一条老街，每年都有一两家店关门，又有新的开进来。这是常态，你不会觉得那年出了什么事。但如果某一年，一半的门脸同时空了，那是另一回事——你会想知道那年到底发生了什么。
化石记录里也是这样。物种一直在消失，大多数时候是涓涓细流。真正想抓的，是那些「一半门脸同时空掉」的年份。
麻烦从第一步就开始：你怎么知道某一段真的异常，而不是你那阵子恰好多走了几趟？这一步没做干净，后面所有解释都建在沙子上。
先说这份数据长什么样。化石能提供的其实是一张出勤表：每一类生物最早出现在哪一层岩石里，最后出现在哪一层。把「最后一次出现」落在某一段时间里的类群数一数，除以那段时间在场的总数，就是这一段的灭绝率。它是数出来的一个比例，不是从石头里直接看见的惨状。
1982 年，两位古生物学家把海洋化石的出勤表统一整理了一遍，照这个算法数下来，几个高峰跳了出来：奥陶纪末、泥盆纪晚期、二叠纪末、三叠纪末、白垩纪末。「五次大灭绝」就是从这儿来的。他们的原文其实比后来的转述谨慎：按当时的数据，泥盆纪那次并没到统计上说得响的程度。
四十年后有人回头重查了一遍，结论更削：按现在这批数据，这五次并不比背景速率显著地高。它们确实是早奥陶纪以来最大的五次，但「大灭绝」和「平常」之间没有一条自然的界线，线是我们划的。这不等于那些事没发生，是说别把它们想成两类东西。
连数字也在缩水。二叠纪末常被引成「96% 的海洋物种消失」。2016 年有人指出那个数把中二叠世和晚二叠世两次事件并在一起算了；把那段时间本来就会照常发生的背景灭绝先扣掉，剩下大约 81%。仍然是有史以来最狠的一次，但这个做法值得单独记住：要说一次事件有多大，你得先知道「平时」有多大，再把它减掉。不减，你量到的是事件加上日常噪声。
还有一个更别扭的偏差。一个物种最后一块化石所在的位置，几乎不可能正好是最后一个个体死掉的时刻——化石本来就稀疏，越少见的物种，「最后一次出现」被记早得越多。于是一场很短时间里发生的灭绝，落进化石记录会被抹成一道缓坡。所以「这次是突然的还是拖了几十万年」，光看形状答不了。寒武纪大爆发那期是同一类偏差的另一个方向：那边是记录突然开了灯，这边是记录把一堵墙涂成了斜坡。
EN What the fossil record actually gives you is an attendance sheet: for each group, the oldest rock layer it appears in and the youngest. Count the groups whose last appearance falls inside a given interval, divide by how many were present, and that is the extinction rate for that interval — a counted proportion, not a scene of carnage read off the stone. In 1982 two palaeontologists compiled the marine fossil record and ran that count. Peaks jumped out: end-Ordovician, late Devonian, end-Permian, end-Triassic, end-Cretaceous. That is where the "Big Five" comes from. Their own paper was more cautious than the retellings: four stood clearly above background, while the Devonian one did not reach statistical significance in those data. Forty years on, a review that went back to check concluded something sharper still — on current data the Big Five are not statistically distinct from background rates. They really are the five largest since the early Ordovician, but there is no natural line between "mass extinction" and "ordinary"; the line is one we drew for convenience. The numbers have shrunk too. The end-Permian is often quoted as killing 96% of marine species. A 2016 analysis pointed out that this figure merged two separate events, the mid- and late-Permian crises; subtract the background extinction that would have happened anyway across that interval, and about 81% is left. Still the worst on record — and the method is worth remembering on its own: to say how big an event was, you must first know how big "ordinary" is, and take it out. Skip that and you measure the event plus the everyday noise. One more awkward bias: the layer holding a species' last fossil is almost never the moment its last individual died. Fossils are sparse, and the rarer the species, the earlier its last appearance gets recorded. So an extinction that really happened fast is smeared in the record into a gentle ramp — a thinning first, then nothing. Which means the question "was this sudden or did it take a few hundred thousand years?" cannot be answered from the shape in the rock alone. The piece on the Cambrian explosion is the same bias running the other way: there the record switched its light on; here the record paints a wall into a slope.
1980 年，几位研究者盯上了意大利古比奥一段石灰岩：白垩纪和古近纪交界处那薄薄一层黏土。他们测的是铱，一种在地壳表层很少、在小行星里相对多的金属。那层黏土里的铱高得离谱。
这本身只是「一个异常」。异常在科学里其实不值钱，随便什么都能给它编一个解释。真正厉害的是为解释它而提出的那个说法——一颗大石头砸了地球——立刻把自己架到了火上：如果这是真的，有几样东西必须存在，找不到就算它错。
三条：这层异常不能只在意大利有，全世界同一层都该有；界线层里该找得到被冲击波压出特定纹路的石英，和熔化的岩石飞出去又落回来凝成的玻璃小球；地球上还得有一个那个年代、那个尺寸的坑。
然后是十年。前两条陆续兑现；第三条最难，因为六千多万年足够把一个坑埋得看不见。1991 年，几位研究者把墨西哥尤卡坦半岛地下一个直径约 180 公里的圆形结构认成了它——地面上什么也看不出来，靠的是重力和磁场的异常，加上石油钻井取上来的岩芯。后来的精确测年把它落在约 6600 万年前，跟界线层对得上，误差只有一两万年。
2010 年，四十多位不同领域的作者合写了一篇综述，结论是这次撞击触发了那场灭绝。但不是全票：同一时期，印度德干高原正喷发一场持续几十万年的巨型火山，放出的气体足以把气候和海水搅乱，这一派至今没被说服，两组人在 2019 年用不同测年方法给出的结论也还打架。目前多数人的读法是撞击那一击最关键、火山把底子搞坏了，但两边各占多少还没定。
这一段真正想留下的不是恐龙，是那个动作：一个说法变硬，不是因为它把已知的事解释得更顺口，而是因为它主动说出了「我要是错了，你去哪儿能抓到我」。
EN In 1980 a group of researchers focused on a thin clay layer at Gubbio in Italy, right at the Cretaceous–Palaeogene boundary. They measured iridium — a metal that is scarce in the Earth's crust but comparatively abundant in asteroids and meteorites — and found it absurdly enriched. On its own that is just an anomaly, and anomalies are cheap; you can improvise an explanation for anything. What made this one matter is that the explanation offered for it — a large rock hit the Earth — immediately put itself on the line. If it were true, certain things had to exist, and failing to find them would count against it. First, the anomaly could not be confined to Italy; the same layer worldwide should carry it. Second, the shock wave would press distinctive internal patterns into quartz crystals, and rock melted and thrown up would fall back as small glass spherules; both should turn up in the boundary layer. Third, there had to be a crater somewhere of that age and that size. Then came a decade. The first two were met. The third was hardest, because sixty-six million years is enough to bury a crater out of sight. In 1991 researchers identified a circular structure roughly 180 kilometres across buried under the Yucatán Peninsula in Mexico — invisible at the surface, recognised from gravity and magnetic anomalies plus cores from oil drilling. Later precise dating puts it around 66 million years ago, matching the boundary layer to within a few tens of thousands of years. In 2010 more than forty authors from different fields wrote a review concluding that the impact triggered the extinction. It was not unanimous. Over the same interval the Deccan Traps in India were erupting on a scale, and over hundreds of thousands of years, sufficient to disturb climate and oceans, and that camp remains unconvinced. In 2019 two teams using two different dating methods reached conflicting answers: one had the largest pulses preceding the impact, the other had more than three quarters of the lava erupting some 600,000 years after the boundary. The common reading now is that the impact delivered the blow while the volcanism degraded the baseline, with the proportions still unsettled. What is worth keeping here is not the dinosaurs but the move: a claim hardens not by explaining known facts more smoothly, but by stating in advance where you should be able to catch it if it is wrong.
「适者生存」在这里最容易被用错：你会以为大灭绝就是把不够强的筛掉，剩下更能打的。数据说的不是这样。
有一批对白垩纪末海洋贝类的研究，把平时和那一刻分开来看。平时——也就是背景灭绝的年头——一个物种能不能撑住，跟几件事有关：幼体能不能在水层里漂很远因而铺得开，这个物种分布多宽，它所在的属下面养着多少个种。这几条在平常年份实实在在地起作用。
到了白垩纪末那一刻，这几条统统失灵。唯一还预测得动存活的，是整个属摊开的地理范围有多大——哪怕它下面每个种的分布都很窄，只要加起来铺在够大的地图上，这个属就更可能有一支活下来。这个模式在多个类群、多次大灭绝里都出现过。
把这句话说明白：平时的优秀在那一刻不折算。不是「更强的活下来了」，也不是纯抽签——是筛子换了一把，新筛子对着的那个维度，平时几乎不被选择直接管。一个属铺得广，多半出于别的原因，跟「将来会有一颗小行星」毫无关系。
撞击之后的鸟把这件事讲得更具体。界线层的花粉记录显示全球森林在那之后崩了——蕨类的孢子暴涨，那是植被从最简单的一步重新开始的信号。2018 年有人把这条环境线索和鸟类的家谱并排放，结果非常干脆：所有树栖的鸟类支系都没留下后代，活到今天的鸟，祖先都是在地面上讨生活的。今天你抬头看见满树的鸟，是灾后从地面重新爬回树上的结果。
所以别把幸存名单读成优秀名单：没有谁预见到那一天，也没有谁为它准备。之后几千万年哺乳动物铺开，同样不是因为哺乳动物更高级——是位置空了，而当时还在场的名单里有它们。
EN "Survival of the fittest" is easiest to misuse right here. You would assume a mass extinction culls the weaker and leaves the tougher. The data say otherwise. Work comparing Late Cretaceous marine molluscs separates ordinary times from the moment itself. In background times, whether a species held on depended on things like whether its larvae could drift far through the water column and so spread it out, how wide the species' own range was, and how many species its genus contained. Those factors genuinely mattered in ordinary years. At the end-Cretaceous moment, all of them stopped working. The one thing that still predicted survival was the geographic spread of the entire genus — even if every species within it was narrowly distributed, a genus whose species collectively covered a large enough map was more likely to get a branch through. The pattern recurs across several groups and several mass extinctions. Stated plainly: ordinary-times excellence does not convert at that moment. It is not that the stronger survived, and not a pure lottery either. The sieve was swapped, and the dimension the new sieve tests is one that selection barely acts on directly in ordinary times. A genus is widespread for other reasons, none of them having anything to do with an asteroid arriving later. The birds after the impact make this concrete. Pollen records across the boundary show global forests collapsed — a spike in fern spores, the signal of vegetation restarting from its simplest stage. In 2018 researchers laid that environmental line alongside the bird family tree and found the outcome blunt: no tree-dwelling bird lineage left descendants. Every bird alive today descends from ancestors making their living on the ground. The birds filling the trees now are the result of climbing back up after the catastrophe. So do not read a survivor list as an excellence list. Nobody foresaw that day and nobody prepared for it; the sieve of that moment happened to test certain traits, and individuals that happened to carry them left more offspring. The mammal radiation over the following tens of millions of years is not a matter of mammals being more advanced either — the seats emptied, and mammals were among those still in the room.