Timelines, in words

The shared past

From the Big Bang to now

Every track that looks forward starts here. The gaps between these steps are the argument: the waits are enormous and they are not shrinking evenly.

1. 13.8 billion yrs ago

   Big Bang

   Spacetime, matter, the first light.

   Lemaître, 1931 · Penzias & Wilson, 1965

2. ↓ 9.2 Gyr later — follows from it

   4.6 billion yrs ago

   Earth forms

   A pale rock around a young star.

   Patterson, meteorite dating, 1956

3. ↓ 800 Myr later — nothing guaranteed it

   3.8 billion yrs ago

   First life

   new substrate for information: the genome

   Chemistry starts to copy itself 800 million years after the planet formed — almost as soon as it had cooled, and only 17% of Earth’s history in. Arriving that fast is the argument: on this planet, life is the easy part. It is also the first place information is ever stored: the genome.

   biogenic carbon, Isua and Nuvvuagittuq, ~3.7–3.8 Gyr · stromatolites, Strelley Pool, ~3.5 Gyr

4. ↓ 1.4 Gyr later — a constraint was lifted

   2.4 billion yrs ago

   The Great Oxidation

   Cyanobacteria had been making oxygen for hundreds of millions of years; once the mineral sinks filled up, it started accumulating. It poisoned most of what was alive — and it made aerobic respiration possible, worth roughly sixteen times more energy per meal. The one release the biosphere produced for itself.

5. ↓ 600 Myr later — nothing guaranteed it

   1.8 billion yrs ago

   The complex cell

   One cell swallows another and keeps it, and everything large and visible on Earth descends from that single symbiosis. This is the bottleneck of the whole line: two billion years of waiting, and it happened once. Oxygen made it affordable and did not make it happen — 600 million years pass after the Great Oxidation before anything uses it. Compare the dot after this one: multicellularity evolved independently dozens of times. A step that happens dozens of times is not a filter. This one is. Dating uncertain (~1.8–2.1 Gyr).

   Lynn Margulis, endosymbiosis, 1967 — rejected by fifteen journals

6. ↓ 1.2 Gyr later — follows from it

   575 million yrs ago

   Many cells, one body

   A billion and a quarter years of almost nothing, and then bodies. The wait is the surprise, not the step: unlike the complex cell, which happened once, multicellularity evolved independently dozens of times — in animals, plants, fungi, algae. The hard part was the cell, not the body. Ediacaran bodies ~575 Myr; the Cambrian, 538.8 Myr, adds skeletons, eyes and predation.

   Avalon assemblage, Mistaken Point ~575 Myr · Cambrian base 538.8 Myr (ICS)

7. ↓ 345 Myr later — nothing guaranteed it

   230 million yrs ago

   Age of dinosaurs

   Another 345 million years, and then 165 million years of reptiles — dominant, various, and with no cumulative culture at all. Intelligence is not where evolution was heading.

8. ↓ 164 Myr later — a constraint was lifted

   66.0 million yrs ago

   The asteroid

   It created nothing. It made room — and what it made room for was size. Mammals had been here for over 100 million years and stayed small the whole time; within 10 million years of the impact they were in every large-bodied niche there was, and a large brain needs a large body to carry it. The stronger claim — that without it we are not here — is not what the evidence says: the placental radiation, primates included, appears to start in the Cretaceous and run straight across the boundary.

   The long version

   The question this dot used to answer too confidently: was the asteroid necessary for us?

   Probably not for our existence, and almost certainly for our size. The molecular evidence puts the origin of the placental mammals — the primate stem among them — back in the Late Cretaceous, which means our lineage was already alive and small when the rock arrived. A birth-death analysis of the radiation finds it running continuously across the boundary rather than being started by it.

   What does change at the boundary is body size. For more than a hundred million years, under dinosaur dominance, mammals are almost all under a few kilograms. Within ten million years after it, they are whales, horses, elephants, and primates large enough to carry an expensive brain. The niche had to be empty first, and the impact is what emptied it.

   It may not even have been the only cause: some analyses argue dinosaur diversity was already falling for millions of years before impact, which is contested. Either way this stays a release — the classification was always the accurate part of this dot.

   Luis & Walter Alvarez, 1980 · Chicxulub confirmed, 1991 · placental radiation uninterrupted by the K-Pg — Liu et al., PNAS 2017 · pre-impact dinosaur decline — contested

9. ↓ 63 Myr later — nothing guaranteed it

   2.6 million yrs ago

   Stone tools — the genus Homo

   Oldowan flakes — though the tools arrive before the toolmaker: Lomekwi 3, in Kenya, is 3.3 Myr, older than the genus itself. Acheulean hand axes follow with H. erectus around 1.76 Myr. Homo sapiens is only the last 12% of this.

   Gona & Ledi-Geraru, Ethiopia · Leakey, Olduvai · Lomekwi 3, Kenya ~3.3 Myr — Harmand et al., 2015 · Kokiselei, Kenya ~1.76 Myr

10. ↓ 1.6 Myr later — nothing guaranteed it

    1.0 million yrs ago

    Fire

    Energy from outside the body, under control — and with cooking, more calories out of the same food. The gut can shrink, the brain can grow, and the night and the cold stop being walls. Nothing in the stone tools guaranteed it: 1.6 million years pass between the flakes and the first hearths anyone can date, and it happened once. Dating contested — ~1 Myr for the earliest burnt deposits, ~400 kyr before habitual use is widespread.

    The long version

    Two different claims get called “fire”, and only one of them is a threshold. Using a fire you found — a lightning strike, a burning coal seam — is opportunism. Making one, on demand, in a place you chose, is the step: it is what turns fire from weather into infrastructure.

    The archaeology can barely tell the two apart. Burnt bone and reddened sediment survive; the act does not. What counts as evidence is repetition — the same spot, used again and again — which is why Gesher Benot Ya’aqov carries more weight at 790,000 years than older single deposits do at a million.

    The anatomical argument runs well ahead of the dirt. Wrangham’s case is that cooking is what paid for us: cooked food gives up more of its energy for less digestion, so the gut can shrink and the brain — which costs about a fifth of our resting metabolism — can grow. Read backwards from Homo erectus’ body, that puts fire near 1.8 million years ago. The hearths are not there. Both halves of that disagreement are inside this dot’s date.

    No gold ring: fire changes what the body can afford, not where information is kept.

    Wonderwerk Cave, South Africa ~1 Myr · Gesher Benot Ya’aqov, Israel ~790 kyr, hearths reused across one occupation · Richard Wrangham, Catching Fire, 2009

11. ↓ 700 kyr later — follows from it

    300k yrs ago

    Homo sapiens

    Anatomically us — Jebel Irhoud, Morocco. And then, for almost all of those 300,000 years, nothing much.

    Jebel Irhoud — Hublin et al., 2017

12. ↓ 235 kyr later — nothing guaranteed it

    65k yrs ago

    Minds that make art

    new substrate for information: culture

    Pigment, ornament, trade networks: the first evidence of a ratchet — knowledge that survives the person who had it. Chimpanzees innovate and never accumulate; we copy even the steps we do not understand. Neanderthals were doing it too (dating contested): two lineages crossed, one survived.

    Blombos Cave · Sulawesi · Chauvet

13. ↓ 53 kyr later — a constraint was lifted

    12k yrs ago

    The climate goes quiet

    The Pleistocene swung 8–15 °C in a few decades, ~25 times over the last glacial. The ice sheets melt, the oscillator switches off, CO₂ climbs from ~190 to ~270 ppm — below which wheat and barley barely yield. Farming becomes physically possible.

    The long version

    Not the polar caps. Greenland and Antarctica sat through the entire Pleistocene and are still here — they were never the oscillator. What melted were the continental ice sheets: the Laurentide, three to four kilometres thick over Canada and reaching down to where Chicago and New York are, and the Fennoscandian over Scandinavia and the Baltic. Those are gone, and they have not come back.

    What your ancestors lived through, 15,000 years ago:

    • Global mean ~4–5 °C colder; sea level 100–120 m lower — dry land from Siberia to Alaska, from Britain to France
    • CO₂ around 200 ppm, air full of dust, northern Europe a cold, windy mammoth steppe
    • At 14,700 years ago the Bølling–Allerød: Greenland warms ~10 °C in a matter of decades
    • At 12,900 the Younger Dryas: Lake Agassiz — a meltwater lake dammed by the retreating Laurentide — bursts into the North Atlantic, the overturning circulation stalls, and Europe snaps back to near-glacial conditions for 1,200 years
    • At 11,700 it ends, again in about a decade. That decade is the start of the Holocene

    A person born then could watch the world change beyond recognition inside one lifetime. Their grandchildren lived somewhere else, without moving.

    Why it stopped: no Laurentide, no proglacial lakes left to burst, so nothing to keep flipping the North Atlantic. Low orbital eccentricity is damping this interglacial too — which is part of why it is unusually placid, and unusually long.

    Greenland ice cores · Richerson, Boyd & Bettinger, 2001

14. ↓ 1 kyr later — follows from it

    11k yrs ago

    Agriculture

    Within ~1,200 years, independently, in at least seven places that had never met. Not a better idea — a constraint lifted.

    Fertile Crescent · Yangtze · Mesoamerica · Andes · New Guinea · Sahel · E. North America

15. ↓ 5 kyr later — follows from it

    3200 BCE

    Writing

    new substrate for information: writing

    Memory leaves the body. Knowledge stops decaying at the speed a generation forgets. Nobody signed it — the first author known by name, Enheduanna, comes 900 years later.

    The long version

    What the earliest tablets say is: how much barley, how much beer, to whom. The British Museum holds one from ~3100–3000 BCE that does nothing else — the sign for beer, an upright jar with a pointed base, three times, beside numerals for the amounts.

    It is not the first tablet, and no single one is: proto-cuneiform begins around 3300 BCE at Uruk and survives in thousands of fragments. But the character of the earliest writing is not in dispute, and it is neither law nor literature. Hammurabi’s code comes around 1754 BCE — some fifteen centuries later, which is the distance between us and the fall of the Western Roman Empire. The famous object is not the first object: Hammurabi is a two-metre basalt stele in the Louvre, and the Uruk tablets are the size of a phone and unreadable as prose.

    And it may have been invented twice. Bone and ivory labels from tomb U-j at Abydos, in Egypt, carry hieroglyphs dated to roughly 3400–3200 BCE — the same centuries, twelve hundred kilometres away, and the tags themselves are strikingly like the clay ones from Uruk. Nobody can say with confidence who was first, or whether either knew of the other. The printing press, further along this line, is the same shape of story read the other way: the same invention, a different context, a different outcome.

    Same point on this axis as Beer as wages on the Alcohol & cocktails track, and not by coincidence: writing begins as a way of counting something worth counting exactly.

    Uruk, Sumer, ~3200 BCE · British Museum, proto-cuneiform tablet recording beer rations, ~3100–3000 BCE · Hammurabi, ~1754 BCE · Abydos tomb U-j labels, ~3400–3200 BCE

16. ↓ 3 kyr later — follows from it

    500 BCE

    Millions, administered

    Writing’s first use at species scale is not literature — it is coordination between people who will never meet. Achaemenid Persia runs tens of millions across three continents, and every instrument it does it with is an information technology: Aramaic as an administrative language nobody had to be born into, the Royal Road, a relay post that moves a message faster than any single man can travel, standardised coinage. Qin China makes the same move from the other end in 221 BCE, by standardising the script itself.

    Cyrus, 550 BCE · Darius I, 522–486 BCE — satrapies, the Royal Road, the angarium relay · Qin unification, 221 BCE · population estimates 17–35M, contested

17. ↓ 605 yr later — a constraint was lifted

    105

    Paper

    Movable type is not what the printing press was waiting for — a cheap surface was. Parchment is livestock: one large book is a herd, which caps the run no matter how cleverly the page is set. Cai Lun’s process is dated to 105; it reaches the Islamic world after Talas in 751, Spain by ~1150, Italy by 1276. Print follows within two centuries of paper being made in Europe, and not before.

    The long version

    The number that makes the argument: the copies of the Gutenberg Bible printed on vellum took something like 170 calfskins each. Around thirty-five were made that way and roughly a hundred and forty-five on paper. Even at the birth of print the parchment edition was the luxury exception — because a press can set a page in an afternoon and a herd takes years.

    So the constraint print broke was never the writing. It was the surface — and the surface was lifted by other people, three centuries earlier, for entirely unrelated reasons. That is what a release is on this axis: nothing is created, a ceiling is removed, and what was waiting goes at once.

    Read this together with The printing press further along the line, which argues that movable type existed in China from ~1040 and in Korea in metal by 1377 and cascaded in neither place. Paper was there too — first. So paper is necessary and not sufficient, which is exactly why the two are separate dots carrying different threads.

    Cai Lun, 105 CE — papermaking archaeologically earlier, ~2nd c. BCE · Samarkand after Talas, 751 · Xàtiva, ~1150 · Fabriano, 1276

18. ↓ 523 yr later — follows from it

    628

    Zero

    This dot is 628, which is neither the first zero nor the one Europe used: it is the year the Indian mathematician Brahmagupta writes down the *rules* — zero as a number you can add to and multiply by, not a gap left in a column — and rules are the part a trade network can carry. The Maya had a positional zero six centuries earlier and nothing carried it. Europe met these digits in Arabic books and named them after the messenger, which is the whole story in one word: the Arabic book it learned them from calls them Indian in its title.

    The long version

    Try multiplying MDCCLXXVI by XIV. Roman numerals record a result; they cannot carry, so the work has to happen on an abacus. Positional notation puts the machinery inside the notation, and a procedure anyone can run without understanding it is what a computation is.

    Around 825, in Baghdad, al-Khwārizmī writes the rules up as On the Calculation with Hindu Numerals. The Arabic is lost; the Latin translation survives as Algoritmi de numero Indorum — which is where Europe got the digits, and where we got the word algorithm, from his name. The name of the thing came the same way: Sanskrit śūnya, empty, becomes Arabic ṣifr, which Fibonacci latinises as zephirum and Italian wears down to zero — while the same Arabic word by the other road gives cifra, and so cipher.

    Then a merchant’s manual, not a treatise. Fibonacci learns the digits as a boy in Béjaïa, where his father ran a Pisan trading post, and in 1202 publishes the Liber Abaci: currency conversion, profit shares, interest, weights and measures, worked in the new notation for people who did those sums for a living. One puzzle in it asks how fast a pair of rabbits breeds, and the answer is the sequence now named after him — the only thing he is popularly remembered for, and not his either. Indian prosodists had it centuries before, from counting the arrangements of long and short syllables in a verse, and Hemachandra set it out plainly thirty years before the Liber Abaci was written. Even the famous part of this story was carried rather than invented.

    Europe was not ignorant of them; it was slow. The digits reach a Spanish manuscript by 976, and from the 980s Gerbert of Aurillac — the future Pope Sylvester II — was promoting them. They had arrived as nine digits without the zero, so there was no positional system to adopt and nothing took: what Europe was missing for two centuries is the thing this dot is named after. Three centuries on, the people they helped most were still refusing: Florence banned Hindu-Arabic numerals in bookkeeping in 1299, too easy to forge, a 0 into a 6 and a 1 into a 7. Most merchants crossed over only by about 1500, taught less by Fibonacci’s book than by the abbaco schools that grew out of it. So the notation that made modern accounting workable spent five centuries in Europe before the bankers would use it, two of them under a ban.

    No gold ring: this is a way of representing number, not a new place to keep information.

    Brahmagupta, Brāhmasphuṭasiddhānta, 628 · al-Khwārizmī, On the Calculation with Hindu Numerals, c. 825 · Fibonacci, Liber Abaci, 1202 · the sequence in Sanskrit prosody: Pingala c. 200 BCE, Virahanka c. 700, Hemachandra d. 1172 · Codex Vigilanus, Spain, 976 — the digits’ first appearance in a Western manuscript · Gerbert of Aurillac, from the 980s · Statuto dell’Arte del Cambio, Florence, 1299 — some Italian accounts date the ban to 1280; 1299 is the better attested · Maya Long Count positional zero, by ~36 BCE · Bakhshali manuscript, dating contested

19. ↓ 822 yr later — nothing guaranteed it

    1450

    The printing press

    Johannes Gutenberg

    new substrate for information: print

    Knowledge becomes cheap, copyable without error, and cumulative. Movable type already existed — Bi Sheng ~1040, Korean metal type by 1377 — but with tens of thousands of characters and no competitive print market it never cascaded. The same invention, a different context, no revolution.

    Johannes Gutenberg, Mainz, ~1450

20. ↓ 237 yr later — follows from it

    1687

    Newton

    Nature turns out to be written in mathematics — and so can be engineered. Unthinkable without two centuries of print: he read Descartes, Galileo and Kepler in books.

    Isaac Newton, Principia, 1687

21. ↓ 82 yr later — follows from it

    1769

    The steam engine

    Thomas Newcomen, then James Watt

    Machines extend the muscle. Newcomen built the first working engine in 1712, decades before anyone could explain why it worked; Watt’s separate condenser made it worth running.

    Thomas Newcomen, 1712 · James Watt, 1769

22. ↓ 62 yr later — follows from it

    1831

    Electricity

    Michael Faraday

    Energy becomes portable, and convertible into anything. Volta’s pile in 1800, Faraday’s induction in 1831, Maxwell’s equations in 1865, the first grids in the 1880s.

    Michael Faraday, 1831

23. 1859

    Darwin

    Charles Darwin · Alfred Russel Wallace

    The point at which this axis becomes thinkable. Every claim on this line depends on it — that intelligence is not where evolution was heading, that two lineages crossed and one survived, that an asteroid made room — and before 1859 none of those sentences can be said. It changed no capability whatsoever, which is why no thread runs through it. Wallace reached the same idea independently, which is why Darwin published in a hurry: the argument was ready, and it was not going to wait for him.

    The long version

    This is the one dot on the trunk that changed nothing about what we can do. No tool follows from it, no ceiling is lifted, nothing gets cheaper. Newton is on this line because nature turning out to be written in mathematics means it can be engineered; there is no equivalent sentence for 1859. What Darwin changed is the reader, not the world.

    It earns the place anyway, because it is this axis’ own epistemology. A timeline that says intelligence was not the destination, that an asteroid made room, that culture is a ratchet, that a species can be replaced by a cousin it interbred with — that is a Darwinian reading of history from end to end. Before the Origin there is no way to draw this chart at all.

    And it is the second of the demotions. Copernicus takes the Earth out of the centre in 1543. Darwin takes the design out of us in 1859. Hubble, in 1924, turns our galaxy into one of billions. Each is the same shape: something we assumed was the point turns out to be an instance. The machines at the end of this line are the current instalment, and the question they raise — whether a mind is a special kind of thing or a common one — is the question this site is about.

    No thread runs through this dot, deliberately: it sits on the axis and outside the causal spine, because it is not a step in the staircase. It is the reason the staircase is visible.

    Darwin & Wallace, joint reading at the Linnean Society, 1 July 1858 · On the Origin of Species, 24 November 1859

24. ↓ 49 yr later — a constraint was lifted

    1908

    Nitrogen from the air

    Fritz Haber · Carl Bosch

    Every agriculture in history ran into the same wall: the reactive nitrogen a crop needs came only from biological fixation, lightning, or a mined deposit — guano, Chilean nitrate — and the deposits were being emptied. Haber makes ammonia from atmospheric nitrogen in 1909, Bosch turns it into a plant at Oppau by 1913, and the wall is gone. About half the nitrogen atoms in the body of anyone reading this came out of that process, and roughly half the people alive are fed by it: no other single step on this line is load-bearing for four billion people. It is also, on its own, one to two per cent of world energy — the release costs power, permanently, which is why it sits between electricity and computing rather than anywhere earlier. And the same reaction, run the other way, is the largest perturbation any species has made to the nitrogen cycle: eutrophied rivers, dead zones at river mouths, and nitrous oxide, which is three hundred times the greenhouse gas that CO2 is.

    Fritz Haber, 1909 · Carl Bosch, BASF Oppau, 1913 · Vaclav Smil, Enriching the Earth, 2001 — the “half the world is fed by it” estimate is his

25. ↓ 29 yr later — follows from it

    1937

    Computing

    Alan Turing · Claude Shannon

    new substrate for information: silicon

    Logic becomes a circuit — and knowledge becomes something that can be executed, not only read. Turing gave it the theory, Shannon showed that switches do Boolean algebra.

    Alan Turing, 1936 · Claude Shannon, 1937

26. ↓ 46 yr later — follows from it

    1983

    The Internet

    Vint Cerf · Bob Kahn

    new substrate for information: the network

    Machines extend the mind. Every copy is instant and free, everywhere at once. Cerf and Kahn designed TCP/IP in 1974; on 1 January 1983 every host on the ARPANET switched over at once, and the separate networks became one. The Web came in 1989.

    the TCP/IP switchover, 1 Jan 1983 · Cerf & Kahn · Berners-Lee

27. ↓ 24 yr later — a constraint was lifted

    2007

    The GPU

    CUDA opens the graphics card to general computation. The chip was built to draw polygons faster for games, and turns out to be the cheapest parallel arithmetic anyone had ever sold. None of that was planned.

    The long version

    Backpropagation was published in 1986 and then largely sat there. The idea was not missing and the data was not missing; what was missing was affordable parallel arithmetic. A graphics card does one thing extremely well — the same simple operation across millions of values at once — because that is what shading a screen sixty times a second demands. It is also, exactly, what training a neural network demands.

    In 2012 Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton trained AlexNet on two GeForce GTX 580s — consumer gaming cards, 3GB each, five to six days of training — and cut the ImageNet top-5 error from 26.2% to 15.3%. The field turned inside a single afternoon, and every model since sits on that lineage.

    Nobody built this for intelligence. The substrate that carries model weights was paid for by people who wanted more frames per second and inherited by something else entirely — which is precisely what a release is on this chart: it creates nothing, it lifts a constraint, and what was waiting goes at once. The same shape as the beer rations under Writing, sixty-two centuries earlier.

    CUDA 1.0, NVIDIA, 23 June 2007 · Krizhevsky, Sutskever & Hinton, ImageNet Classification with Deep CNNs, NeurIPS 2012 · two GeForce GTX 580, 3GB each

28. ↓ 10 yr later — follows from it

    2017

    Generative AI

    new substrate for information: model weights

    Knowledge stops being written and starts being learned — carried in weights, not sentences. Backpropagation in 1986, AlexNet out of Hinton’s lab in 2012, then the Transformer removes the sequential bottleneck and scale becomes a dial you can turn. ChatGPT in 2022 is the grid, not the discovery — it has its own dot, two along. The corpus is the web itself: no Internet, no model.

    Vaswani et al., “Attention Is All You Need”, 2017

29. ↓ 5 yr later — follows from it

    2022

    ChatGPT

    The capability was years old in the labs; this is the week it reached everyone, and the dot before this one already draws the distinction — ChatGPT is the grid, not the discovery. What changed on 30 November 2022 is that talking convincingly to a machine stopped being a demonstration and became something a hundred million people did inside two months, the fastest adoption of any consumer software on record. A technology nobody outside the field could evaluate became one everybody had an opinion about, which is a different kind of event and belongs on a different dot.

    OpenAI, ChatGPT, 30 November 2022 · ~100M users by January 2023

30. ↓ 3 yr later — follows from it

    2025

    AI works for hours

    Conversation is a short act. What changed at the turn of 2025 is duration: models began holding a goal across long autonomous stretches — writing and debugging code, driving tools, recovering from their own mistakes — without a person in the loop at every step. METR measures it as a task horizon, the length of job a model finishes unsupervised about half the time: nine seconds for a GPT-3 agent in 2020, fourteen and a half hours by February 2026, doubling every seven months for five years and nearer every four since 2024. Not a better conversation. A longer one with itself.

    The long version

    The number that turns this from an impression into a measurement. METR gives frontier models a set of real software and reasoning jobs, each labelled with how long a human professional takes on it, and reports the task horizon: the length of job the model finishes unsupervised about half the time.

    • 2020, an early GPT-3 agent — about 9 seconds
    • February 2026, a frontier model — about 14.5 hours

    That is roughly a doubling every seven months across six years, and nearer every four months since 2024. It is the cleanest quantitative version of what the right-hand end of this trunk keeps claiming, and unlike a benchmark score it has a unit anybody can feel: how long you can walk away.

    Deliberately no model name on this dot. Versions date within months and choosing one is an argument; the horizon is a measurement, and it is the thing that actually changed. The step is also why the dot after this one was possible at all — a machine cannot produce a result nobody has if it cannot hold a problem for longer than a conversation.

    METR, Measuring AI Ability to Complete Long Tasks, March 2025 · HCAST, 50% reliability · 9 s (2020) → 14.5 h (Feb 2026)

31. ↓ 1 yr later — nothing guaranteed it

    2026 — now

    AI starts innovating

    On 20 May 2026 a reasoning model disproves Erdős’s unit distance conjecture, open since 1946. Learning and creating are not the same act, and nothing about the first guaranteed the second.

    The long version

    The question was never whether a machine could do mathematics. It was whether it could produce knowledge nobody had. Erdős asked in 1946: place n points in a plane, how many pairs can be exactly one apart? For eighty years the belief was that a square grid is essentially the best you can do. It is not. The model found an infinite family of constructions from algebraic number theory that beat the grid by a polynomial factor — later sharpened by Will Sawin at Princeton to n^1.014.

    What makes it count is that humans checked it. A digested, human-verified version was published within days, with reflections from Noga Alon and Tim Gowers among others. Gowers called it a milestone; Gil Kalai set it beside Appel and Haken’s computer-assisted proof of the four-colour theorem in 1976 — the last time mathematics had to decide what it would accept as a proof it could not read in one sitting.

    It did not come from nowhere. AlphaTensor found a faster matrix multiplication in 2022; FunSearch improved a bound on the cap set problem in 2023, the first time a language model moved an open problem; AlphaEvolve beat Strassen’s 1969 algorithm for 4×4 complex matrices in 2025, after fifty-six years. Those were searches over a space. This is closer to an argument.

    What the nine of them actually checked matters. The paper is a digest of edited reasoning, not of the raw output — which is the honest limit of this dot, and the subject of an argument still running about whether a result you can verify but cannot follow is understanding or only automation.

    And here is what does not count, so that the criterion is legible rather than implied. August 2026 produced two results that look like this dot and are not. Moderna and Merck’s individualised mRNA cancer vaccine cleared Phase 3 on 19 August, with a model picking up to 34 neoantigens from each patient’s own tumour — medically enormous, and AI as a selector inside a pipeline people designed. Generalist’s GEN-1.5 learned robot tasks from a single short demonstration. That is a machine learning better. Learning and creating are not the same act, which is the entire line this dot stands on.

    Held loosely, and here is the reason to: several 2026 claims of this kind have not survived contact with referees, and at least one turned out to be a rediscovery of something already in the literature. The dot marks a result that named mathematicians have gone through by hand. If the standard slips, this is the dot to revisit.

    Erdős, 1946 · disproof announced 20 May 2026 · “Remarks on the disproof of the unit distance conjecture”, arXiv:2605.20695 — Alon, Gowers et al. · Will Sawin, δ = 0.014 · precursors: AlphaTensor, Nature 2022 · FunSearch, Nature 2023 · AlphaEvolve, 2025 · on the limits: “Automation Without Understanding”, arXiv:2607.06377 · what does not count: Moderna/Merck intismeran autogene Phase 3, 19 Aug 2026 · Generalist GEN-1.5, one-shot robot learning

32. ↓ 0 yr later — follows from it

    2026 — now

    You are here

    2026 — the present. The thread stops here: nothing guarantees the next gate opens.

Our futures

Our timeline

A future track continues from the shared past above; only what it adds is listed here.

1. → 2026

   Next Starship launch

   waypointtoward: First permanent Moon basedate updates live in the explorerEstimate only

   Another integrated flight test.

   The long version

   The next integrated flight test of Starship — SpaceX’s fully-reusable Super Heavy booster + Starship upper stage, the largest and most powerful rocket ever flown.

   • Planned as Starship’s first fully orbital flight
   • A first attempt to catch the Ship at the tower (“chopsticks”) — the booster has been caught before, the upper stage not yet
   • Lift-off from Starbase, Boca Chica, Texas

   Date is a NET / “TBD” placeholder and updates live from the launch manifest — expect it to move.

   estimate

2. → 2028

   Artemis IV — crew on the Moon

   waypointtoward: First permanent Moon basetarget: 1 March 2028Likely (slips)

   First humans on the Moon since 1972. (Artemis III was rescoped in 2026 to a low-Earth-orbit lander test.)

   NASA

3. → 2029

   First Starship to Mars (uncrewed)

   waypointtoward: First human landing on Marstarget: 1 November 2028Uncertain

   A cargo lander in a Mars transfer window.

   SpaceX

4. → 2030

   AGI

   thresholdwindow: 2030 – 2040Contested

   A symbolic marker — there is no consensus on the date.

   Expert forecasts

5. → 2033

   Commercial fusion power

   thresholdwindow: 2033 – 2045Uncertain

   Net-positive fusion energy on the grid.

   CFS / Helion

6. → 2035

   First crew departs for Mars

   waypointtoward: First human landing on Marstarget: 1 January 2035Speculative

   Humans leave Earth’s system for another world.

   NASA / SpaceX

7. → 2038

   First permanent Moon base

   thresholdwindow: 2035 – 2045Speculative

   Not a visit — a place people live. The Moon stops being a destination and becomes an address.

   NASA Artemis / CNSA ILRS

8. → 2040

   First human landing on Mars

   thresholdwindow: 2040sSpeculative

   Boots on another planet.

   NASA / SpaceX

9. → 2045

   The Singularity?

   thresholdwindow: 2045 · KurzweilContested

   Intelligence folds in on itself — Kurzweil’s date.

   Ray Kurzweil

10. in 250 million yrs

    Pangaea Ultima

    The continents reassemble into one landmass and it is a furnace: no ocean to moderate the interior, volcanism pushing CO2 up, a brighter Sun above it. Between 8% and 25% of land stays suitable for mammals, against about 66% today.

    The long version

    The extinction and the inconvenience are not the same claim, and the paper only makes the first. Farnsworth models mammals as animals: bound by thermal physiology and, above all, by connectivity. A hyperthermal continental interior cuts the landmass into refuges that populations cannot cross, and for wild fauna that is a mass extinction with a mechanism.

    For a species with air conditioning it is arithmetic. Eight per cent of that landmass is still around twelve million square kilometres — larger than Russia. And “over 40 °C in the warmest months, with little rain” describes places tens of millions of people live in now. This dot is a wall for the biosphere and a logistics problem for a technological species, which is a distinction the dot after it does not allow.

    It also assumes high CO2 from assembly volcanism. On a 250-million-year horizon any civilisation still present is a geological actor — so read the number as the climate a species would have to answer, not as the verdict on it.

    Farnsworth et al., Climate extremes likely to drive land mammal extinction during next supercontinent assembly, Nature Geoscience, Sept 2023

11. in 600 million yrs

    Photosynthesis stops

    The first floor that is not about where you stand. The Sun brightens about 1% every 110 million years; a hotter planet weathers rock faster, and weathering buries CO2. Below roughly 150 ppm the C3 pathway fails. Not a smaller habitable area — no base to the food chain.

    The long version

    Why this is the deadline and Pangaea Ultima is not. The dot before this one reduces where you can live. This one removes what you live on, everywhere, at once.

    The mechanism has no geography in it. Solar luminosity rises about 1% every 110 million years; a warmer surface weathers silicate rock faster, and weathering is how carbon leaves the air for the sea floor. The thermostat that kept this planet habitable for four billion years runs one way at the end — CO2 falls, and below roughly 150 ppm the C3 pathway, which is most plants and every cereal, can no longer fix carbon at all. C4 plants concentrate CO2 before fixing it and hold out somewhat longer.

    Everything downstream of a plant goes with it. Free oxygen is a by-product of photosynthesis and is chemically unstable without it, so over the following hundreds of millions of years the atmosphere reverts. Past this dot the line is describing a planet, not a biosphere.

    Artificial photosynthesis and sealed atmospheres are not ruled out by any of this, which is the difference between a floor and a wall. This is a floor: you can build on it, but nobody stands on the ground any more.

    Caldeira & Kasting, Nature, 1992 · Ozaki & Reinhard, Nature Geoscience, 2021

12. in 1.3 billion yrs

    The oceans go

    The same brightening, further along. Water vapour reaches the stratosphere, hydrogen escapes to space, and Earth leaves the habitable zone from the inside.

    Leconte et al., Nature, 2013 · Wolf & Toon, 2015 — estimates 1–1.5 Gyr, sensitive to cloud feedbacks

13. in 4.5 billion yrs

    Andromeda arrives

    The Local Group’s two spirals merge into one elliptical, and the sky is rebuilt before the Sun dies. Long called a certainty; a 2025 reanalysis makes it closer to a coin flip.

    van der Marel et al., 2012 · Sawala et al., Nature Astronomy, 2025 — ~50% within 10 Gyr

14. in 5.0 billion yrs

    The Sun leaves the main sequence

    Core hydrogen runs out, the Sun swells, and by around 7.6 billion years from now the Earth is inside it. The address stops existing.

    Schröder & Smith, MNRAS, 2008 — engulfment at ~7.59 Gyr

15. in 100 billion yrs

    The end of cosmology

    Expansion carries every galaxy outside the Local Group past the horizon, and the observable universe becomes one island in an apparently static void. An observer there, with perfect instruments, cannot find the expansion, the background, or the Big Bang. The mirror of Darwin — the point at which this chart becomes unthinkable again, and we are inside the window.

    The long version

    This is the dot on the chart that should be the most unsettling, and it is not about us at all. Everything we know about the origin and the shape of the universe rests on two observations: that distant galaxies recede, and that the sky glows faintly at 2.7 kelvin. Both are temporary. Neither is available to anyone living after the horizon closes.

    Those observers are not primitive. They can have better instruments than ours, more history, more mathematics. The evidence is simply gone — redshifted past detection and beyond a horizon that recedes faster than light can cross it. Their best cosmology, done honestly, describes a single eternal island of stars in an empty static space. It is wrong, and there is nothing available to them that would say so.

    Which puts a date on something this whole axis assumes: that the past is knowable. It is knowable from here. That is a property of when we are standing, like the retrograde loops under Copernicus — and it is the strongest argument on this chart for the value of looking now.

    Loeb, 2002 · Krauss & Scherrer, The return of a static universe and the end of cosmology, 2007

16. in 100 trillion yrs

    The last stars

    Star formation ends — the gas left is too thin and too hot to collapse. The last red dwarfs burn on a while, then the universe goes dark for good. Decade +14 of 100: the part of the future that can hold anything is its first seventh.

    Adams & Laughlin, A dying universe: the long-term fate and evolution of astrophysical objects, Rev. Mod. Phys., 1997

17. in 10¹⁵ yrs

    Planets go free

    Nothing accelerates here — the wait simply runs out. A star passing close enough to unbind a planetary system is very rare, and at this galaxy’s density the expected wait for one is about this long: a hundred thousand times the present age of the universe. Dead stars pass dead stars, planets are stripped from them, and “solar system” stops being a kind of object.

    Adams & Laughlin, 1997 — cosmological decade η≈15, the encounter timescale at galactic disk density

18. in 10²³ yrs

    Galaxies dissolve

    The same encounters, given longer: most remnants are flung into intergalactic space and the rest spiral into the central black hole. Everything on this chart happened inside a galaxy.

    Adams & Laughlin, 1997 — η≈19–24

19. in 10³⁷ yrs

    If the proton decays

    Then the remnants go early: black dwarfs and neutron stars evaporate into radiation and no baryon is left anywhere. This is an early exit, not the deadline — that is two dots along, and it does not need this to be true. Predicted by grand unified theories, never observed, and every year the detectors run without a signal moves this dot right.

    Super-Kamiokande, 0.37 Mton·yr exposure, Phys. Rev. D 106, 072003 (2022) — bound past 10³⁴ yr · Hyper-Kamiokande heading for ~10³⁵ · Adams & Laughlin, 1997 — η≈34–39

20. in 10⁶⁷ yrs

    Black holes evaporate

    Hawking radiation takes the smallest first, and the smaller they get the faster they go. By here only the giants at the galactic centres are left.

    Hawking, 1974 · Adams & Laughlin, 1997 — solar-mass evaporation at η≈65–67

21. in 10⁷⁸ yrs

    The last matter

    The ceiling on matter, and it holds whether or not the proton decays. Pair production follows from the curvature of spacetime rather than from an event horizon, so a white dwarf radiates too, and ends in an explosive instability. Two clocks run on the end of matter: the earlier one is a maybe, this one is the deadline.

    The long version

    Two clocks run on the end of matter, and until 2025 only one of them was on the board. The first is proton decay: predicted, never seen, and if it turns out not to happen the whole mechanism is absent. The old fallback in that case was quantum tunnelling — iron stars assembling by pycnonuclear reaction, black holes forming by tunnelling — on timescales around 10¹¹⁰⁰ years. Numbers that large are not on this axis and could not be: the ruler would need eleven times its current length.

    The second clock does not need the proton to be unstable. Hawking’s calculation is usually told as a property of event horizons, but the pair production it describes follows from the curvature of spacetime, which anything with mass has. So a white dwarf radiates — unimaginably slowly, and then, as it loses mass and destabilises, not slowly at all.

    Which is why this dot is where the axis can stop. Without it the honest right-hand end of this chart is 10¹¹⁰⁰ years and the log axis breaks. One paper in 2025 made the future drawable.

    Falcke, Wondrak & van Suijlekom, An upper limit to the lifetime of stellar remnants from gravitational pair production, JCAP 05 (2025) 023, arXiv:2410.14734

22. in 10¹⁰⁰ yrs

    Heat death

    Maximum entropy. The supermassive black holes at the centres of dead galaxies are the last structures in the universe, and the largest take around 10¹⁰⁰ years to radiate away. After that there is no gradient left anywhere — and a gradient is what every dot on this line was made of.

    The long version

    Eighty-six decades separate the last star from this dot, and everything in them is something ending. The whole distance from the Big Bang to now is ten. The future is not long compared to the past; it is long compared to everything, and almost none of it contains anything.

    Read the axis as a proportion and the shape is this: everything — stars, chemistry, planets, life, minds, and whatever follows minds — happens in a narrow band at the far left of the future, and then there is dark for a length of time that makes the entire history preceding it a rounding error. We are not early in the universe. We are early in the interesting universe, which is a much smaller thing and nearly over.

    And the thread stops well before here, deliberately. Nothing on this chart guarantees the next step; a line drawn to maximum entropy would be claiming to know that there is nothing to do about any of it.

    Adams & Laughlin, 1997 · proton decay unobserved, lower bound >10³⁴ yr (Super-Kamiokande)

Looking back

Forecasts

A future track continues from the shared past above; only what it adds is listed here.

1. 1969

   Verne — men to the Moon

   happenedsaid in 1865

   Three men, a capsule, launched from Florida. Apollo 11 went in 1969, from Florida. He was looking 104 years ahead.

   Jules Verne, 1865

2. 1985

   The population bomb

   never arrivedsaid in 1968

   “In the 1970s hundreds of millions of people will starve to death.” The Green Revolution overtook it.

   Ehrlich, 1968

3. 1997

   Terminator — Skynet awakes

   never arrived

   Skynet becomes self-aware in August 1997. It didn’t.

   The Terminator, 1984

4. 1999

   Space: 1999 — Moonbase Alpha

   never arrived

   A permanently crewed Moon base by 1999.

   1975

5. 2000

   Y2K — the millennium bug

   never arrivedsaid in 1997

   A civilization-ending glitch at midnight 2000. Fixed quietly, at great expense; nothing happened. The shortest arrow here — and still wrong.

   forecast through the 1990s

6. 2001

   2001: A Space Odyssey

   never arrived

   Orbital hotels, Moon colonies, and HAL 9000 by 2001.

   Kubrick / Clarke, 1968

7. 2007

   Tesla — the phone in your pocket

   happenedsaid in 1926

   “A man will be able to carry one in his vest pocket.” The iPhone arrived in 2007 — 81 years later.

   Nikola Tesla, 1926

8. 2015

   Back to the Future II

   never arrived

   Hoverboards, flying cars, self-lacing shoes by 2015.

   1989

9. 2015

   The end of death from cancer

   deadline passedsaid in 2003

   In February 2003 the director of the US National Cancer Institute set a formal target: not to cure cancer, but to “eliminate the suffering and death due to cancer” by 2015. US cancer mortality has since fallen by roughly a third from its 1991 peak — real progress, and the deadline passed anyway.

   NCI, Andrew von Eschenbach, Feb 2003

10. 2019

    Blade Runner

    never arrived

    Bioengineered replicants and off-world colonies by 2019.

    1982

11. 2026 — now

    Metropolis — we are here

    never arrived

    Fritz Lang set his stratified mega-city and its machine-being in 2026. The city rhymes; the robot doesn’t — yet.

    1927

Themes

Climate

A future track continues from the shared past above; only what it adds is listed here.

The planet’s own history — the second band, drawn below the axis on the explorer, running underneath the shared past above rather than following from it

1. 4.0 billion yrs ago

   The faint young Sun

   The Sun was 20 to 30% dimmer than today and the ocean was liquid anyway. Nothing accounts for that except a far thicker greenhouse — CO₂ and methane at concentrations that would be unbreathable now. The greenhouse effect is not a hypothesis about the future: it is the reason there was ever an ocean to put life in.

   Sagan & Mullen, Science, 1972 — the faint young Sun paradox

2. ↓ 1.6 Gyr later — follows from it

   2.4 billion yrs ago

   The first ice age

   Directly above, on the trunk, the mineral sinks have just filled and oxygen starts accumulating. It destroys atmospheric methane — the Archean’s strongest greenhouse gas — and the planet freezes. The first glaciation in the record is a side effect of the first great biological invention, and the only dot on this band caused by life rather than by rock, orbit or Sun.

   the Huronian glaciation, ~2.4–2.1 Gyr

3. ↓ 1.7 Gyr later — follows from it

   717 million yrs ago

   Snowball Earth

   Ice to the equator, twice. The Sturtian runs 717 to 661 Myr — fifty-six million years of it — and the Marinoan follows, now dated to about four. What ends them is the same knob turned the other way: with the rock weathering that removes CO₂ shut down under the ice, volcanoes keep loading the atmosphere until the albedo loses.

   Sturtian 717–661 Myr · Marinoan ~650–635 Myr, duration ~4 Myr — PNAS, 2025

4. ↓ 272 Myr later — follows from it

   445 million yrs ago

   Hirnantian ice

   The end of the Ordovician, and the first of the five great extinctions. The glaciation arrives on a world whose CO₂ had been falling — how far, and why, is still argued — and it is on this band because it is the first time a climate swing is the cause of an extinction rather than a footnote to one.

5. ↓ 145 Myr later — follows from it

   300 million yrs ago

   The coal forests

   The coldest the Phanerozoic ever gets: about 12 °C global mean, nearly two degrees below pre-industrial. The Carboniferous forests bury carbon faster than anything can rot it, CO₂ falls, ice reaches the tropics. Three hundred million years later we dig that carbon up and burn it — the Late Palaeozoic Ice Age *is* the coal.

   PhanDA — Judd et al., Science, 2024

6. ↓ 48 Myr later — follows from it

   252 million yrs ago

   The Great Dying

   The Siberian Traps load the atmosphere with carbon on a scale nothing since has matched and the global mean rises about 10 °C. Ninety per cent of marine species go. This is the one dot here that is unambiguously a carbon-release extinction, which is why the mechanism is not in doubt anywhere on this band — only the rate ever differs.

   end-Permian, 252 Myr — the Siberian Traps

7. ↓ 157 Myr later — follows from it

   95.0 million yrs ago

   Cretaceous hothouse

   About 34 °C global mean — twenty degrees above pre-industrial — no permanent ice anywhere, and shallow seas across the continents: the Western Interior Seaway cut North America in two. The dinosaurs lived in it for a hundred million years, which settles one thing and must not be allowed to settle another. Hot is survivable for a biosphere with millions of years to arrive at it. But the high sea is not a preview of ours: the estimates run from about +100 m to +240 m, all the ice on Earth is worth only 65, and the difference is ocean-basin volume — young hot crust and fast spreading ridges, on a tectonic clock we do not have.

   PhanDA — Judd et al., Science, 2024 · Haq, 2014 (+240 m) against Miller & Kominz (~+100 m) · total ice ≈ 65 m: Antarctica 57.9, Greenland 7.4

8. ↓ 39 Myr later — follows from it

   56.0 million yrs ago

   The PETM

   The closest natural analogue there is, and the comparison is the entire point of having it. Five to seven degrees of warming, thousands of gigatonnes of carbon, an acidified ocean, whole biotas on the move — and the carbon went in over something between one and twenty thousand years. Present emission rates are roughly ten times the PETM onset. On the current trend we add a PETM’s worth of carbon in one to three centuries.

   Gingerich, Paleoceanography and Paleoclimatology, 2019 — modern rates ~9–10× the PETM onset

9. ↓ 22 Myr later — follows from it

   34.0 million yrs ago

   Antarctica freezes

   CO₂ falls past a threshold at the Eocene–Oligocene boundary, a permanent cap forms, and the planet stops being a hothouse. Everything after this dot happens in an icehouse world, us included. The ice sheets that fix every coastline we have are 34 million years old — the youngest thing on this band, and the thing our cities are arranged around.

10. ↓ 31 Myr later — follows from it

    2.6 million yrs ago

    The glacial oscillator

    Northern ice sheets begin advancing and retreating on orbital beats — 41,000 years, then 100,000. This is the climate every ancestor of ours evolved in, and the trunk dot almost directly above says what leaving it was worth: the oscillator switching off is what makes farming physically possible.

11. ↓ 2.6 Myr later — follows from it

    3974 BCE

    The mid-Holocene question

    The one number on this band that nobody can give you. Proxy syntheses put a global peak a few tenths of a degree above pre-industrial around here and a long cooling after it; a data-assimilation reconstruction finds no global peak at all, and has the world 0.17 °C *cooler* at 6,000 years than just before industrialisation. The suspicion is that the proxies record Northern Hemisphere summers rather than the globe — which is also why the warmth was real where it was measured, and why it is not the comparison people reach for it to be. Today is above both answers.

    Marcott et al., 2013 · Kaufman et al., Scientific Data, 2020 · Osman et al., Nature, 2021 · Bova et al., Nature, 2021 · Nature, 2022 — Revisiting the Holocene global temperature conundrum

12. ↓ 6 kyr later — follows from it

    1650

    The Little Ice Age

    Two tenths of a degree, globally, and it froze the Thames. It is on this band as a calibration: a swing small enough to sit inside the uncertainty of every curve here reorganised European agriculture for three centuries. Nobody lives in the global mean.

13. ↓ 200 yr later — follows from it

    1850

    The baseline

    1850–1900, and about 278 ppm. Everything on this chart is measured from here, and it is worth saying that this is a choice rather than a fact: not the Holocene average, but the earliest window with enough thermometers in enough places. It also sits at the cold end of the last two thousand years — the Little Ice Age has barely finished.

14. ↓ 108 yr later — follows from it

    1958

    The Keeling curve

    March 1958, Mauna Loa. Not the discovery — Arrhenius had the physics in 1896 and Callendar the trend in 1938 — but the first time anyone could watch the atmosphere change, month by month, with the seasonal breath of the northern forests visible inside the annual rise. Everything on this band after this dot is measurement instead of reconstruction.

    Charles David Keeling, Mauna Loa, 1958 · Arrhenius, 1896 · Callendar, 1938

15. ↓ 66 yr later — follows from it

    2024

    Past 1.5 °C

    2024 is the first calendar year above 1.5 °C, and 2023–2025 the first three-year period whose average is above it. The Paris figure was never a single year — it is a long-term mean — so this is not the limit being breached. It is the limit coming into view about a decade earlier than the agreement that named it had assumed.

    WMO — Jan–Aug 2025 at +1.42 ± 0.12 °C; 2015–2025 the eleven warmest years in the 176-year record

16. ↓ 76 yr later — follows from it

    → 2100

    2100 — current policies

    predictedwindow: +2.1 to +3.4 °CCurrent policies

    Central estimate +2.6 °C, and it has not moved across four consecutive assessments. Sea level 0.6 to 0.9 m above pre-industrial. Read against the band it sits on: warmer than any point in the last three million years, and cooler than every dot here older than thirty-four million. Both halves of that are true, and only one of them is reassuring.

    Climate Action Tracker, 2025 update · IPCC AR6 SSP2-4.5 (0.44–0.76 m on a 1995–2014 baseline)

17. ↓ 200 yr later — follows from it

    → 2300

    The sea keeps rising

    predictedwindow: by 2300 · 0.3 – 3.1 mCommitted

    Temperature can stop rising within decades of emissions stopping. Sea level cannot: the ocean is still taking up heat and the ice sheets answer over centuries. This is the one place the two metrics come apart, and the reason the band does not end in 2100 — the last dot of the human line above it is nine years old, and this one is still moving in the year 2300.

    IPCC AR6 WG1, Chapter 9 — SSP2-4.5 to 2300

Themes

Population & energy

1. 11k yrs ago

   Four million of us

   The entire species, worldwide, at about the population of one large city today — and the only energy source is fire. Every dot after this one is the same two quantities going up together, until the last few, where they come apart.

   estimates for global population before agriculture

2. ↓ 8 kyr later — follows from it

   0

   A city of a million

   Imperial Rome reaches something like a million people, which is not matched anywhere for seventeen centuries. A city that size is an energy problem before it is anything else: grain fleets, aqueducts, and a hinterland stripped of firewood.

3. ↓ 2 kyr later — nothing guaranteed it

   1712

   The steam engine

   The first energy that is not muscle, wind, water or wood. It is a gate because nothing before it lifted the ceiling: for ten thousand years the power available per person barely moved, and the whole of the graph above is flat until here.

   Newcomen, 1712 · Watt’s separate condenser, 1769

4. ↓ 88 yr later — follows from it

   1800

   One billion

   Around 1800, and it took the whole of prehistory and history to get here. The next billion takes 127 years; the one after that, 33.

5. ↓ 108 yr later — a constraint was lifted

   1908

   Haber–Bosch

   new substrate for information: nitrogen

   Nitrogen out of the air, at industrial scale — the constraint that had capped every agriculture in history lifted in one process. About half the nitrogen in the body of anyone reading this came out of it, and roughly half the people alive are fed by it. It is also, on its own, about 1–2% of world energy: the release costs power, which is the whole subject of this track.

   Fritz Haber, 1909 · Carl Bosch, BASF, 1913 · Smil, Enriching the Earth, 2001

6. ↓ 19 yr later — follows from it

   1927

   Two billion

   1927. From here the doublings arrive inside a single lifetime, and so does the energy: the total has never once fallen, in any year, for any source on the chart.

7. ↓ 43 yr later — follows from it

   1970

   The Green Revolution

   Dwarf wheat and rice, fertiliser, irrigation: yields per hectare double and the famines that had been forecast for the 1970s do not arrive — the prediction on another track that this dot falsifies. It is not free. It is Haber–Bosch, diesel and water, which is to say it is energy converted into food.

   Norman Borlaug, Nobel Peace Prize, 1970

8. ↓ 52 yr later — follows from it

   2022

   Eight billion

   November 2022. Twenty-one terawatts of power — energy per unit time, which is the unit the Kardashev scale is written in — and over 80% of it fossil — and on Sagan’s continuous reading of Kardashev’s scale that puts the species at about 0.73.

   UN, 8 billion on 15 November 2022 · Energy Institute, 2025 — ≈186,000 TWh in 2024

9. ↓ 63 yr later — follows from it

   → 2085

   Peak population

   thresholdwindow: mid-2080sLikely

   The UN’s 2024 revision has the curve turning over at about 10.3 billion in the mid-2080s and falling after it. Sixty-three countries had already peaked by 2024. It would be the first fall in this line with no catastrophe underneath it — every previous one had a plague or a famine under it.

   UN World Population Prospects, 2024 revision

10. → 2427

    Kardashev I — and the wall

    thresholdwindow: ≈2430 at 2.3%/yr · ≈2155 super-expPhysics, not forecast

    Two things that turn out to be one number, about 2×10¹⁷ W. It is Type I as the scale is now used — all the stellar energy arriving at the planet — and it is also the thermal ceiling, because using as much power as Earth receives means radiating as much as Earth receives, and the surface has to get hot enough to do it. No efficiency touches that and neither does fusion: fusion changes the fuel and the carbon, not the heat. At the historical 2.3% a year we arrive in about four centuries; on a super-exponential path — the growth rate itself climbing towards 10% — in about 130 years. Faster growth does not clear the wall, it reaches it sooner — so everything above this line has to happen somewhere that is not a planetary surface. And this is not how Kardashev defined Type I: in 1964 he put it at 4×10¹² W, “close to the level presently attained on the Earth”. We passed his figure before 1970.

    Tom Murphy, Galactic-Scale Energy, 2011 · Kardashev, 1964 · the modern figure ~2×10¹⁷ W

11. → 3369

    Kardashev II — the star

    thresholdwindow: ~1,300 yrs at 2.3%/yrSpeculative

    The Sun’s entire output, 4×10²⁶ W. About thirteen hundred years at the historical rate — and on this site’s own axis that is nothing at all: the Sun has five billion years left and photosynthesis has six hundred million. If a civilisation keeps growing at all, it arrives at its star almost immediately in astronomical terms. The catch is the dot above: this is nine orders of magnitude past the point where a planetary surface cooks, so a Type II is not a bigger us — it is something that does not live on a planet.

    Kardashev, 1964 — Type II at 4×10²⁶ W

12. → 4483

    Kardashev III — the galaxy

    thresholdwindow: ~2,500 yrs at 2.3%/yrSpeculative

    Four times ten to the thirty-seventh watts. Kardashev described three types and stopped; Type IV, the observable universe at 10⁴⁵–10⁴⁶ W, is a later addition by other authors and is not on this line. The scale was never a ranking of civilisations either — the 1964 paper is about how much power you could put into a radio beacon. Everyone else made it a league table.

    Kardashev, Transmission of Information by Extraterrestrial Civilizations, 1964

Themes

The Economy

1. 1150

   Capital, in Genoa

   The word arrives before the concept it is now made to carry. In twelfth-century Genoa *capitale* is the principal of a loan or the stake in a venture — a sum at risk in a named voyage, with a named counterparty, for a named return. It is not yet a factor of production, and it is not yet a stock of anything. Everything the term later has to do is added by people who never met a Genoese merchant.

   Mario Fabbri, La fabbrica delle illusioni, Rubbettino — which traces the idea from here

2. ↓ 616 yr later — follows from it

   1766

   Turgot

   Anne Robert Jacques Turgot

   The step that makes modern economics possible and, on one reading, the mistake at its root: development comes from profit-capital generated by saving. Adam Smith takes it up, and from there it is furniture. Fabbri’s book is an argument that this is where the false move enters — that the idea of capital as a self-standing productive substance was built on a merchant’s bookkeeping term and then asked to carry a physics. That reading is contested and worth marking as a reading; what is not contested is that the concept dates from here rather than from the thing it describes.

   Turgot, Réflexions sur la formation et la distribution des richesses, 1766 · Adam Smith, 1776

3. ↓ 168 yr later — nothing guaranteed it

   1934

   GDP is invented

   Simon Kuznets

   Before 1934 there is no number. Kuznets builds the first national accounts for the US Senate in the middle of the Depression, because nobody could say how bad it was; Keynes and Stone then rebuild them in Britain to work out how much war the country could afford. This is a gate and not a chain: nothing about economic theory produced it, an emergency did, and it was designed to answer a question about wartime capacity rather than about welfare. Kuznets himself spent the rest of his life warning against the use it was immediately put to.

   Simon Kuznets, National Income 1929–32, US Senate, 1934 · Keynes & Stone, UK, 1941

4. ↓ 10 yr later — follows from it

   1944

   Bretton Woods

   The number is three years old and it is already the unit of international account: quotas, aid and adjustment all get denominated in it. Fixed rates against a dollar fixed to gold, so a government’s room to borrow is bounded by a metal.

5. ↓ 27 yr later — a constraint was lifted

   1971

   The Nixon shock

   15 August 1971: the dollar comes off gold and the constraint goes. It is a `release` in this chart’s vocabulary — nothing new is created, a limit is removed, and what was waiting arrives at once. Every dot after this one is about what a state can do when the amount it may owe is a political question rather than a physical one.

   15 August 1971

6. ↓ 2 yr later — nothing guaranteed it

   1973

   The oil shock

   The date every curve on this site turns. Power per head stops growing, the machine-level efficiency gains had already run out in the 1960s, and from here the residual does the work. Whether the deceleration is the oil price, the end of post-war catch-up, or the exhaustion of the conversion gains is not settled — the coincidence of the three is why 1973 keeps appearing.

   OPEC embargo, October 1973

7. ↓ 6 yr later — follows from it

   1979

   Volcker

   Rates to twenty per cent to break inflation, and a forty-year decline in the cost of borrowing begins on the other side of it. For most of what follows the real rate sits below the growth rate, which is the arithmetic that lets a state outgrow its debts instead of repaying them. That arithmetic has recently inverted, and the `close the deficit` view is what it looks like when it does.

   Paul Volcker, Federal Reserve, 1979–81

8. ↓ 29 yr later — follows from it

   2008

   2008

   Gross world government debt is 59% of GDP the year before. The response makes the deficit a permanent instrument rather than a cyclical one, and the level never returns.

9. ↓ 12 yr later — follows from it

   2020

   The deficit step

   2020: the world deficit reaches 9.5% of GDP in a single year and gross debt goes to 97%. It settles at five per cent and ninety-two — which is to say the step down never finished. This is the plinth the `1970 break` view draws.

   IMF — world general government balance −9.5% in 2020, −5.0% in 2023

10. ↓ 6 yr later — follows from it

    2026 — now

    You are here

    Unresolved

    2026. World gross government debt around 92% of GDP, the world deficit around 5%, real growth around 3% — of which, on the identity, roughly two thirds comes from a residual nobody can audit physically. Whether that is productivity, composition, mismeasurement or financing is the open question, and the metric view on this track is an attempt to bound it rather than to answer it.

Themes

Medicine & longevity

1. 1796

   The first vaccine

   happened

   Jenner inoculates against smallpox — the birth of preventive medicine.

   Edward Jenner, 1796

2. 1846

   Anaesthesia

   happened

   Ether ends the agony of surgery; operations become survivable, then routine.

   William Morton, Boston, 1846 (Crawford Long, 1842, unpublished)

3. 1865

   Germ theory & antisepsis

   happened

   Pasteur and Lister: disease has a cause you can wash away. Infection stops being fate.

   Pasteur / Lister, 1860s

4. 1921

   Insulin

   happened

   Diabetes goes from a death sentence to a managed condition.

   Banting, Best, Collip & Macleod, Toronto, 1921

5. 1928

   Penicillin — the antibiotic era

   happened

   Fleming’s mould: for the first time we beat bacterial infection, and life expectancy leaps. It took Florey and Chain another twelve years to turn it into a drug.

   Alexander Fleming, 1928 · Florey & Chain, 1940s

6. 1953

   The double helix

   happened

   The structure of DNA — read off Franklin’s Photo 51.

   Watson, Crick, Wilkins & Rosalind Franklin, 1953

7. 1978

   First IVF baby

   happened

   Louise Brown — life begun outside the body.

   Steptoe & Edwards, 1978

8. 1980

   Smallpox eradicated

   happened

   The only human disease ever wiped from the Earth.

   WHO — D. A. Henderson, 1980

9. 1996

   Dolly, the cloned sheep

   happened

   An adult mammal copied from a single cell.

   Wilmut & Campbell, Roslin Institute, 1996

10. 2003

    The human genome

    happened

    The first full read of our own code.

    Human Genome Project (Collins) · Celera (Venter), 2003

11. 2012

    CRISPR-Cas9

    happened

    Editing the code, cheaply and precisely.

    Jennifer Doudna & Emmanuelle Charpentier, 2012

12. 2022

    A pig heart in a human

    happened

    David Bennett received the first gene-edited pig heart in January 2022 and lived 60 days. It was compassionate use, not a trial — and for three years it stayed that way, one case at a time. The longest any pig organ has lasted in a person is 130 days.

    University of Maryland, Jan 2022

13. 2023

    First CRISPR therapy approved

    happened

    Casgevy — sickle-cell disease, edited away. Eleven years from the discovery to a licensed medicine.

    Vertex / CRISPR Therapeutics, 2023

14. 2026 — now

    You are here

    2026.

15. → 2030

    Longevity escape velocity?

    predictedContested

    Life expectancy rising faster than time passes.

    Kurzweil / de Grey

16. → 2035

    Aging as a treatable disease?

    predictedSpeculative

    Reversing biological age, not just slowing it.

    Sinclair / Altos Labs

17. → 2045

    Homo Deus — a-mortality

    predictedContested

    Indefinite life as the next human project.

    Yuval Noah Harari

Themes

One of many

1. 1543

   Copernicus

   Nicolaus Copernicus

   The Earth is not the centre of anything. It is the first demotion, and it took a century to be believed — the book came out in 1543, the year he died, and the argument was mathematical rather than observed: nothing anyone could see through in 1543 settled it.

   The long version

   Why this is the first one and not an earlier one. Aristarchus of Samos put the Sun at the centre in the third century BCE and it went nowhere — no instrument could test it and no network carried it. The same shape as the positional zero on the trunk: the invention is the cheap half.

   What Copernicus actually offered was not proof, it was economy. His system still used circles and still needed epicycles; it was not obviously more accurate. What it was, was simpler about one specific thing — the retrograde loops of the planets stop needing an explanation and become a side effect of where the observer is standing. The argument is that the strangeness was never in the sky. It was in the seat.

   That is the move every dot on this track repeats: something we read as a property of the world turns out to be a property of our position in it.

   De revolutionibus orbium coelestium, 1543

2. ↓ 67 yr later — follows from it

   1610

   Galileo’s moons

   Galileo Galilei

   The first demotion anybody could look at. Four moons going round Jupiter means there is a centre of motion that is not the Earth — not an argument, a thing in an eyepiece. In the same months the Milky Way resolves into individual stars and the Moon turns out to have mountains: the heavens stop being a different kind of place from here.

   Sidereus Nuncius, March 1610 — the Jovian moons observed from 7 January 1610

3. ↓ 178 yr later — follows from it

   1788

   Deep time

   James Hutton · Charles Lyell

   We are not most of the story, or much of it. Hutton reads the rock as a process with no visible beginning — “no vestige of a beginning, no prospect of an end” — and the Earth goes from thousands of years old to unbounded. This is the demotion that makes the next one possible: natural selection needs time it did not have before 1788.

   James Hutton, Theory of the Earth, read 1785, published 1788 · Charles Lyell, Principles of Geology, 1830–33

4. ↓ 71 yr later — follows from it

   1859

   Darwin

   Charles Darwin · Alfred Russel Wallace

   Not designed — a branch. And not the tip of the branch either: there is no tip. This is the demotion that lands hardest because it is the one about us specifically, and the only one on this track that also earns a dot on the trunk, where it sits off the causal thread for exactly the reason it belongs here: it changed what we could see, not what we could do.

   Darwin & Wallace, Linnean Society, 1 July 1858 · On the Origin of Species, 24 November 1859

5. ↓ 41 yr later — follows from it

   1900

   The unconscious

   Sigmund Freud

   Not even in charge of ourselves. Freud’s machinery — the drives, the topography, most of the clinical apparatus — largely did not survive as science, and saying so is part of the dot. What survived is the claim underneath it: most of what a mind does is not available to the mind doing it. Cognitive science spent the following century confirming that and rebuilding the explanation from scratch.

   The long version

   This is the shakiest dot on the track and it is here on purpose. Freud is not a reliable scientist and the track would be more defensible without him. But the demotion he named is real, and it was named by him first: introspection is not a window, it is a press release.

   What replaced the theory keeps the finding. Split-brain patients confabulate reasons for choices made by a hemisphere that cannot speak. People asked to explain a preference produce fluent accounts of causes that experiments show were not operating. The mind reports on itself and the report is a construction.

   Which is the reason this track ends where it does. “The machine is only predicting the next token, it has no access to what it is doing” is an argument that stops being decisive the moment you apply it symmetrically.

   Die Traumdeutung, 1899 (dated 1900) · the mechanism contested; the inaccessibility of most mental process is not

6. ↓ 24 yr later — follows from it

   1924

   A galaxy among billions

   Edwin Hubble

   The Milky Way is not the universe. Hubble finds Cepheid variables in Andromeda, measures the distance, and it is far outside our own system — those smudges are other galaxies. Within a decade the count is in the millions, and now it is hundreds of billions. Copernicus moved us off the centre of the solar system; this moves the solar system off the map.

   Cepheids in M31, 1923–24, announced January 1925 · the Shapley–Curtis debate, 1920

7. ↓ 7 yr later — follows from it

   1931

   Gödel

   Kurt Gödel

   Even the reasoning is not complete. Any formal system strong enough to do arithmetic contains true statements it cannot prove — so mathematics, the one thing that looked like it could be finished, cannot be. The demotion here is not of us against nature but of our best instrument against itself.

   Über formal unentscheidbare Sätze der Principia Mathematica und verwandter Systeme I, 1931 · Hilbert’s programme, 1920s

8. ↓ 22 yr later — follows from it

   1953

   The double helix

   Not a different kind of stuff. The code is chemistry, and it is the same chemistry in a bacterium, a redwood and a person — four bases, the same triplet dictionary, read the same way everywhere. Life turns out to have one implementation and we are a configuration of it.

   Watson, Crick, Wilkins & Rosalind Franklin, 1953 — read off Franklin’s Photo 51

9. ↓ 7 yr later — follows from it

   1960

   Tools in the wild

   Jane Goodall

   Not designed, not central, not special — and not the only toolmaker either. Goodall watches a chimpanzee at Gombe strip a twig and fish for termites, and Louis Leakey answers with the sentence that gives this whole track its method: “Now we must redefine tool, redefine Man, or accept chimpanzees as human.” Man the Toolmaker had been the working definition for a century. It did not die: it retreated, to language, to art, to invention. Every dot after this one is that retreat continuing, and the demoting is no longer done by astronomy — it is done by looking harder at other animals, and later by building something.

   Jane Goodall, Gombe, 1960 · Louis Leakey, in correspondence · Gallup, the mirror test, 1970 · Weir & Kacelnik, tool manufacture in New Caledonian crows, 2002

10. ↓ 35 yr later — follows from it

    1995

    Planets everywhere

    Michel Mayor · Didier Queloz

    Not a special address. The first planet round a Sun-like star turns up in 1995, and it is a hot Jupiter in a four-day orbit — nothing our own system would have predicted. Thirty years later the count is in the thousands and planets are the rule, not the exception. Our arrangement is one draw, and not a distinguished one.

    51 Pegasi b — Mayor & Queloz, Nature, October 1995, Nobel 2019 · pulsar planets, Wolszczan & Frail, 1992

11. ↓ 15 yr later — follows from it

    2010

    We are a mosaic

    Svante Pääbo

    Not even a clean lineage. The draft Neanderthal genome shows that between 1% and 4% of the DNA of everyone outside Africa came from another species of human — and Oceanian populations carry Denisovan on top of it. “We” is not a branch that stayed separate; it is a confluence, and the other species are not entirely gone, they are partly us.

    Green et al., A Draft Sequence of the Neandertal Genome, Science, May 2010 · Svante Pääbo, Nobel 2022

12. ↓ 12 yr later — follows from it

    2022

    Language

    The first fallback position after the tools went, and it held for sixty years: whatever else an animal does, reasoning carried in language is ours. On 30 November 2022 it stopped holding in public. The trunk carries this same date for a different reason — there it is the week the capability reached everyone; here it is the week the category stopped being ours, and the two are not the same claim.

    ChatGPT, 30 November 2022 — see the trunk

13. ↓ 2 yr later — follows from it

    2024

    Art

    Images first, in 2022, and within two years a generative model was an ordinary step in an ordinary production pipeline. Then music, then video. Three defences in about thirty months, and the interesting part is that the objection was rewritten each time rather than dropped: not art because derivative, then not art because unintended, then not art because there is nobody behind it. Those may all be true. None of them is a capability, which is the only currency this track deals in.

    Stable Diffusion and DALL·E 2, 2022 · text-to-music, 2024 · text-to-video, 2024–25

14. ↓ 2 yr later — follows from it

    2026 — now

    Innovation

    The last fallback, and the one that was worth defending: not doing the thing well, but making something nobody had. On 20 May 2026 a reasoning model disproved a conjecture Erdős posed in 1946. The trunk carries it as a gate, because nothing about learning guaranteed creating. Here it is the demotion that lands on the narrowest ground the category had left.

    see “AI starts innovating” on the trunk, 20 May 2026

15. ↓ 0 yr later — follows from it

    2026 — now

    What is still ours?

    Unresolved

    Not “are we the only minds” — a chimpanzee with a twig answered that in 1960, and the four dots before this one are the fallback positions falling one after another. The question actually open is what the claim even is, once every candidate capability has turned out to be an instance. What is left is not a capability. It is experience, and standing: whether there is something it is like to be the thing, and whether that obliges us to anything. Those are the two we have no test for, which is why this dot is a question and stays one.

    The long version

    The pattern of this track is not that we keep being humbled. It is that the thing we treated as a category turns out to be a quantity. The centre of the universe became a coordinate. Design became a process. Life became chemistry. Our galaxy became one of a hundred billion. Toolmaking became something a chimpanzee does with a twig. Every time, something held as a kind turned out to be an instance.

    What is different about the last four dots is who is doing the demoting. Copernicus and Hubble and Goodall all found the instance already out there; language, art and invention were demoted by something we built on purpose, in under four years, having decided in advance what to aim it at. The record of this track is unbroken, and the argument for an exception this time is the argument that was made every previous time.

    And it is still not finished, because the remaining claim was never really about capability. Experience and moral standing are not things a benchmark can settle: no result on any test tells you whether there is anyone home. That is not a gap in the evidence that more evidence closes — it is a question of a different kind, and pretending otherwise in either direction is the one move this page will not make.

    This is the dot where the site’s own subject and its timeline meet. It is the only one written as a question, and it stays a question.

    Qualcuno Risponde

Themes

Alcohol & cocktails

1. 10.0 million yrs ago

   A liver that can drink

   Ethanol is a toxin and we are built to burn it.

   The long version

   One amino-acid change in the ADH4 enzyme, in the last common ancestor of humans, chimpanzees and gorillas, made it roughly forty times better at oxidising ethanol — ten million years before anyone brewed anything. Alcohol is not a habit civilization picked up. It is a metabolic inheritance civilization later found a use for.

   The question this line exists to answer: alcohol is a poison, so how did it end up at the centre of human social life?

   Because it was there first. Apes that came down from the canopy fed on fallen fruit, and fallen fruit ferments where it lies — the ethanol plume is a reliable long-range signal that sugar is ripe and available, and it carries 7 kcal/g, more than the carbohydrate it came from. The A294V substitution in ADH4 is dated to the ancestor of the African great apes, ~10 Myr ago, and raises ethanol oxidation about fortyfold.

   Two consequences that shape everything downstream:

   • Ethanol is the only recreational drug for which we already had the enzymes. Every other one meets a body with no plan for it.
   • The dose our biology was calibrated to is the dose fermentation can reach — and yeast dies in its own waste at about 15%. That ceiling holds for ten million years, and then Aqua vitae on this line removes it.

   "Poison" is a claim about dose, not about a molecule. What makes ethanol unusual is not that it is toxic — it is that we spent ten million years learning to metabolise it and then, in eight hundred, built a delivery system that outran the adaptation.

   Carrigan et al., PNAS, 2015 · Robert Dudley, the “drunken monkey” hypothesis

2. ↓ 10.0 Myr later — follows from it

   13k yrs ago

   Beer before bread

   Stone mortars at Raqefet Cave, on Mount Carmel, carry residues of a fermented cereal drink ~13,000 years ago — Natufian hunter-gatherers, at least a thousand years before cereals were domesticated.

   The long version

   The order is the finding: the drink comes first, the crop second. Three thousand years later the limestone tubs at Göbekli Tepe hold 160 litres, at a monument raised by people with no permanent houses — communal drinking older than the village, and the feast is what recruited the labour. Whether alcohol was a reason to farm is still argued; the dates no longer rule it out.

   Liu et al., J. Archaeol. Sci. Reports, 2018 · Katz & Voigt, 1986 · Göbekli Tepe vats — Dietrich et al., Antiquity, 2012 (residues contested)

3. ↓ 4 kyr later — follows from it

   7000 BCE

   Jiahu

   The oldest chemically identified alcoholic drink anywhere: rice, honey and fruit, in pottery jars in Henan, ~7000 BCE. Nine thousand years ago, and the recipe is already a blend of three things.

   The long version

   Grape wine follows the same curve at the other end of Asia — resinated, so it would keep, at Hajji Firuz Tepe ~5400 BCE, and by ~4100 BCE Areni-1 in Armenia is a winery with a press, vats and cups: a scale that implies customers.

   McGovern et al., PNAS, 2004 · McGovern, Nature, 1996 · Areni-1, Armenia, ~4100 BCE

4. ↓ 4 kyr later — follows from it

   3200 BCE

   Beer as wages

   The earliest writing is not poetry. The Uruk tablets are accounts, and a great many of them count out rations of barley and beer.

   The long version

   Alcohol is among the first things worth measuring exactly: a wage, a ration, a unit of obligation — and the men who built the Giza pyramids were paid partly in it. Same point on the axis as Writing on the trunk, and not by coincidence.

   Uruk IV tablets, ~3200 BCE · Alulu beer receipt, ~2050 BCE · Hymn to Ninkasi, ~1800 BCE

5. ↓ 2 kyr later — follows from it

   974 BCE

   The genome answers

   Farming put cereal alcohol in front of everyone, daily — and selection responded, this time in the other direction.

   The long version

   The ADH1B His48 variant, which clears ethanol fast and makes drinking unpleasant, swept East Asian and Middle Eastern populations within the last ~7,000 years; the ALDH2 flushing variant is more recent still, and both are among the strongest signals of recent selection in the human genome. Alcohol reshaped our enzymes twice, in opposite directions. The second time is not finished.

   Peng et al., BMC Evol. Biol., 2010 · ADH1B rs1229984 · ALDH2 rs671

6. ↓ 2 kyr later — a constraint was lifted

   822

   Hops

   Hops are in the accounts of Corbie Abbey by 822, and Hildegard of Bingen writes down the reason around 1150: they preserve. A drink that survives a journey is a drink that can be shipped, taxed and sold far from where it was made.

   The long version

   Beer stops being a household act and becomes a trade. (The companion claim — that everyone drank beer because the water was unsafe — is largely a modern invention, and is not on this line.)

   Statutes of Adalhard of Corbie, 822 · Hildegard von Bingen, Physica, ~1150

7. ↓ 328 yr later — nothing guaranteed it

   1150

   Aqua vitae

   Yeast cannot make a drink stronger than about 15%: at that point it dies in its own waste. For ten million years that was the ceiling, and every human biology on Earth was calibrated to it.

   The long version

   Alchemists at Salerno are boiling wine and condensing the vapour by the twelfth century; Taddeo Alderotti reaches near 90% in the thirteenth. The first substance in our history our metabolism has no experience of is not a new molecule — it is an old one, concentrated.

   Distillate does not arrive as a drink. It arrives as a medicine — aqua vitae, the water of life — and the oldest structure in the entire cocktail canon is the apothecary’s formula, unchanged:

   • the active ingredient — the spirit
   • something to make it palatable — sugar
   • an aromatic — bitters
   • dilution — water

   That is the Old Fashioned, and it is the root of everything else on the shelf. Bitters are the surviving evidence: Angostura was formulated in 1824 by a German army surgeon in Venezuela as a stomach tonic for Bolívar’s troops, and it is still on every back bar, still labelled as though it were a remedy.

   The Sazerac in New Orleans, around 1838, is a pharmacist’s drink in the most literal sense — Antoine Peychaud compounded the bitters himself. The story that he served it in a French egg cup, a coquetier, and that this is where the word "cocktail" comes from, is charming and almost certainly folklore; the etymology has at least six candidates and no winner.

   Six centuries pass between this gate and the word being written down. For most of them the concentrated dose is sold by chemists.

   Magister Salernus, Salerno, 12th c. · Taddeo Alderotti, 13th c. · the alembic, Jabir ibn Hayyan, 8th c.

8. ↓ 150 yr later — follows from it

   1300

   Baijiu

   Distillation reaches China through Middle Eastern trade and meets a fermentation nobody else had: *qū*, a mould starter working on solid grain in earth pits rather than a liquid mash.

   The long version

   It is the largest spirit category on Earth and this chart had no dot for it. Baijiu is roughly 99% of the liquor drunk in China, and China drinks enough of it that by volume it outsells whisky, vodka, gin, rum and tequila together. A page that gave Jerry Thomas a dot and baijiu none was not describing the world, it was describing one bar in New York.

   The technology is genuinely different, not just the flavour. Everywhere west of here, fermentation is a liquid: a mash, a must, a wort. Baijiu ferments solid grain — sorghum, packed into pits lined with stone or clay, some of them in continuous use for centuries so that the pit mud itself is the culture. The saccharification is done by qū, a cake of mixed moulds and yeasts attested from the Han, rather than by malting. Two independent answers to the same problem: how do you get a starch to become a sugar before anything can eat it.

   Dating is a range, not a year: distillation arrives during the Song–Yuan window, either through Middle Eastern trade from around 960 or with the Mongol conquest. The dot sits in the middle of that window rather than pretending to a date the evidence does not support.

   Distillation reaches China, Song–Yuan, c. 960–1368 · qū (曲) starters attested from the Han · baijiu ≈ 99.5% of spirits consumed in China

9. ↓ 193 yr later — a constraint was lifted

   1493

   Sugar, citrus

   Cane goes to the Caribbean in 1493 and the plantations follow; within two centuries sugar is cheap in Europe for the first time, and citrus travels the same routes.

   The long version

   The Navy’s ration of spirit cut with water, sugar and lime juice is grog — the first mixed drink made at scale, and the direct ancestor of every sour on every menu. None of it is a discovery about taste. It is what an empire’s cargo made affordable, on the backs of enslaved labour.

   What this release produces is not a drink. It is an arithmetic — two parts spirit, one part sour, one part sweet — and once limes and cheap sugar are landing in the same port, the whole family follows from it:

   • Daiquiri — rum, lime, sugar. Jennings Cox, an American mining engineer at Daiquirí in Cuba, 1898. Three ingredients, and every one of them is a product of the same colonial freight.
   • Gimlet — gin and Rose’s Lime Cordial, patented in 1867 to preserve the Navy’s anti-scurvy ration. A drink that exists because of a shipping regulation.
   • Margarita, Sidecar, Whiskey Sour — the same ratio, a different spirit. One template, a whole menu.

   Grog is the industrial ancestor: from 1740 the Royal Navy issues spirit already cut with water, and later sugar and citrus, to a hundred thousand men a day. It is the first mixed drink produced at scale anywhere, and it was a ration, not a pleasure.

   The sour was never invented. It was waiting for the logistics — and the logistics ran on enslaved labour, which is the part of this dot that does not soften with retelling.

   Columbus’ second voyage, 1493 · Vernon’s grog order, 1740 · Navy lime ration, 1867

10. ↓ 237 yr later — follows from it

    1730

    The Gin Craze

    London drinks something like ten litres of grain spirit a head per year at the 1743 peak.

    The long version

    Distillation had sat in Europe for six centuries as a medicine and a curiosity; cheap grain, urban poverty and no excise turned it into the first mass encounter between a population and a dose it had no evolved defence against. Eight Gin Acts between 1729 and 1751 before it broke. The pattern recurs whenever a delivery technology outruns the rules for it.

    Gin Acts, 1729–1751 · Hogarth, Gin Lane, 1751

11. ↓ 76 yr later — a constraint was lifted

    1806

    Ice

    Two things in 1806: a New York paper prints the first definition of a “cock-tail” — spirits, sugar, water, bitters — and Frederic Tudor ships New England ice to the Caribbean. Water is in the recipe from day one.

    The long version

    On 13 May a reader of The Balance and Columbian Repository, in Hudson, New York, asks what a “cock-tail” is; the editor answers: spirits of any kind, sugar, water and bitters. Four parts — and the water is one of them. Dilution is in the definition from the first day it is written down. That same year Frederic Tudor ships 130 tons of New England pond ice to Martinique and loses most of the cargo; within forty years the trade moves hundreds of thousands of tons annually, and in 1876 von Linde’s machine makes ice without a winter. The word is the idea. The cold is the release — chilling and dilution are one act, and no amount of taste could have invented a supply chain for frozen water.

    Cold is an ingredient, and it is one you can measure. Stir a Martini and it takes on something like a fifth of its finished volume in water; shake a Daiquiri hard and it is nearer a quarter, plus the air. A recipe written without that water in it is not the drink — which is why the same four ounces of gin and vermouth, stirred or shaken, are two different things.

    The drink that sold ice to the world was not a cocktail at all. The Sherry Cobbler — sherry, sugar, citrus, and a mound of cracked ice — swept America in the 1830s and 40s; Dickens gave it a page in Martin Chuzzlewit. The drinking straw became common because of it: you cannot get at a drink buried in ice any other way.

    Three answers to one question — how much water, and how fast:

    • Old Fashioned — one large piece of ice, whose only job is to melt slowly. Less surface per volume, so it dilutes late. The shape of the ice is part of the recipe, not decoration.
    • Mint Julep / Cobbler — crushed ice, maximum surface, deliberately fast. Meant to be drunk while it is still changing.
    • Highball — the dilution poured in on purpose, as soda, so the ice only has to hold the temperature.

    Every one of these is a technique for controlling a dose. That is the same job the four-part definition of 1806 was already doing, in a sentence.

    The Balance and Columbian Repository, 13 May 1806 (earlier mention: London, 1798) · Frederic Tudor, first cargo 1806 · Wyeth’s ice cutter, 1825 · Carl von Linde, 1876 · dilution figures: Dave Arnold, Liquid Intelligence, 2014

12. ↓ 56 yr later — follows from it

    1862

    Jerry Thomas

    *How to Mix Drinks, or The Bon-Vivant’s Companion* — the first cocktail book, and the first time the American bar has a written standard rather than a set of local habits.

    The long version

    A recipe becomes copyable, and a drink can exist in two cities at once. The same step print took for knowledge, four centuries earlier and one shelf up.

    The canon is a compression, not a list. That is why the book matters more than any drink in it. Embury argued in 1948 that there are six basic drinks; the modern restatement, Cocktail Codex in 2018, names six roots and derives the rest:

    • Old Fashioned — spirit, sugar, bitters, water. The apothecary’s formula (see Aqua vitae).
    • Martini — spirit aromatized with wine, stirred. First printed recipe 1888, Harry Johnson; the Martinez comes earlier. The Manhattan, New York in the 1880s, is the same root with a different spirit — not a different idea.
    • Daiquiri — spirit, citrus, sugar. The sour.
    • Sidecar — the sour with a liqueur doing the sweetening.
    • Whisky Highball — spirit and a mixer, lengthened.
    • Flip — spirit with egg or dairy for body.

    A three-hundred-drink menu carries perhaps six ideas. Everything else is a substitution, a split base, or a garnish — and once that is written down, a bartender in San Francisco and one in London can build the same drink without ever meeting.

    The same step print took for knowledge four centuries earlier: not a better recipe, a copyable one.

    Jerry Thomas, 1862 · David Embury, The Fine Art of Mixing Drinks, 1948 · Death & Co, Cocktail Codex, 2018

13. ↓ 58 yr later — follows from it

    1920

    Prohibition

    The 18th Amendment closes the American bar on 17 January 1920; the 21st reopens it on 5 December 1933.

    The long version

    What it could not do was delete the knowledge — the bartenders left and the canon went with them, to Harry MacElhone in Paris and Constante Ribalaigua in Havana. Thirteen years of suppression made the American cocktail an international form, and cost it two generations of technique at home. The year after repeal, Don the Beachcomber opens in Hollywood selling Caribbean rum under a Polynesian fiction belonging to neither place: tiki, the first cocktail style built on a heritage that does not exist, with recipes coded so the staff mixing them could not read them.

    The spirit was bad, so the recipes hid it. That constraint is the reason these drinks are still on menus a century later — they had to work with something barely drinkable:

    • Bee’s Knees — gin, lemon, honey. Honey because it covers more than sugar does.
    • Southside — gin, lime, mint. Same logic, more camouflage.

    What left the country was the knowledge. Harry MacElhone publishes the ABC of Mixing Cocktails in Paris in 1922 and the Sidecar goes into the canon from there. In Havana, Constante Ribalaigua at El Floridita puts the Daiquiri in an electric blender: the frozen Daiquiri, in the 1930s, is the first cocktail whose defining ingredient is a machine.

    Then repeal, and within a year tiki — the Zombie, and in 1944 the Mai Tai: rum blends of real complexity sold under a heritage that never existed, with the recipes written in code so the staff pouring them could not learn them.

    Thirteen years of prohibition produced more durable recipes than the sixty years that followed it. Scarcity was a better editor than abundance.

    18th Amendment in force 17 Jan 1920 · repealed 5 Dec 1933 · Don the Beachcomber, 1933–34 · Trader Vic · the Mai Tai, 1944 · MacElhone, ABC of Mixing Cocktails, 1922

14. ↓ 67 yr later — follows from it

    1987

    Fresh juice

    Dale DeGroff

    Dale DeGroff, at the Rainbow Room in New York, rebuilds the drinks out of Jerry Thomas’s book with fruit squeezed that day. Sixty years of sour mix, and the correction is not an innovation — it is a restoration.

    The long version

    In 1999 Sasha Petraske’s Milk & Honey turns it into a discipline: hand-cut ice, no menu, and a standard the rest of the world copies.

    The revival’s signature technique is three hundred years old. Clarified milk punch: pour milk into an acidic punch, let it curdle, and filter out the curds — they carry the colour, the tannin and the harshness away with them, and what is left is clear, silky, and keeps for months without refrigeration.

    • The earliest known recipe is Mary Rockett’s, dated 1711.
    • Benjamin Franklin sent one in a letter to James Bowdoin in 1763.
    • Then it is forgotten for two centuries, and re-derived from the archive by the craft revival around 2010.

    The other half of the revival is instruments rather than history: fat-washing (infuse a spirit with butter or bacon fat, freeze it, filter the fat out and keep the flavour), forced carbonation, centrifuge clarification, ice cut for density. Dave Arnold’s Liquid Intelligence, in 2014, is the point at which a bar starts measuring what it had been guessing.

    The first move of the craft revival was archaeology, not invention — squeezing fruit again, and reading the old books. Only the second move was a laboratory.

    Dale DeGroff, Rainbow Room, 1987 · Milk & Honey, NYC, Dec 1999 · Mary Rockett’s recipe, 1711 · Franklin to Bowdoin, 1763 · Dave Arnold, Liquid Intelligence, 2014

15. ↓ 14 yr later — follows from it

    2001

    Ginza

    Whisky Magazine runs its first Best of the Best in 2001: 62 judges, 47 whiskies from Britain, the United States and Japan, tasted blind. The top two are Japanese.

    The long version

    Yoichi 10-year takes first place overall and Hibiki 21 second — the first time Japanese whisky beats Scotland on a blind international panel. The distilleries are not new: Masataka Taketsuru studied chemistry in Glasgow and worked in Speyside and Campbeltown from 1918, came home, and opened Yamazaki with Shinjiro Torii in 1923. It took the rest of the world seventy-eight years to taste it without knowing what it was.

    The other half is the bar, and it is a different argument from the American one. The Ginza tradition treats service as a craft with a technique to be drilled: ice hand-carved from a single block into a sphere or a diamond so it melts slowly and evenly, the stirring counted, the shake taught as a specific motion rather than left to the wrist. Where the American revival was archaeology — go back to the book, squeeze the fruit — this one is closer to a discipline handed down by demonstration.

    Two standards for the same bar, arrived at independently. Most of the world now works somewhere between them.

    Whisky Magazine, Best of the Best, 2001 — Yoichi 10 first, Hibiki 21 second, 62 judges, 47 whiskies · Yamazaki distillery, 1923 — Shinjiro Torii & Masataka Taketsuru

16. ↓ 22 yr later — follows from it

    2023

    No safe level

    The WHO states that no amount of alcohol is safe for health, and the J-shaped curve that had made a daily drink look protective turns out to be largely an artefact: the abstainer group was full of people who had stopped because they were ill.

    The long version

    Ethanol has been a Group 1 carcinogen since 1988. The oldest recreational drug in our lineage is the one whose case has weakened fastest.

    WHO, Jan 2023 · Zhao et al., JAMA Netw. Open, 2023 · IARC Group 1, 1988

17. ↓ 3 yr later — follows from it

    2026 — now

    You are here

    2026. Consumption per head has been flat or falling for a decade across most high-income countries, and the fastest-growing category in the trade is drinks with none of it in them.

    The long version

    A technology that has outlived the problems it solved: hydration, calories, and trust between strangers all have better substrates now.

    Four laboratories, four different parts of the drink. None of them is trying to invent a taste — which is the same thing this whole line has been saying since the ADH4 mutation. They are attacks on time, on wood, on feedstock and on authorship.

    • Time. Bryan Davis’s THEA reactor (Lost Spirits, unveiled 2015) suspends oak in young spirit under high-intensity light and heat and produces something chemically close to twenty-year-aged whisky in about six days. The barrel was only ever a slow reactor kept in a cold building.
    • Wood, skipped entirely. Glyph (Endless West, San Francisco, 2018) is designed backwards: pick the molecules an aged spirit ends up with, then assemble them. No distillation, no cask, hours instead of years.
    • Feedstock. Air Company (Brooklyn, 2019) makes vodka whose carbon comes out of captured CO₂, reacted with hydrogen from water electrolysis. It won NASA’s CO₂ Conversion Challenge — first phase May 2019, final round 2021 — because the same reactor is how you would make consumables on Mars. Ten million years of ethanol out of sugar, and then some out of the air.
    • The recipe. Mackmyra Intelligens (May 2019, with Microsoft and Fourkind) is the first whisky whose recipe a model wrote: some 70 million candidate recipes generated against the warehouse’s actual casks, and master blender Angela D’Orazio choosing among them. AI-generated, human-curated — and the second half of that phrase is the interesting one.

    And one that is pure curiosity. In October 2011 Ardbeg sent vials of new-make spirit and shards of its casks to the International Space Station. In January 2012 the crew broke the wall between them and maturation began; the vials came home in September 2014, against a control batch that had sat on Islay the whole time. Microgravity turned out to extract the hard-to-reach syringic compounds less well, and the space sample tasted measurably different. It remains the only maturation experiment whose control group was four hundred kilometres below it.

    And one that may never arrive. David Nutt’s group has spent over a decade chasing a GABA-selective compound that would reproduce alcohol’s social disinhibition without the toxicity or the withdrawal. It held a dot on this axis until 13 Aug 2026 and lost it: a molecule that does not exist is not a step, and the line already ends on two that are documented. If it ever works it separates, for the first time in ten million years, what alcohol was for from the poison it arrived in.

    Note what none of these is: a new flavour. Three of them are ways to stop waiting, and the fourth changes who decides — which makes it, on this site of all places, the one worth watching.

    Lost Spirits THEA reactor, 2015 · Endless West Glyph, 2018 · Air Company, 2019 — NASA CO₂ Conversion Challenge · Mackmyra Intelligens, May 2019 · Ardbeg ISS experiment, Oct 2011 – Sep 2014 · David Nutt / Alcarelle

The Italian lineage — the second band, drawn below the axis on the explorer

1. 1786

   Vermouth di Torino

   Antonio Benedetto Carpano

   Antonio Benedetto Carpano, Turin: wine fortified, sweetened and aromatised with wormwood and some thirty botanicals. Wormwood wines are ancient — what is new is the style and the occasion.

   The long version

   This is a drink built to come before the meal, and the bitterness is the point of it: it is not there to be pleasant, it is there to make you hungry. Traditionally dated, and the first recorded recipe for the sweet Turin style. Two hundred and thirty-one years later the category is given legal borders — the decree of 22 March 2017 reserves the name to Piedmont, Italian wine, and Piedmontese artemisia.

   Antonio Benedetto Carpano, Turin, 1786 · Vermouth di Torino IGP, ministerial decree, 22 Mar 2017 · EU recognition, 2019

2. ↓ 74 yr later — follows from it

   1860

   Campari

   Gaspare Campari

   Gaspare Campari, Novara. The bitter stops being a house recipe and becomes a product — the same colour, the same bitterness, the same bottle, in every bar.

   The long version

   That is the step an assembled drink actually needs: you cannot standardise a cocktail on an ingredient that changes from one counter to the next. The assembly follows almost at once, at the Caffè Campari under the Milan Galleria: the bitter from Milan poured into the vermouth from Turin and called Milano–Torino, then given soda and renamed Americano around the turn of the century. Still nothing distilled, still under 15%.

   Gaspare Campari, Novara, 1860 · Caffè Campari, Galleria Vittorio Emanuele, Milan, 1867

3. ↓ 59 yr later — follows from it

   1919

   Negroni

   Camillo Negroni · Fosco Scarselli

   Count Camillo Negroni asks Fosco Scarselli, at Caffè Casoni in Florence, to strengthen his Americano; the soda comes out and gin goes in.

   The long version

   It is the one point where this line touches the spirituous cocktail, and the strength roughly doubles. The same year, at the Padua fair, the Barbieri brothers show Aperol at 11% — a third of a Negroni. 1919 hands the Italian aperitivo both its strongest drink and its weakest, and it is the weak one that will travel. Look straight up from here: the American bar is closing the same year.

   Caffè Casoni, Florence, 1919 — Luca Picchi, *Sulle tracce del conte* · Aperol, Barbieri brothers, Padua International Fair, 1919

4. ↓ 51 yr later — follows from it

   1970

   The bitter goes in

   The step that actually makes the drink, and it is fifty years after the bottles it is made from. Through the 1800s and long afterwards a *spritz* is wine with sparkling water in it — Habsburg practice, and the word is German: *spritzen*, to spray. It is still precisely that in Trieste and Gorizia, where ordering a spritz gets you white wine and seltzer and the bitter has to be asked for by name. What happens in the Veneto in the 1960s and 70s is that the bitter goes into the glass: white wine, soda, and Select in Venice or Aperol in Padua. Both liqueurs had existed since 1919–20 and neither had been this. The soldiers-splashing-water story behind the word is repeated everywhere and documented nowhere; the etymology is solid, the anecdote is folklore.

   Aperol — Barbieri brothers, Padua, 1919 · Select — Fratelli Pilla, Venice, trademark registered 29 May 1920 · the wine-and-seltzer spritz survives unchanged in Trieste

5. ↓ 33 yr later — a constraint was lifted

   2003

   Spritz goes global

   Not a birth — the dot before this one is the birth. This is the year a regional habit acquires a global marketing budget: Campari Group buys Aperol, fixes the recipe at three prosecco, two Aperol, one soda, and starts selling the hour rather than the bottle. Within fifteen years the lowest-proof drink in the canon is among the most ordered on Earth — the exact inverse of the spine above, where every step is about breaking a ceiling. Italy’s most successful drinking export turns out not to be a spirit but a time of day.

   Campari Group acquires Barbero 1891, and with it Aperol, 2003 · the 3-2-1 recipe dates from that ownership · “It Starts With Aperol Spritz”, from c. 2017