fully resolved tree can keep branching through as many nested levels as it takes. “Phylum, class, order, family, genus, species” gives you only six or seven.
The clash is structural, not an accident caused by some troublesome group. A tree and a catalogue of ranks do not have equivalent shapes. Two sister lineages stand on equal terms in the tree, even if one later produces thousands of species and only three remain in the other. The tree puts them on the same level; the table requires someone to decide which gets to be a class and which a family. Strict adherence to monophyly consequently tends to leave only three ways out: change the boundaries of traditional groups, insert a great many intermediate ranks, or abandon some ranks altogether.
Beginning in 1990, two scholars proposed that third route: define a name directly as a particular clade, for example, “the most recent common ancestor of A and B, and all of its descendants,” without first stipulating its rank. This proposal eventually became a formal code of phylogenetic nomenclature, with names that can actually be registered and used.
It operates alongside the existing zoological and botanical codes. It is institutionalised, allows registration and has had a clear influence, but it has not acquired the general jurisdiction of the traditional codes.
The objections fall into five groups, none of them trivial. If the hypothesised tree changes, the scope of a node-defined name may jump about. Eliminating ranks would lose the quick cues that “class,” “order” and “family” have provided for so long.
Converting centuries of names into new definitions would itself cause another round of instability. Where fossils sit uncertainly in a tree, the reference points on which definitions depend can shift too. Having two sets of rules in parallel would raise the cost of communication.
Supporters reply that the traditional system also changes names as trees change, and that ranks never had a uniform biological scale in the first place. The dispute is therefore not between stability and instability. It is over what should remain stable: a name's position in the hierarchy, or the clade to which it refers.
Another set of terms changed the boundaries of names. In 1979, the terms crown group, stem group and total group were established. A crown group consists of the last common ancestor of the members alive today, together with all of its descendants.
The stem group comprises the relatives that branched off before that ancestor and have no descendants left alive today. Crown and stem together make the total group. Episode 20 used a fan as an analogy: the crown group lies above the handle; the branches that broke off below it form the stem group. One restriction matters: stem members are closer to this crown than to any other living group. Not every extinct branch that split earlier belongs to its stem.
Classification must then answer a frequently blurred question: does a name mean the crown alone, or also its earlier stem relatives? The crown contains the last common ancestor of its living members and all that ancestor's descendants, including extinct ones. A strict crown definition can exclude traditional members, but “crown” does not mean merely “alive today.”
What changes is not the number of labels but what their boundaries mean. We should also be clear about what this language does not solve. A stem group is itself a paraphyletic assemblage defined relative to a particular crown group; move a fossil's position in the tree, and its status as a stem-group member moves with it. A tree can keep branching without limit. A table has only six or seven tiers of boxes. To fit the tree into the table, something always has to be forced.
5 When a Tree Is Not Enough
The harder problem is that some histories were never trees to begin with.
Episode 2 described how horizontal gene transfer tangles the roots of the tree of life almost beyond recognition as a tree. It happens away from the roots too. Hybridisation, recombination and genes moving sideways between lineages cannot be represented by strictly bifurcating branches.
Hence the development of phylogenetic networks. On paper, their difference from trees is immediately visible: the lines of a tree only part; they never rejoin. In a network, two lines that have separated can be linked again by an arc. That arc represents hybridisation, or a stretch of genetic material passing from one lineage into another.
Then there are “forests.” The word has two uses. In one, different genes yield different trees, so researchers begin with a whole collection of trees rather than a single one.
Studies of bear genomes provide a ready example. Build trees from the genome one stretch at a time, and what you get is a collection, not one tree. There are two reasons.
First, the ancestral population already contained several versions of a gene, and chance played a part in which lineage inherited which version when they separated. Second, hybridisation continued after the split, allowing genes to cross back over.
Together, those two processes can leave many loci supporting different trees. The other use of “forest” is more formal: several non-intersecting trees, plus connecting arcs, represent the reticulate relationships between lineages.
A “forest” expresses the heterogeneity of history itself, not a choice of drawing style. There are “rings,” too. In 2004, a “ring of life” was proposed for eukaryotes, using a circular relationship to represent two ancestral genomes converging in a single lineage. The incorporation described in Episode 5 cannot be drawn as a tree, because trees divide but never join. This was a model of particular deep relationships, not a proposal that all life should be drawn as a ring. Its data and methods of inference have also drawn criticism since.
When, then, should we give up on a universal tree of life? One answer is very clear: if different genes, processes and levels did not share a single tree-shaped history, we should not force every pattern into one topology just to preserve “one tree.” Different questions can legitimately have different representations.
What the tree loses is automatic universality, not validity. For a great many local questions dominated by vertical inheritance, it remains an efficient, testable model. We simply need to ask, before using it, whether the history at issue actually branches like a tree.
The tree also comes with a set of counting units, and those units are made too. The three levels from the first section still apply: a unit defined by people does not make whatever is counted with it fictitious. Keep that distinction in mind through the next few paragraphs.
Different species concepts and delimitation criteria can divide the same biological continuum into different numbers of species. This is even clearer at the level of the genus. Genera do not have equivalent scope across research traditions and groups of organisms, and no uniform theory dictates how much difference a genus should contain. The thirty-five genera in Episode 22 are an immediate example: one had been merged out of existence, one removed wholesale from the list of apes, and one newly split off.
Constructed counting units do not invalidate statistics. Three practical things are needed: make the taxonomic practice explicit, test sensitivity by comparing genus-level and species-level results, and account separately for revisions to names and actual changes in diversity. Whether genus-level data can stand in for species-level trends is a question for the particular dataset.
Phanerozoic diversity curves have been rebuilt in just this way, step by step. The earliest version came from a database of marine invertebrates. It showed a rapid rise in the early Paleozoic, renewed expansion from the Mesozoic into the Cenozoic, and several troughs of mass extinction along the way.
From 2001 onwards, researchers pointed out the close coupling of raw taxon counts with sampling intensity and began standardising samples. Several peaks and troughs changed, as did the long-term slope. Later work addressed uneven spatial sampling and found that some Paleozoic peaks in the global curve, especially the Permian peak, became markedly smaller. Sampling more regions from a period and that period genuinely having more taxa worldwide are two different things.
Nor did the corrections leave a flat line across the Phanerozoic. The drops caused by mass extinctions, the recoveries that followed, and the broad expansion from Mesozoic to Cenozoic all remained. The safest conclusion avoids certainty at either extreme. “It rose monotonically throughout” goes too far; so does “after correction, there is no long-term increase at all.”
Another layer lies beneath the curves. Today they are drawn from public paleontological databases, and every record in those databases comes from a paper and from someone who entered it.
A database curve therefore holds four things at once: the original geological record, the history of publication, taxonomic revisions and the database's own rules. Open sources, traceable references and review by multiple people are precisely the means of reducing individual subjectivity, layer by layer.
One final, very practical matter: this tree is built on specimens, and possession of those specimens is very unevenly distributed.
Large numbers of type specimens and crucial comparative specimens entered major European and North American museums during the colonial era. Later comparative material and historical literature accumulated there too. So even if illegal collecting stopped altogether today, a researcher's ability to re-examine the specimen behind a paper would still depend on travel costs, visas, funding and access to collections. When a private collection is sold, lost or closed to researchers, that chain breaks.
In paleontology, reproducibility does not mean “making another dinosaur.” It means whether someone else can take that specimen out and look at it again.
6 A Procession
The third construct is the telling of a “history of life” itself. In 1965, an American popular natural-history series published Early Man, edited by Clark Howell.
The book contained a foldout illustration spanning several pages, drawn by Rudolph Zallinger and titled “The Road to Homo Sapiens.” Fifteen figures stood in a row from left to right. The last was a modern human.
The first five of those fifteen were Pliopithecus, Proconsul, Dryopithecus, Oreopithecus and Ramapithecus.
All were Miocene, and Episode 22 has just discussed four of them. Proconsul was split in 2015: most of the material from Rusinga Island was placed in a new genus, leaving the old name with a much narrower scope. The boundaries of Dryopithecus have expanded and contracted several times. It once held material from many places; today it refers only to a small European group.
Oreopithecus was an extraordinarily specialised lineage on an Italian island. It has been placed in several entirely different positions on the phylogenetic tree. Ramapithecus was merged in its entirety into Sivapithecus in 1982. The genus name disappeared.
As for Pliopithecus, first in line, it gave its name to the broad group of pliopithecoids. Under today's classification, that group diverged before apes and monkeys parted company. The genus that Episode 22 described as being removed from the ape list in 2024 was moved into this group. The first figure in the queue is not an ape. The original illustration also had two features that were later cut away.
One was the coloured time bars at the top, marking when the groups existed and for how long. Those bars were already saying something contrary to the procession: several lineages were alive at once. One did not leave the stage before the next stepped on.
The other was the explanatory text. It was quite explicit: Oreopithecus was “probably a side branch”; Paranthropus was “an evolutionary dead end.”
Under Ramapithecus, meanwhile, the caption said that some experts now considered it the oldest direct ancestor of humans. Seventeen years later, the genus no longer existed.
In the version that later spread, the time bars were gone, the words were gone, and the fifteen figures had been cut to five or six. All that remained was a row of bodies moving from stooped to upright, from short to tall. None of the original's warnings that this was a side branch or that was a dead end survived.
Advertisements, cartoons and posters of every kind copied the row over and over, taking it further and further from the original explanation. Some have called this visual convention, quite directly, a mistaken iconography of evolution.
This was not one person's lapse. A visual grammar had overridden the words. And its effects can be measured.
In 2008, researchers analysed evolutionary illustrations in thirty-one biology textbooks, from secondary school through university. Later papers reported these figures for the sample: images with clear linear elements appeared in about sixty-seven per cent of secondary-school textbooks, fifty per cent of introductory university textbooks for non-biology majors, and eighty-three per cent of introductory textbooks for biology majors. These images were no isolated occurrence, then; most textbooks had one. And they were especially concentrated on two examples: humans and horses.
The horse procession deserves a look too. Textbooks often line up horse fossils from small to large, their toes dwindling in number as their bodies grow from the size of a dog to that of a modern horse.
The history itself, however, has the shape of a bush, not a line. Many lineages existed together. Some grew larger, others smaller; some retained three toes until very late; most left no descendants. The few arranged in a row were picked out of that bush.
An experiment in 2011 was more direct. One hundred and eight university students read three kinds of diagram: a linear illustration of the kind common in textbooks, a branching diagram drawn as a ladder, and a branching diagram drawn as a tree.
The last two had exactly the same topology; only the way they were drawn differed. The linear textbook image was significantly more likely to be read as “one species turned into the next.” The tree-shaped diagram was least likely to elicit that reading. The authors also found that students with stronger backgrounds in biology were not wholly immune.
In other words, the sense of direction is more than a critic's intuition. It is something that really happens when people read images. Does that make a tree safe? No.
At every node in a tree, the left and right branches can be rotated without changing its topology at all. The top and the right-hand edge of the page do not mean “later” or “closer to a goal.”
All the tips surviving today have passed through the same amount of time. Put one lineage in the upper-right corner, and the topology is correct, but the picture looks like a ranking.
“Side branch” is a dangerous term too. In the topology of a tree, two living sister lineages descending from a common ancestor have no inherent status as main line and sideline.
A drawing in which “the trunk runs all the way here while the others peel off along the way” is often merely a layout choice. Only when we are explicitly discussing a particular crown group does “stem” carry the strict meaning given in the previous section. Time scales have a similar problem.
Compressing Earth's history into one year has the advantage of keeping the proportions exact: a position in the year corresponds directly to a percentage of the total duration. On that scale, the Cambrian begins around 18 November, the dinosaurs go extinct on 26 December, and the Miocene ends, where Episode 22 left off, sometime on the afternoon of 31 December.
The trouble is attention. Billions of years are packed into the earlier months, while a mass of familiar recent events crowds into the last few days of December, or even the final seconds. Compression into a single day makes it worse: differences measured in seconds are barely perceptible. Readers may remember that “human time occupies just one second before midnight” and forget the enormous differences in duration between the earlier stretches.
Lay time out along a corridor, a rope or a path across a campus, and walking it can restore a bodily sense of proportion. But how the route is divided, where the landmarks stand and where the guide stops to speak at greater length will redistribute attention all over again. Equal time per metre does not mean equal narrative weight per metre.
Which analogy distorts least? I found no large comparative study that could settle the question.
7 The Phase-Transition Column Changed Three Times
There is a more scholarly way to line up the history of life: make a list of major transitions.
The 1965 illustration was for the public. In 1995, two biologists proposed a much more serious framework, “the major transitions in evolution,” usually given as eight items. Molecular replicators to groups of molecules in compartments.
Independent replicators to chromosomes. RNA serving as both gene and enzyme to a division of labour between DNA and proteins. Prokaryotes to eukaryotes. Asexual clones to sexual populations.
Single-celled to multicellular organisms. Solitary individuals to societies with non-reproductive castes. The final item was primate societies to societies with language.
The framework addressed a problem spanning scales. Again and again in life's history, changes have turned units that replicated independently and competed with one another into parts of a larger unit, while new forms of inheritance and information transfer emerged. One language would connect chromosomes, cells, eukaryotes, multicellularity, sociality and language.
This was also precisely where the framework later drew its sharpest criticism: it was a mixed bag. Establishing a higher-level relation between parts and wholes, changing a mode of reproduction, creating a new system of information encoding and undergoing endosymbiosis are not obviously the same kind of phenomenon. Philosophical reviews explicitly identify this as one of the framework's chief conceptual problems.
What happened next is interesting. Around 2011, a collection of essays appeared devoted to reassessing the framework. Meanwhile, another group of researchers narrowed its attention to a single, more focused question: how do units that could once reproduce separately and undergo selection separately form a new individual capable of participating in selection at a higher level?
This approach is narrower than the original list, but its criteria are much more consistent. In 2026, researchers were still debating whether such transitions can be reversed, how to define the criteria, and whether the transitions can be produced in a laboratory. This is no settled historical framework. It is a research programme still running.
In 2015, one of the original authors undertook a revision himself. He explicitly proposed rebuilding the theory of twenty years earlier. Instead of forcing every candidate under one criterion, he would treat transitions in individuality and innovations in systems of inheritance and information as two separable dimensions.
The result: sex was demoted. The move from asexual clones to sexual populations was no longer treated as a major transition in the same sense as the formation of a new, higher-level individual. The endosymbiotic origin of plastids, meanwhile, was added.
Twenty years after a pioneering list appeared, its own author had removed an item, added another and changed the criteria.
By 2022, a philosophical review stated the point outright: over thirty years, “major transition” had acquired several meanings. Without explicit criteria, “major” can easily dwindle into “the researcher thinks this is especially important.” The safer practice now is to specify, each time, whether the subject is a new level of individuality, a new inheritance system, a new structure of cooperation or a large-scale ecological impact.
This bears directly on our series, because this series has a list of its own.
Episode 10 defined a “phase transition” as a change in the unit of selection and broke that definition into three testable questions. Have units that once replicated independently lost the ability to reproduce on their own? Has a new reproductive entity emerged that bears fitness as a whole? Has some mechanism systematically suppressed conflict among the lower-level units? Only when all three hold do we call it a change of level.
Those three questions cover only part of the 1995 list. They concern levels of individuality alone: who reproduces, and who bears fitness. They leave the information-systems half entirely untouched. Sexual reproduction, discussed in Episode 6, is therefore an item on the classic list but not under this series' criteria. That essay described it as “reinforcement.” The two lists are not interchangeable. We need to remember the mismatch.
Once that criterion was in place, it sent us back to change this series' own outline three times.
Episode 12. The outline originally had “yes” in the phase-transition column. Test the move onto land against the three questions, one by one, and none holds. When fish entered shallow water, plants came ashore and arthropods began breathing air, they did not combine into new reproductive individuals at a higher level.
That entry became three lines. Change in the unit of selection: no. Changes in ecosystem feedback and the state of Earth's surface: yes, and very strong ones. Environmental outputs becoming the material basis of later niches: direct material evidence.
Episode 16. The outline also said “yes.” After the three questions were applied, “eusociality” as a whole could not be given a single yes or no. In some lineages, workers never develop functional ovaries at all. Even when a colony loses its reproductives, the workers cannot switch back.
In others, suppression is merely behavioural and hormonal: once the main reproductive dies, some members begin laying eggs. The entry changed to “completed in some lineages, only partly completed in others.”
Essay 21 reserved a place for reversals. Applying the three questions to five candidates excluded three and gave two partial passes.
An earlier problem had been embedded in the outline from the start. “Changing levels” had been doing two jobs: one was a change in the unit of selection; the other was “the remainder of one level becoming building material for the next.” Oxygen was waste from cyanobacteria; later it rebuilt the chemical environment of the entire planet.
This is exactly the kind of mixture for which the 1995 list was criticised. The tightening of the definition in Episode 10 was the same move as the 2015 revision, on a much smaller scale. Lists of this kind have another flaw that is harder to guard against.
If a list includes only changes that form higher-level wholes, leaving out simplification, disintegration, failed alliances and lineages that go for long stretches without moving up a level, its visual arrangement points in one direction by default. This series' defences are five essays deliberately featuring no phase transition, Episode 21 devoted entirely to subtraction, and Episode 8's account of more than twenty independent origins of multicellularity, most of which did not go far.
As for the hindsight implicit in “only transitions that left descendants make the list,” I found no central paper that had tested it quantitatively. What we can say is that “major” is often a judgment made looking back.
A framework that can be revised by its own criteria does not leave us knowing nothing. Historical sciences have their own way of working: start with the traces that remain today, compare hypotheses about a common cause in the past, then look for the crucial trace that can exclude competing explanations. Ash beds, isotope anomalies, shocked minerals, bone pathologies and molecular sequences are all traces.
The fact that a past event cannot be replayed does not make hypotheses about it untestable. One past cause can leave effects in several records at once: lithological, biological, chemical and magnetic. It leaves more effects than are needed to establish it, so new evidence for testing it may still turn up later.
8 The Remainder Lies on the Extended Line
Now turn the camera on this series itself. This is Episode 23, and it is the last.
There are twenty-three essays, their framework arranged by time. The first addresses the boundaries of life; the last is the one you are reading. Placing twenty-two essays in a row performs four acts in its own right, none of which is written into any individual essay.
First, selecting events. Out of everything that happened in more than three billion years, twenty-two things were chosen to become essays. Why is there one on colour and none on smell? That choice is invisible in the text.
Second, allocating scale. The first nine essays cover more than three billion years, from life's beginnings to before the Cambrian. Ten later essays cover the Cambrian through the Miocene, a little over half a billion years. Three occupy no time slot: two concern method, and one concerns a direction of change present from beginning to end.
On average, one of the early essays covers as much time as seven of the later ones. Compressing Earth's history into a year has a defect everyone notices: recent events all crowd into the last few days of December. This series' contents have the same defect, in a milder form.
Third, deciding which events to link into a causal chain. Is there a line between the Great Oxidation and the eukaryotic cell? This series drew one. That line was chosen; it did not grow there of its own accord in the evidence.
Fourth, looking back from the present to decide which things in the past count as important.
A scholar of narratives of evolutionary history distinguished these four acts in 1992. He also made a point that matters even more to this series: historical reconstruction must organise scattered evidence into an account of succession and ancestry, but narrative structure must not manufacture ancestor–descendant relations on the evidence's behalf.
Narrative is not in itself teleology. Across the sciences, narratives organise processes of development, and a narrative can contain time and causation without a purpose, a protagonist or an endpoint. A “history” can stand without anyone having to win in the end.
Something else must be put in plain view. At its opening, this series said that, for the first time, it would take the role of chiseler away from humans. Natural selection cuts up the space of possible forms without purpose or bias, and still the construct cannot close.
That is true at the first level. At the second, the chiselers have never ceased to be human: the people classifying, the people voting, the people drawing diagrams, and the person arranging these twenty-three essays. A construct made by a chiseler without intentions is described by a group of chiselers who have intentions.
The series tries to resist that reading: five essays contain no transition to higher-level individuality, not no events at all. Multicellularity arose repeatedly without every origin leading to complex specialization. An entire essay is devoted to subtraction.
Episode 20 ended with no one waiting for a vacancy. Episode 22 dealt with the vast majority that had been pruned away. Yet one thing cannot be resisted.
Essay 22 stops with Miocene apes, without developing the later human branch. Yet a sequence of points ordered in time may still invite readers to connect them and extend the line beyond its final point.
This is not conjecture. The experiment in Section 6 addresses precisely this: figures set in a row lead people to read “one became the next.” Arranging twenty-two essays chronologically is the same act, on a much larger scale, using words instead of pictures.
The procession in Section 6 is what that extended line looks like once it has been drawn. Episode 22 discussed almost all of its first five figures.
This series stops exactly between the procession's fifth and sixth figures. Of the ten that follow, this series has written not a word. A reader may nevertheless supply the remaining procession in imagination.
The remainder is not in the evidence, the ruler or the model. It lies on the extended line. Hence the sixteenth kind: the remainder lies on the extended line.
Keep it distinct from Episode 12's “remainder in the sequence.” That essay argued that the order supplied by the record is not necessarily the order in which events occurred, because footprints and bones are two sieves opening onto different environments.
This kind has nothing to do with the record. Even if the sequence is entirely correct, the act of arranging it in a row gives readers a direction, and they will take one more step along it. In the former case, the error lies in the sequence as arranged; in the latter, in the step taken once the arranging is done.
On the scale that compresses everything into a year, this series' main account stops on the afternoon of 31 December. Of the remaining hours, not a word.
Here, too, the series' law acts upon the series itself. The construct cannot close: twenty-three essays cannot lay out more than three billion years in full, nor can they block that extended line.
The remainder cannot be eliminated. Leaving it entirely unwritten in Episode 22 did not make the extra step disappear. It merely moved the step off the page and into the reader's mind. Spread out the sixteen kinds of remainder, and they fall roughly into four families.
In evidence and records: outside the evidence, in the sieve, in the sequence, in the mismatch between the two prunings. In rulers and coordinates: in the coordinate axes, in the shape of the grid, in a ruler that does not exist, a quantity without a unit, in a coordinate system that shifts as it is occupied.
In explanations and models: in explanation itself, two irreducible descriptions, in the act of making the unmeasurable measurable, in independence, in the null hypothesis. In history and its telling: in the counterfactual, on the extended line.
The next time you read a conclusion, you can work through those four families and ask: in which one does the remainder fall?
But this map is itself a construct. Grouping sixteen kinds into four families makes another choice. The grouping could differ; the number of families could differ. Once the grouping is done, whatever fits awkwardly will start to look unimportant.
An essay about constructs is itself a construct. That is not a modest disclaimer; it has specific content. This essay selected three things to call “constructs this series built for itself” and left other things unchosen: the unit called a species, for instance, the word “extinction,” or the metaphors the series has used throughout. Finally, we return to the spike.
To describe time together, people choose a reference point in the rock. The layers already had an order; the spike gives them a shared chronological language across oceans. Here the series' metaphor becomes almost literal: a hammer, a brass spike, a rock face.
The layers beside the spike do not lose their names: each belongs to the chronological unit on its side. What the names do not exhaust is what happened in those centimeters, or all the reasons for choosing this point rather than another. A boundary gives us a shared address, not a finished history of what happened there.
This series has driven two spikes of its own. One, in Episode 1, fixed where to begin. The other, in Episode 22, fixed where to stop. Between them lie these essays. Outside them lies the remainder.
The next spike will go where Episode 22 stopped. That place is empty for now, as empty as the twenty-one places still unfilled among the hundred and two.
This series has always written from outside the ladder; even when Section 6 dismantled the procession, it stood outside it. The next series, on ancient humans, will write about what lies on that extended line. It cannot stand outside.
Its first task will be the same as Section 6's: take the procession apart and look at what each figure is in its own right, one by one, rather than at its place in the queue. This series has already handed over the tools.
Observation, methodological judgment and decisions about standardisation must remain three distinct levels. The remainder has sixteen kinds of source. Crown group and stem group provide a language for drawing boundaries. And the most important thing to take away is this: whenever you put a set of figures in a row, readers will extend that row for you.
Another paragraph will not abolish the extension; it moves it onward. Wherever the next series ends, another line will invite continuation. We will choose another point and remember that the named layers on either side exceed their names.
The story stops here. The rocks remain, and the tree keeps branching. For now, it is enough to look back at the measures we used.