The End of the Cretaceous: Two Causes, One Extinction, No Simple Split

I The Boundary Is a Surface in the Rock, Not a Number

Let us begin with something that may come as a surprise. The official definition of the line between the Cretaceous and the Paleogene is neither a numerical age nor “an iridium concentration above such-and-such a level.” It is a surface in the rock on a hillside in northwestern Tunisia.

The place is called El Kef. The International Commission on Stratigraphy approved the proposal in 1990, and the International Union of Geological Sciences formally ratified it in 1991. The boundary point lies at the base of the “boundary clay,” where deposition of the lowermost layer rich in impact material begins. The iridium anomaly, the mass extinction of planktonic foraminifera, and the disturbance in carbon isotopes are all immensely important markers. But they help us recognize the line. They are not the line itself.

We will need this distinction throughout the essay, so let us establish it now: the operational definition is a physical interface; the criteria for identifying it are a set of signals that can corroborate one another. The two are not interchangeable.

There is something else that runs against intuition. El Kef was not chosen because its boundary clay was the thinnest, as close as possible to “an instant.” Quite the opposite: the boundary interval there is unusually expanded. New drilling and older exposures put the clay at fifty to fifty-five centimeters thick, whereas at many open-ocean sections it measures only millimeters to a few centimeters. An expanded record has its uses. It gives the changes of the very earliest Paleogene somewhere to be recorded.

And “boundary clay” does not mean a rock made to the same recipe everywhere. The “Fish Clay” at Stevns Klint in Denmark and the clay at El Kef in North Africa differ in mineral composition: montmorillonite, mixed-layer illite–montmorillonite, kaolinite, calcite, and iron oxides occur in different proportions. What allows the layers to be correlated around the world is a combination of things: their stratigraphic position, iridium and other platinum-group elements, impact glass and spherules, shocked minerals, and an abrupt change in the fossil succession.

Now for the iridium. In 1980, four researchers published a paper entitled “Extraterrestrial Cause for the Cretaceous-Tertiary Extinction.” They had not set out to find a catastrophe at all. They wanted to use iridium as a timer. Cosmic dust falls continuously onto Earth; a high concentration of iridium would indicate slow sedimentation, a low concentration rapid sedimentation.

Then the measurements came back far too high. At the three sites originally studied, iridium in the boundary layer was enriched by about thirty, one hundred and sixty, and twenty times, respectively, relative to the background in the adjacent sediment.

“Slow sedimentation” could not explain concentrations on that scale. They required an influx of extraterrestrial material over a very short time. The researchers also specifically considered, and ruled out, a nearby supernova. Instead, they proposed a chain of events: a large asteroid impact, dust blocking sunlight around the world, and the collapse of food chains. A marker intended to serve as a stopwatch became evidence of an event because its reading went off the scale.

How often was that iridium anomaly found again? By the time of a global synthesis in 2010, several hundred Cretaceous–Paleogene boundary sites had entered the impact-layer database. The iridium anomaly itself had been reported repeatedly at more than a hundred terrestrial, shallow-marine, and deep-sea sections. A layer of clay a few centimeters thick was independently recognized at more than a hundred places around the world. And it was the same layer.

One more finding arrived in 2021: an iridium-rich layer was identified in drill cores from inside the Chicxulub crater itself. This matters a great deal. We will need it in the next section.

Iridium is not the only marker. The planar deformation features in shocked quartz require transient pressures far beyond those an ordinary volcanic explosion can generate. Around the boundary, the grain size of that quartz generally decreases with distance from Yucatán, consistent with a field of material thrown outward from there.

Nickel-rich spinels formed as molten droplets of the impactor entered the atmosphere at high speed, oxidized, and crystallized. Their trace elements and isotopes are closer to those of meteoritic material. Chromium isotope measurements examine isotopic composition, not merely abundance, and further weaken the possibility that the iridium came from ordinary terrestrial volcanism.

One marker deserves separate attention, because it will come back to bite us: osmium isotopes. Across the boundary, seawater osmium shifts toward less radiogenic values, which can record a sudden addition of material.

But extraterrestrial material and mantle-derived basaltic volcanic material can both supply low osmium isotope ratios. Osmium is therefore an excellent global tracer, but not a fingerprint that can distinguish “impact” from “volcanism” on its own.

Finally, the age. In the 2026 edition of the International Chronostratigraphic Chart, the boundary is still marked at exactly 66 million years. That is a standardized, rounded boundary age, not the reading from a particular experiment. High-precision dating yields another set of numbers. A combined argon–argon result published in 2013 was about 66.04 million years, with an analytical uncertainty of roughly 11,000 years, increasing to about 43,000 years once systematic uncertainty in the timescale was included.

How did the “65 million years” in the textbooks become “66 million years”? At least two kinds of revision were involved, and they are often muddled together.

One comes from better techniques: purer single minerals, chemical-abrasion uranium–lead dating, better mass spectrometry, and better age–depth models. The other comes from changes to the ruler itself: recalibration of the ages of mineral standards and cross-calibration of decay constants in the potassium–argon system. The second kind can shift an entire set of old measurements together. It does not mean that new samples have independently made the boundary half a million years older.

Here is a rule worth keeping. When an age is revised, the first question is not “Who measured it?” It is “Did the sample change, or did the ruler?”

II A Crater an Oil Company Found First

The 1980 paper proposed an impact. It did not have a crater. At the time, nobody knew where the crater was.

In fact, people had already encountered it. They simply had not thought of it in those terms. During the 1950s and 1960s, wells drilled in Yucatán by Mexico’s national oil company struck unusual breccias and rocks that had crystallized from melts. They were interpreted at the time as volcanic rocks or basement structures.

In 1978, petroleum exploration geophysicists Glen Penfield and Antonio Camargo identified a nearly circular anomaly in aeromagnetic surveys of northern Yucatán. Penfield then combined the offshore magnetic anomaly with old gravity data from onshore. Together they outlined a ring-shaped structure roughly 180 kilometers across.

The immediate purpose was to find oil, not to test the Alvarez hypothesis. Once again, the evidence existed before the question. It had simply been filed under another name. This has the same shape as the scene that opened Essay Seventeen. There, tiny bodies under a microscope were taken for bacteria. Here, a ring-shaped structure was taken for a volcano.

The circle was not closed until a decade later. Penfield had tried to attract the attention of impact researchers, but the real turning point came in the late 1980s and around 1990. Thick impact ejecta were found in the Caribbean and Haiti. Alan Hildebrand and others began looking nearby for a large crater. The journalist Carlos Byars supplied the lead about the Yucatán anomaly, and samples from the old oil-well cores revealed shock metamorphism. In 1991, Hildebrand, Penfield, Boynton, Camargo, and their colleagues formally proposed Chicxulub as a candidate for the Cretaceous–Paleogene impact crater.

Gravity and magnetic surveys, seismic reflection, drilling, and petrology subsequently established its impact origin together. Modern geophysical data generally put its diameter at 180 to 200 kilometers. It is a multiring basin, with an inner peak-ring structure characteristic of large, complex impact craters.

The crater lies beneath the northern Yucatán Peninsula and the adjacent sea, covered by younger sedimentary rocks. Its top is broadly buried several hundred meters to about a kilometer deep. That is why no ring is visible at the surface. When did the impact happen? The strongest constraints come from two complementary approaches.

One dates impact glass directly. Gorgonilla Island in Colombia preserves unusually fresh glass spherules. Their argon–argon age is about 66.05 million years, with an uncertainty of roughly 31,000 years, rising to about 54,000 years when systematic uncertainty is included.

The other cross-calibrates impact glass from Haiti, volcanic ash near the boundary in North America, and the boundary timescale, yielding about 66.04 million years. Within their uncertainties, both results are indistinguishable from the formal boundary. But this essay wants to emphasize something else: the strongest evidence that the impact and the boundary belong to the same event layer is not a numerical age. It is their geometry.

Outside the crater, impact ejecta sit directly on the extinction boundary, from thick spherule beds close to the Gulf of Mexico to iridium-rich clay far away. Inside the crater, iridium enrichment appears in the earliest sediment above the impact melt rock and breccia. The 2021 discovery placed “formation of the crater” and “settling of the global iridium layer” within one continuous sequence inside the crater.

Numerical ages can share systematic errors and shift when the ruler is recalibrated. Stratigraphic relationships provide another kind of constraint, one that recalibration does not automatically rewrite. But erosion, reworking, and depositional gaps can place deposits of different ages next to each other. “Lying together” still has to be read within a sedimentary history.

The 2016 Chicxulub drilling expedition recovered cores from the peak ring. In 2019, about 130 meters of impact melt rock, breccia, and rapidly deposited infill in those cores were interpreted as a sequence formed on the day of the impact and in its immediate aftermath. It records seawater flooding back in, tsunami backwash, and the arrival of material from land.

“Part of the same event as the impact” is a very firm conclusion. Translating a particular stretch of centimeters or meters into “how many hours after the impact” depends on models of sediment dynamics and is much less firm.

Another section pushes the resolution to an even more extraordinary level. In the Tanis event bed in North Dakota, a sudden surge buried large numbers of fish and other organisms. About half the fish examined had glassy impact spherules lodged in their gill rakers. In other words, these fish were already dead while the spherules were still falling through the air or the water.

According to the models, this is a record of the first tens of minutes to hours after the impact. The link between the spherules and the boundary impact is strong. Whether the arriving surge was the tsunami itself or a standing wave in an inland body of water excited by seismic waves, and the exact timing of the successive waves, remain matters of debate.

A 2022 study examined skeletal material from three paddlefish and three sturgeons. Growth lines, changes in bone cells, and carbon-isotope records pointed to boreal spring. These are complementary clues from a shared body of material, not three independent reruns of the event. On this study’s reading, on that day sixty-six million years ago, spring was arriving in the Northern Hemisphere.

Material with subannual resolution is rare. It cannot tell us which year that year was. The uncertainty around “66 million years” is still tens of thousands of years wide. Yet within that breadth of tens of thousands of years, there is one day whose season we can identify.

III What It Hit Matters More Than How Hard It Hit

We can calculate the energy released when an object some ten kilometers across strikes at cosmic speed, and calculate it quite accurately. But the point of this section is that energy is not what determines the outcome, or at least not the only thing that does.

The impact site was not a clean slab of granite. It was a shallow-marine carbonate platform: thick carbonate rocks interbedded with sulfur-rich evaporites, especially anhydrite, along with organic matter and hydrocarbon-source material.

A shock wave can decarbonate carbonate rock, releasing carbon dioxide. It can vaporize and break down sulfates, ultimately producing sulfur-bearing gases and sulfate aerosols. And sulfate aerosols are among the leading actors in an impact winter.

This leaves us with a problem. The crucial original sedimentary layers at the center of the impact have been vaporized, melted, and blasted away. There is no intact, undisturbed drill core that can tell us directly what percentage of the total volume struck was anhydrite. That has to be reconstructed from regional stratigraphy and numerical models. The proportions of the target rocks are therefore part of the uncertainty in the amount released, rather than an input we can fix first and then use to calculate the consequences.

To see how much this matters numerically, look at the changing estimates for sulfur. More recent high-pressure impact models once yielded instantaneous releases of about 325 billion tonnes of sulfur and 425 billion tonnes of carbon dioxide, with an uncertainty of around 130 billion tonnes for sulfur. Earlier models covered a still wider range.

In 2025, however, researchers used sulfur isotopes from five boundary sections in North America, together with a sulfur mass balance, to reconstruct the release anew. They obtained about 67 billion tonnes, with a margin of roughly 39 billion: a range from 28 to 106 billion tonnes. The new central value is about one-fifth of the values from some major models of recent years.

This gap does not tell us who got the calculation wrong. It tells us that we know the impact energy much more securely than we know how much sulfur actually entered the climate system. And the latter is the number climate models really need.

This has the same shape as the argument about large igneous provinces. The consequences depend not only on how much energy there is, but also on what it strikes. Put the same energy into sulfur-poor crystalline basement, or into deep ocean water, and the global loads of sulfate aerosols and soot would be very different.

The angle comes along to complicate matters too. A three-dimensional numerical simulation published in 2020 compared asymmetries in the crater’s structure with possible impact trajectories. The best fit was a relatively steep impact, about forty-five to sixty degrees above the horizontal, arriving roughly from the northeast.

That is close to the range of angles that can efficiently eject deep, sulfur-rich and carbonate-rich target rocks into the atmosphere. But the angle is the result of structural inversion combined with hydrodynamic modeling. It is not an ancient angle of approach that we can measure directly.

Now for “impact winter.” It is really three different things, and the strength of the evidence differs for each.

The first is fine silicate dust. In 2023, researchers fed measured grain-size distributions from boundary sediments into a climate model and brought fine dust back to prominence. Micrometer-scale particles can remain aloft for years. In the model, global mean surface temperature at one point fell by about fifteen degrees Celsius, sunlight usable for photosynthesis was almost shut off for roughly two years, and the climatic influence of fine dust could last more than a decade. “There was a great deal of impact dust” is a firm conclusion. “Fifteen degrees,” “two years,” and “more than a decade” are quantities highly sensitive to particle size, injection altitude, and the climate model.

The second is sulfate aerosols. Older models used sulfur quantities several times the new estimate and produced extremely strong cooling lasting many years. The much lower estimate from 2025 does not remove this mechanism. But it substantially weakens the version in which sulfur alone accounts for all the darkness and cold, increasing the relative importance of dust and soot.

The third is soot. Polycyclic aromatic hydrocarbons and combustion products in boundary deposits and sediments inside the crater show that extensive high-temperature burning occurred. But organic geochemical work on crater and distant deposits in 2020 found that much of the soot was more consistent with rapid pyrolysis of sedimentary organic matter and hydrocarbon-source rocks at the impact site than with “every forest on Earth catching fire at once.” The evidence that soot entered the global atmosphere is strong. The evidence that most of it came from wildfires around the world is much weaker. These two statements are often treated as one.

Global wildfire itself remains contested. Early models proposed that the infrared heat pulse from reentering ejecta could ignite vegetation worldwide. Critics pointed out that charcoal in many nonmarine boundary beds does not show the globally uniform abundance that would be expected, and that the polycyclic aromatic hydrocarbons could have come from hydrocarbons in the target area.

New thermal-radiation simulations in 2026 offered another possibility. A thick curtain of dust lingering after the impact could increase the absorption and downward radiation of heat from reentering spherules, making the heat load at the surface several times higher than calculations based on spherules alone. Widespread, even near-global, ignition would once again become possible. This is a very recent numerical result from a single team. It cannot be written up as the end of the argument.

We can move more quickly through the other consequences. A huge tsunami spreading around the world was an inevitable near-field outcome of the impact dynamics. Around the Gulf of Mexico, exceptionally thick redeposited sandstones, fragments of land plants, and evidence of returning water are preserved.

Models estimate local incoming waves at coasts at roughly fifty to a hundred meters, or a hundred to three hundred meters in some scenarios. Global models estimate that the tsunami’s initial energy may have been about thirty thousand times that of the 2004 Indian Ocean tsunami. The impact energy coupled into Earth’s crust was equivalent to an extreme earthquake of around moment magnitude ten. That is an energy conversion, however, not a reading from an ancient seismograph.

Acid rain followed from sulfur oxides and nitrogen oxides, over timescales of months to years. But the buffering capacity of the enormous ocean rules out the idea that “the entire deep sea instantly became strongly acidic.” Studies using boron isotopes and other evidence in 2019 pointed to rapid acidification of surface seawater after the boundary. That is much more direct evidence than simply inferring from sulfur in the target rocks that “there must have been acid rain.”

There was also an effect in the opposite direction. The carbon dioxide released instantaneously by the impact, together with organic matter heated and burned, would leave a greenhouse forcing after the initial cold, over hundreds to thousands of years or longer. Older calculations often put the long-term warming from impact-derived carbon dioxide alone at less than one degree Celsius. “Cold first, warmth afterward” is physically reasonable. Its magnitude should not be presented as a settled value.

IV The Other Cause Was There All Along

Now let us look the other way.

In western India lies a vast expanse of basalt. About 500,000 square kilometers survive; reconstructions usually put its original extent at one million to one and a half million square kilometers, or more. The total erupted volume, combining what is preserved and what is inferred, is on the order of a million cubic kilometers. In the main sections of the Western Ghats, lava is piled layer upon layer, several kilometers thick. Individual giant lava units can extend laterally for hundreds to more than a thousand kilometers.

These are the Deccan Traps. They are a classic continental flood-basalt province: not one volcano, not one eruption, but a long interval of volcanic activity made up of many lava-flow units and pauses between them. The essential fact about the timing fits in one sentence: it began before the impact and continued after it.

High-precision radiometric dating and paleomagnetism both place the main volume within an approximately million-year window spanning the end of the Cretaceous and the very beginning of the Paleocene. A new paleomagnetic synthesis in 2026 put the principal phase on a timescale of 700,000 to 800,000 years and confirmed that the eruptions crossed the boundary.

The boundary itself falls within an interval of reversed magnetic polarity, allowing distant lava sequences to be attached to a common framework. But paleomagnetism supplies only the order and the broad framework. A polarity interval can last hundreds of thousands of years; it is not a stopwatch that reads to the nearest thousand. The real disagreement lies between two stopwatches, whose results appeared in the same issue of a journal in 2019.

One is uranium–lead dating of zircon. Basalt itself seldom contains zircon suitable for high-precision uranium–lead dating, so researchers mainly use volcanic ash, red-bole horizons, or zircon-bearing volcaniclastic beds between the lava flows to build age–stratigraphy models. The isotopic precision of an individual zircon grain can be very high. But its crystallization or depositional age must still be translated into the actual eruption ages of the basalt packages above and below it.

The other is argon–argon dating. Researchers separate potassium-rich plagioclase and other minerals directly from Deccan basalt flows, making the link between an age and a particular flow more direct. But absolute ages depend on the calibration of monitor standards and the potassium decay system. They are also affected by the low potassium content of young basalt, alteration, and excess argon.

Both clocks can achieve internal precision on the scale of tens of thousands of years. But if we want to compare the two “absolute clocks” to within ten or twenty thousand years, systematic uncertainties cannot be left out.

Here are the results. The uranium–lead group reconstructed roughly four high-volume eruptive phases. They concluded that substantial eruption pulses had begun before the boundary and placed the boundary relatively high in the main Deccan sequence. Combining their age model with the volume reconstructions then available, the proportion already erupted before the boundary could approach half.

The argon–argon group likewise concluded that more than ninety percent of the main volume erupted within less than about a million years. But their reconstructed time–volume curve was much more continuous than “four discrete pulses.” Their model placed about three-quarters of the main lava volume after the boundary and one-quarter before it, and inferred a rise in average lava flux after the boundary.

The crucial disagreement between the two papers published together was therefore not whether the Deccan eruptions straddled the boundary. Both agreed that they did. It was where the boundary fell within that stack of lava, and whether the volume flux had been reorganized after the impact.

A 2021 reassessment of both dating datasets and the methods used to calculate eruption rates concluded that the disagreement had not gone away. Deriving “instantaneous eruption flux” from discrete age measurements depends heavily on the age–depth model, the estimated volumes of the different stratigraphic units, and whether age uncertainties are turned into intervals of quiescence.

The same set of ages can look pulsed under one statistical treatment and more continuous under another. The conflict of 2019 cannot simply be reduced to “one paper was later shown to have measured the dates wrong.”

Work in 2025 and 2026 narrowed the disagreement over the broad framework. Correlations using astronomical cycles and marine records lent further support to a sulfur-rich eruptive phase already underway a little over 300,000 years before the boundary.

New reconstructions of volumes through space and time also supported a substantial proportion of volcanism continuing after the boundary. Paleomagnetism tightened the main phase to 700,000–800,000 years. Yet by 2026, different methods still had not converged on a single percentage for how much erupted before the impact, roughly half or roughly a quarter, or agreed on whether there was a genuinely discrete jump in rate afterward.

How much gas did the Deccan release? It is best not to answer with just one number. Estimates of total carbon dioxide span at least an order of magnitude, with the broader model ranges amounting to trillions to tens of trillions of tonnes. Include deep magmatic degassing and intrusions, and the total can rise further.

A study of melt inclusions in olivine in 2021 directly constrained the carbon dioxide content of early Deccan magmas. Yet it found that a simple degassing calculation based on the surviving lava volume might not suffice, by itself, to account for all the warming in the late Maastrichtian. This suggested that large quantities of carbon dioxide may already have escaped during magma transport at depth, before eruption.

There is a comparison here that is very easy to get wrong. The impact released its hundreds of billions of tonnes of carbon dioxide over hours to days. The Deccan could accumulate a far greater total, but spread over hundreds of thousands to a million years.

Climatic consequences depend not only on the total but also on the release rate and the carbon cycle’s response. The ocean and silicate weathering partly absorb slow inputs; very high-flux pulses can drive abrupt changes. “The Deccan released much more carbon dioxide in total” can therefore be true without implying stronger short-term forcing than the impact.

The same applies to sulfur. Estimates from melt inclusions suggest that a typical cubic kilometer of Deccan basalt released about 3.5 million tonnes of sulfur dioxide, or more than five million tonnes in sulfur-rich scenarios.

Some of the most sulfur-rich lavas from before the boundary had distinctly higher original sulfur contents than many lavas from afterward. The integrated total may have been enormous, but the aerosols lived for only years. What really determines the cooling, then, is how many cubic kilometers erupted within a few years or decades, and how much sulfur reached the stratosphere. Those are precisely two of the hardest quantities to recover from the surviving flood-basalt record.

A more useful contrast is a single immense impact pulse versus repeated volcanic pulses over hundreds of thousands of years. The totals can be compared, but without release times and the gases’ subsequent fate, they cannot by themselves compare climatic consequences.

Finally, there is the environmental disturbance before the boundary. This is the firmest evidence that “the Deccan had effects.” Multiple marine oxygen-isotope records identify late Maastrichtian warming that began or peaked about 150,000 to 300,000 years before the boundary. Estimates of global or deep-ocean warming range from 2.5 to 5 degrees Celsius, followed by a cooling interval as the boundary approached.

Its timing overlaps the initial part of the main Deccan eruptions, and that correlation is fairly firm. Quantitatively assigning all the warming to a particular Deccan flow formation is another matter, limited by uncertainty in the timescale and the carbon dioxide release rate.

Mercury anomalies are often used as a proxy for volcanism, because large modern volcanoes do emit considerable amounts of mercury. Several late Maastrichtian marine sections have reported mercury anomalies, and some peaks coincide with the warm interval. But mercury can move during diagenesis, changes in redox conditions and clay content, and weathering. There are several steps between “a mercury anomaly is present” and “every mercury spike equals a Deccan eruption.”

V Two Causal Diagrams

Put the timing of the two processes side by side, and the most secure relationship is this: the Deccan began first; the impact occurred within its main eruptive interval; the Deccan continued afterward.

High-precision radiometric ages, paleomagnetism, and marine climate records all argue against the idea that “the Deccan suddenly started from nothing only after the impact.” The main eruptions began at least hundreds of thousands of years earlier and continued for hundreds of thousands of years beyond the boundary. Two different causal diagrams can be drawn over this same arrangement.

The first has two independent arrows. Deccan volcanism points toward environmental stress; Chicxulub points toward environmental stress. They overlap in time but have nothing to do with each other.

The second adds an arrow. Chicxulub points toward a change in the Deccan’s eruptive state, which in turn points toward additional volcanic forcing. The impact’s direct climatic effects remain in the diagram as well.

The second diagram did not come from nowhere. A proposal in 2015 suggested that when the powerful, world-traveling seismic waves generated by Chicxulub passed through India, they could have changed fracture permeability, magma foaming, and transport through dikes in a magmatic system already near a critical threshold. Notice the form of the claim. It does not conjure a volcanic province out of nothing. It switches an already active system from one eruptive state to another.

Is that physically possible? Numerical estimates suggest that, upon reaching the Deccan, Chicxulub’s seismic waves could still have carried dynamic energy densities on the order of 0.1 to 1 joule per cubic meter. In modern volcanic and geothermal systems, distant large earthquakes have indeed been observed to trigger microseismicity, permeability changes, and small eruptions. So “a powerful, distant earthquake disturbs a magmatic system near a threshold” does not violate physics.

But extrapolating from modern systems measuring tens to hundreds of cubic kilometers to a flood-basalt province on the scale of a million cubic kilometers is an enormous leap in scale. Physically possible does not mean historically actual.

The evidence for the second diagram comes mainly from eruption rates. One synthesis concluded that lava flux increased two- to threefold after the impact. The argon–argon group’s placement of about three-quarters of the main lava volume after the boundary points in the same direction. In 2018, researchers also analyzed global seafloor gravity anomalies and proposed that magmatic activity at mid-ocean ridges increased near the boundary. This might be another example of the same seismic waves triggering additional magma in decompression-melting systems around the world.

There are two problems. First, the uranium–lead group did not reconstruct the same history of rates. Second, the resolution of the mid-ocean-ridge evidence, involving ocean-crust ages, sediment cover, spreading rates, and inversion of gravity anomalies to estimate lava volume, is far coarser than the instantaneous timescale of the impact.

Most damaging of all is the precision of the timing itself. It has been proposed that the Deccan magmatic system changed state within 50,000 years after the impact. That sounds close.

But uncertainties in the different timescales, the volumes of flow formations, and models of quiescent intervals are themselves on the order of tens of thousands of years. The conclusion that “the state change occurred zero to fifty thousand years from the impact” is on the same scale as its own uncertainty. Numbers like that cannot establish the direction of causation. And something else needs to be made explicit, because it is so often overlooked: “pure coincidence” is a genuine null hypothesis.

The literature arguing that the impact was the primary cause does not require it to have triggered the Deccan. The Deccan could have begun and continued independently through processes involving the Indian Plate and the mantle. Chicxulub simply happened to fall within that roughly million-year volcanic window.

The many-author synthesis of 2010 and the climate and ecological models published from 2020 onward can still identify the impact as the primary cause precisely because they treat the ongoing volcanism as background forcing, while matching the sharp moment of extinction to the impact layer. In logical terms, coupling is not necessary. Does being “so close in time” count as evidence in itself, then?

That depends on how the question is framed. If the main Deccan phase lasted 700,000 to a million years, and Chicxulub occurred within that window, the statistical surprise of their overlap depends entirely on the window used to define a large igneous province event.

Conversely, a change in eruption flux within a few tens of thousands of years after an impact can be identified as impact-triggered only if we can show that it exceeded the volcanic system’s spontaneous fluctuations, and if several independent clocks all place it after the impact. The second condition has not yet been met. We can now set out three positions and give each its strongest version.

The first makes the impact the primary cause. Its strongest version requires no coupling at all. The global ejecta layer, the iridium layer inside the crater, and the extinction layer are synchronous at the resolution currently available. The Deccan lasted much longer, both before and after the extinction, yet not every major eruptive phase corresponded to a biological extinction on the same scale. That is how the 2010 synthesis organized its case.

The second makes the Deccan the primary cause. Its strongest version need not deny the impact. Late Maastrichtian warming is empirically documented; so are mercury anomalies; and so are high-flux pre-boundary eruption pulses within the uranium–lead age model. This position holds that volcanic environmental stress had already pushed the system past its threshold before the impact.

Its proponents also point to something that is genuinely true: the near-field stratigraphy is complicated. Some Gulf of Mexico sections contain multiple spherule beds. Tsunamis, erosion, and redeposition make it unsafe to treat all the spherule-rich layers as recording one and the same instant.

That warning is reasonable. The stronger version, which shifts the main extinction wholesale to before the impact, was seriously weakened by the high-precision ages of 2013 and the discovery of iridium inside the crater in 2021.

The third argues for a combined effect, and this category contains very different accounts. In some, the Deccan first produced prolonged warming, acidification, and ecological stress, and the impact delivered the final, extreme, short pulse.

In others, the impact caused the main extinction, then amplified the pressure afterward by triggering the Deccan. Other work identifies more than one pulse of biological crisis in regions such as Antarctica and attempts to match them separately to volcanic warming and the boundary impact. What these accounts share is a refusal to assign the other process zero environmental effect.

VI What a Verdict Requires, and What Is Missing Here

The preceding sections have been laying out the material. This one asks a question about method: what kind of evidence would separate the causes? There are four kinds. Let us see, one by one, what each can do here.

The first is nonoverlapping time windows. The ideal situation would be an extinction occurring within the window when Cause A was effective, while Cause B had not yet begun or had already ended. Time itself would then separate them.

The end of the Cretaceous does not meet that condition. Chicxulub lies very close to the extinction layer, but Deccan volcanism and the late Maastrichtian environmental disturbance had already begun. We therefore need to distinguish what precision in time can do here from what it cannot.

It is enough to argue against the strong version that “Chicxulub preceded the boundary by hundreds of thousands of years and had nothing to do with the main extinction.” The absolute ages of impact glass, the global ejecta layer, and the iridium layer inside the crater all pin the impact to the boundary. It is also enough to argue against “all of the Deccan occurred after the impact,” because large volumes of lava and the late Maastrichtian warming clearly predate it.

Dates alone cannot answer “seventy percent impact, thirty percent volcanism,” because volcanic effects overlap the pre-impact background. Sharper timing often rules out scenarios; it does not automatically yield each cause’s share. Estimating contributions also requires mechanisms, exposure, and ecological responses, with their assumptions tested. Ruling out a scenario and assigning causal weight are different jobs.

The second kind is a pattern of selectivity. This comes closer to a test of causation than mere simultaneity, because different mechanisms predict different lists of casualties.

A brief, intense episode of darkness and cold should hit especially hard those organisms that depend directly on living photosynthetic primary producers, cannot fast for long, lack seeds or detritus as alternative food, and need large amounts to eat. Prolonged warming and chronic carbon dioxide forcing should instead produce patterns associated with heat tolerance, latitude, habitat temperature, metabolism, ocean deoxygenation, and sensitivity to acidification.

What does the actual list look like? Marine plankton dependent on photosynthetic food chains, ammonites, and large terrestrial food webs were devastated. Freshwater detrital food webs, many turtles, crocodyliforms, and amphibians were relatively buffered. That is highly consistent with a sudden collapse in light.

But acidification can also selectively harm marine calcifiers. Volcanic sulfur pulses can also produce short-term cooling. Deccan carbon dioxide can also impose climatic and acidification stress. “The selectivity fits an impact winter” does not mean that only an impact winter could produce it.

The third kind is spatial pattern. In theory, Chicxulub’s mechanical destruction, thermal pulse, and tsunami should diminish markedly with distance from the crater. Stratospheric dust and aerosols would become global. The Deccan’s climatic forcing could also become global, but prolonged warming usually differs with latitude, between land and sea, and with ocean depth.

In theory, then, we can ask whether extinction spread outward from the Gulf of Mexico. Did high latitudes provide a buffer? Were deep-sea bottom dwellers and surface plankton affected at the same time?

The practical problem is that the age uncertainties of most fossil sections far exceed the travel times of shock waves, tsunamis, and even the spread of impact winter. The extremely thick ejecta and tsunami deposits around the Gulf of Mexico are clearly more substantial than those far away: a spatial fingerprint of the impact. But once we ask whether a particular Antarctic group went extinct two thousand years after its North American counterpart, or at the same time, the correlation uncertainties of most sections already exceed the difference we are trying to measure.

The fourth kind is the order of ecological responses. An ideal impact-winter sequence would run like this: photosynthetic primary producers collapse rapidly; herbivores and filter feeders decline next; consumers at higher trophic levels follow. On land, forests collapse and spore-bearing plants produce a “fern spike.” In water, detritus-based chains persist relatively well.

Records of marine planktonic foraminifera, calcareous nannofossils, and terrestrial pollen do support a reorganization of primary-production systems at the boundary. But sediment mixing and the time averaging of fossils generally distinguish only what came before and after the same event layer. They cannot reliably separate producer–consumer lags of a few years. Having examined the four kinds of evidence, we can draw a very specific boundary between what can and cannot be decided.

Four things can be judged with high confidence. A Chicxulub impact occurred. It was synchronous with the global boundary layer and the main extinction at the resolution currently available. The Deccan had begun before it and continued afterward. Significant late Maastrichtian climate disturbance was already underway before the impact.

Four other things cannot be judged with equal confidence. Without the Deccan, would the impact still have caused an extinction on the same scale? Without the impact, would the Deccan have caused a mass extinction at that same moment? Did the impact trigger the later phase of the Deccan? How much causal weight belongs to each process?

The first four are observational constraints. The second four are counterfactual questions. This distinction is the heart of the essay, and Section VIII will return to it. While we are here, let us also separate two meanings of “we cannot decide,” a phrase often abused.

If it means “we have absolutely no idea which event the main extinction is more closely associated with,” it does not represent the current mainstream view. The synchrony between the global impact layer and the extinction is markedly stronger than the association with any single dated Deccan flow formation. The impact as primary cause is currently the relatively dominant synthesis. The 2010 review involving forty-one authors, along with several marine-climate and ecological-niche models from 2020 onward, points in that direction.

If it means that we cannot assign reliable quantitative weights to impact and volcanism, cannot determine whether the impact triggered the Deccan, and cannot know how much weaker the extinction would have been without the other event, then it expresses a broadly shared methodological position.

This essay uses the second meaning.

VII Were They Already in Decline Before the Boundary?

One question has hovered beside this debate all along: were non-avian dinosaurs already declining before the boundary?

It matters because, if the answer is yes, the impact looks more like a final blow than the sole cause. But before looking at any data, we need to know about a trap. Otherwise the data will mislead us.

Imagine a hundred species all dying out on the same day. The same day, with no sequence at all. Sixty-six million years later, we go digging for their fossils. What will we unearth? A curve that looks like “gradual decline.”

The reason is simple. Common species leave many fossils, so their last fossil will fall close to that day. Rare species leave few, and their last one may lie more than a hundred thousand years before it. Stack the hundred species’ last appearances by stratigraphic height, and a staircase thinning progressively upward emerges quite naturally. In reality, they all died on the same day.

A genuinely instantaneous extinction will grow a decline curve of its own in the fossil record. This is what the tool from Essay Eleven looks like in this setting.

Putting numbers to it makes the scale clearer. Suppose a species left, on average, only twenty identifiable occurrences per million years during its final interval. After an instantaneous extinction, its ninety-five-percent last-occurrence window could extend downward by about 150,000 years. At fifty occurrences per million years, that becomes about 60,000 years. At a hundred, about 30,000. These are Poisson-model calculations assuming approximately independent sampling at a constant rate, not universal confidence windows for every dinosaur.

So “the last dinosaur fossil occurs more than a hundred thousand years below the boundary,” taken on its own, tells us nothing.

And this is precisely what makes the dinosaur question so difficult. Marine planktonic microfossils can distinguish an extinction pattern lasting a thousand years from one lasting a hundred thousand. Large land-dwelling dinosaurs cannot.

Deep-sea and shelf carbonates can accumulate continuously, with abundant individual microfossils in every gram of sediment. Magnetostratigraphy, carbon isotopes, and the impact layer allow correlation across oceans. Terrestrial river deposits, by contrast, suffer scour, shifting channels, and brief gaps. Large vertebrates are rare to begin with, and a fossil locality often represents an average across a considerable span of time.

Yet dinosaurs are precisely the group everyone most wants to ask about. With that in mind, let us look at the data.

In 2016, researchers used time-calibrated dinosaur phylogenetic trees and birth–death models to estimate speciation and extinction rates. They found that, in several major dinosaur lineages, extinction rates gradually overtook speciation during the last tens of millions of years of the Cretaceous. The overall decline could be traced back more than twenty million years before the boundary.

In 2020, another group reanalyzed evolutionary rates on the same kind of phylogenetic trees. They specifically tested how sensitive the earlier result was to tree topology, branch lengths, time calibration, and model choice. Their conclusion was that there was no robust evidence for a sustained decline in the overall dinosaur diversification rate before the boundary.

In 2021, another study compiled about 1,600 fossil occurrence records worldwide, covering 247 species in six relatively well-preserved dinosaur families. A Bayesian fossil birth–death model directly estimated preservation, speciation, and extinction rates. It found extinction overtaking speciation after about 76 million years ago, earlier among herbivores and somewhat later among carnivores.

Three approaches. Three answers that do not agree.

None of these three conflicting results escapes the trap described above. The material still comes predominantly from Laurasia, especially North America, and the amount of fossil-bearing rock exposed in one region can turn directly into a proxy for “global diversification rate.” Work using paleoclimate and habitat-suitability models has found that North America’s low Maastrichtian species counts can be produced by a reduction in exposable fossil-bearing rock. They do not require a matching decline in actual ecological diversity.

Two examples make the weight of sampling especially clear.

Only a few pterosaur families had once been confirmed from the latest Cretaceous, making it easy to imagine them as “long past their prime.” But late Maastrichtian material from Morocco reported in 2018 raised the tally for roughly the final million years to at least three families and seven species, spanning a broad range of body sizes and functional ecologies. Rarefaction and sampling analyses did not support a significant loss of diversity from earlier intervals to the Maastrichtian.

Add more sampling, and those families “gradually disappearing” extend back up toward the boundary. Ammonites supply the other half of the picture. In 1996, an analysis of confidence intervals on the last occurrences of latest Cretaceous outer-shelf macrofossils from the western European Tethys divided the results into four categories. There was one major, abrupt extinction near the boundary.

A marine regression somewhat before the boundary may have been associated with the disappearance or changes in abundance of as many as about nine ammonite species. Roughly six more showed background extinction during the latest Cretaceous. Most inoceramid bivalves, meanwhile, had disappeared progressively well before the boundary.

The inoceramids are especially important because they provide a genuinely gradual comparison. The same statistical method, applied to the same succession of rocks, identifies some groups as victims of a boundary pulse and others as genuine losses before the boundary. This shows that the method does not “prove sudden extinction whatever record you give it.”

Planktonic foraminifera give us a third shape. In continuous marine sections with high sedimentation rates, many specialized, morphologically complex late Maastrichtian planktonic foraminifera persist until very close to the boundary, then undergo a major loss of species near the iridium-rich layer. Some large or ecologically specialized species, however, had already declined during earlier climate fluctuations.

On the marine side, then, the answer is an abrupt boundary change superimposed on earlier changes in ecological assemblages, rather than a uniform fading away over hundreds of thousands of years.

On land, one of the nearest approaches to a conclusion comes from trophic networks. A 2022 study built such networks from about 1,600 vertebrate occurrence records and more than 470 genera in western North America. It found that the influence of large herbivorous dinosaurs within trophic networks was declining in the latest Cretaceous, while medium-sized herbivorous and omnivorous dinosaurs were becoming more important. Yet the realized ecological niches of the large herbivorous dinosaurs themselves showed no corresponding instability.

A community being rearranged is not the same as an entire system heading for collapse. The binary choice between rise and decline is not enough here.

How far can we presently take the answer? There is substantial material supporting reorganization or decline in some lineages and some regions. There is no comparable evidence that all non-avian dinosaurs were already on an irreversible path to extinction by 66 million years ago.

VIII The Remainder Lies in the Counterfactual

Both protagonists of this essay are real. The impact is real. The crater is there, the ejecta are there, the iridium layer is there, and so is the iridium layer inside the crater. The Deccan is real too: the 500,000 square kilometers of basalt are there, the pre-boundary warming is there, and so are the lavas on both sides of the boundary.

They overlapped in time, and both had mechanisms capable of producing environmental stress. But the extinction happened only once. The two columns set out in Section VI are the whole conclusion of this essay: four observational constraints, four counterfactual questions.

The column of observational constraints can still be tightened with more sections, better dating, and more detailed models. Its history is the history of the past forty-odd years: from one iridium anomaly to hundreds of boundary sections; from one magnetic anomaly to drill cores from the peak ring; from “65 million years” to “somewhere around 66.04 million.” It really is tightening, and doing so beautifully.

The counterfactual column can also be narrowed by new evidence, but in a different way.

“Without the Deccan, would the impact still have caused an extinction on the same scale?” “Without the impact, would the Deccan have caused a mass extinction at that moment?” We cannot set beside this Earth a history with either event removed. The geological record contains only the history that happened. Yet mechanisms, models, and comparisons with other events can still constrain the alternatives. What this essay adds to the catalogue is the part of that inference which depends on assumptions and cannot be checked by directly replaying Earth: the remainder lies in the counterfactual.

Let us also note its place. This is the eleventh kind in the series’ catalogue of remainders. The first ten lay outside the evidence; in the coordinate axes; in the shape of the grid; in the sieve; in the order; in the explanation itself; in a ruler that does not exist; in a quantity with no unit; between two irreducible descriptions; and in the act of making the unmeasurable measurable. The eleventh lies in the counterfactual.

It offers a contrast with the kind in Essay Seventeen, so the two should be considered together. That earlier remainder grew along with our abilities. Only after color became measurable could we first ask how much diagenetic shrinkage had occurred, or in which taxon a mapping ceased to work. The stronger the capability, the more specific that kind of remainder became.

More sections, better dates, and finer climate models will not produce a second actual history of Earth. They can, however, rule out some counterfactual scenarios and narrow others. The inability to replay Earth directly is a principled limit; how tightly we can constrain alternative histories changes as methods improve. Confuse the two, and “we cannot replay it” becomes the mistaken claim that we can never learn more.

There is another rule worth setting down here, one that will outlast this essay’s particular subject.

When candidate causes overlap in time, finer resolution often helps rule out accounts without being enough to apportion their effects. The end-Cretaceous illustrates this: four decades of improved dating have eliminated many accounts, but have not supplied a model-independent percentage split between impact and volcanism. Models can estimate contributions; the estimates must travel with their assumptions.

Let us return to selectivity, because it is the easiest part of this essay to write badly. The casualty list has a structure. It was not drawn from a hat. A quantitative analysis of end-Cretaceous marine bivalves in 1995 found that genera with wide geographic ranges were significantly less likely to go extinct than narrowly distributed genera. This was a quantitative result spanning regions. It shows that catastrophe still obeyed some of the ordinary rules of extinction risk.

Yet on that same list, the widely distributed ammonites all went extinct. So did the mosasaurs, top marine predators of the late Maastrichtian; the pterosaurs, still represented by at least three families in the latest Cretaceous; the dominant large vertebrates in North American terrestrial ecosystems; and the enantiornithines, an important Cretaceous lineage of flying birds. Being common before the catastrophe was not enough to carry a group through an instantaneous global disturbance.

As for those that crossed the boundary—freshwater turtles, semiaquatic crocodyliforms, frogs and salamanders, small omnivorous and insectivorous mammals, a few bird lineages, fish in detrital food chains, ferns, and plants with dormant propagules—we must be precise here, in keeping with this series’ discipline.

They were not “superior,” nor were they “the best adapted.” They happened to possess certain life-history traits that provided a buffer under some combination of short-term darkness, food shortage, fire, or acidification. Before the catastrophe, those traits did not necessarily confer any ecological advantage.

A low metabolism, the ability to fast for long periods, the ability to use detritus already accumulated rather than that day’s living production: in a world with its normal sunlight, these are no extraordinary accomplishments.

There is no shift of level to identify in this essay. The unit of selection did not change across the boundary. This is not that kind of problem. What it offers instead is a view of what remains, irreducibly, when a causal question has been pursued to the end.

It takes the form of a column of observational constraints beside a column of counterfactuals. Both can improve. The second must always show which assumptions made the next step possible.

The thin clay layer is still there. It has left us a day, not ready-made answers to every “what if.”