公园草丛里露出一截绳子，孩子先跳开，再回头看清楚。回到家，插座就在墙上，插进去真会出事，你提醒过一百遍，他照样往里塞东西。
绳子从没害过他，插座天天在那儿。按「吃过亏才会怕」这个说法，两件事该反过来才对。
于是有人接一句：怕蛇是刻在我们身上的，插座太新了来不及刻。这句话听着很顺，而且里面大概率有一部分是真的。麻烦在于，它顺得不需要任何证据也能说出口——一门靠这种句子吃饭的学科，很难跟一个会讲故事的人区分开。
桌上四张卡片，每张一面印字母、一面印数字。你现在看到的四个面是：E、K、4、7。有人告诉你一条规则：一面是元音字母，那另一面必须是偶数。要检查这条规则有没有被破坏，你最少得翻哪几张？
多数人翻 E，再翻 4。正确答案是 E 和 7。E 的背面要是奇数，规则就破了；7 的背面要是元音，规则也破了。翻 4 什么都说明不了——规则没说偶数背面一定得是元音。这道题从 1960 年代起被反复做，做对的人常年在一成到两成半之间，结果很稳。
换个说法再来一次。你是酒吧的服务生，店里的规矩是：喝含酒精的饮料，必须满 18 岁。四位客人，你手上各有一张卡，一面写他在喝什么，一面写他多大。你看到的四个面是：在喝啤酒、在喝可乐、25 岁、16 岁。你该去查哪两位？
几乎人人当场答对：查喝啤酒那位多大，查 16 岁那位在喝什么。不用纸笔，不用想。
这个落差本身没人否认。它至少说明一件事：脑子不是一台通用的推理机器，题目骨架不动、内容换掉，输出质量会变。
1989 年 Leda Cosmides 把这个落差解释成：我们身上有一套专门盯着「谁拿了好处却没付代价」的处理方式。这里得把话说准。所谓「专用回路」，不是脑子里有个抽屉写着「作弊」，而是碰上某一类内容时，跑起来的不是同一套程序。也别说「演化让我们长出了作弊检测器」——账不是这么记的：在祖先那种反复拿好处换好处的处境里，如果认得出谁白拿了的个体确实留下了更多后代，这套处理方式占的份额就会一代代升上去。是不是真这么回事，得另外查。
争论也正在这儿。落差是真的，「落差是因为社会契约」只是其中一种解释。有人把功劳记在「这个说法你本来就熟」上；还有人发现带警告性质的规则（「进车间必须戴安全帽」）一样能把成绩拉起来，可那里面并没有谁占谁便宜。这个到今天还在争。
EN Four cards on a table, each with a letter on one side and a number on the other. You see E, K, 4, 7. The rule: if one side is a vowel, the other must be even. Which cards must you turn? Most people pick E and 4. The answer is E and 7. Turning 4 proves nothing — the rule never said even numbers must have vowels. Since the 1960s, only about 10 to 25 per cent get this right. Now change the wording. You are a bartender; the rule is that anyone drinking alcohol must be over 18. The four cards read: drinking beer, drinking cola, 25 years old, 16 years old. Almost everyone instantly checks the beer drinker's age and what the 16-year-old is drinking. The logical skeleton is identical; only the content changed, and scores jump to roughly 65 to 85 per cent. So the mind is not a general-purpose reasoning engine: swap the content, keep the structure, and output quality shifts. In 1989 Leda Cosmides read that gap as evidence of machinery tuned to spotting people who took a benefit without paying the cost. Be precise about what that claims. Not a drawer in the head labelled "cheating", but a different routine running on a certain kind of content. And not "evolution gave us a cheater detector" — the ledger does not work that way. What it would mean is that among ancestors who repeatedly traded favours, individuals who noticed the free riders left more descendants, so that way of processing grew as a share of each generation. Whether that is what happened is a separate question. The gap is not disputed; the social-contract reading of it is only one candidate. Others credit familiarity, and rules about hazards — hard hats in a workshop — lift scores too, with nobody cheating anyone. This is still argued about.
1979 年，Gould 和 Lewontin 借吉卜林一本童话书的名字，管这类解释叫「原来如此的故事」——书里讲大象的长鼻子是被鳄鱼拽出来的，讲得有鼻子有眼，你还真挑不出它哪句错。（适应与约束那期讲拱肩，用的就是这篇文章。）
举个真的。2007 年有一项研究，让两百来人在成对的颜色里连着挑，结果女性的平均偏好偏向偏红的那一端，男性偏向蓝绿，这个差异在统计上相当结实。研究之外，媒体上接的解释是：祖先分工里女性采集浆果，红色在绿叶背景上最跳。
这个解释的毛病不在于它荒唐，在于它安全得过分。假如数据反过来——女性偏蓝绿——现成的故事一样等着：水源、天色、平安的信号。一个不管结果朝哪边都能讲的解释，讲出来等于没讲。
还有一层更直接的麻烦。1918 年，美国一本童装行业刊物白纸黑字写着：通行的规矩是男孩穿粉、女孩穿蓝，理由是粉色更强烈更有决断，蓝色更娇嫩秀气。要到 1940 年代之后，今天这一套才固定下来。
别把这一条用过头。它推不出「颜色偏好里没有任何生物学成分」——那项研究量的是对色调的偏好，不是「粉色属于女孩」这个符号，两码事。它推出的只有一件事：那个采浆果的故事，套在一个几十年前还是反过来的现象上也一样顺——说明它根本不挑事实。
反过来也一样。「这是个事后编的故事」这句指控本身，可能同样便宜。你说一个解释是编的，就得把话说到位：它哪一条预测会输？答不上来，这句指控和它要批评的东西是一路货。
EN In 1979 Gould and Lewontin borrowed the title of a Kipling children's book and called this kind of account a "just-so story" — the book explains that the elephant got its trunk when a crocodile pulled on it, and you cannot point to a false sentence anywhere in it. A real example: a 2007 study had a couple of hundred people choose repeatedly between pairs of colours. Women's average preference leaned toward the reddish end, men's toward blue-green, and the difference was statistically solid. The explanation that travelled through the press was that ancestral women gathered berries, and red stands out against green leaves. The trouble is not that this is absurd; it is that it is far too safe. Had the data run the other way, a story was waiting: water, sky, signals of safety. An account that works whichever way the numbers fall has told you nothing. There is a blunter problem too. In 1918 an American children's-clothing trade journal stated plainly that the accepted rule was pink for boys and blue for girls, pink being the stronger and more decided colour and blue the daintier one. Today's convention only settled after the 1940s. Do not overplay this. It does not show that colour preference has no biological component — the study measured hue preference, not the cultural sign that pink belongs to girls. It shows one thing: the berry story fits just as smoothly onto a pattern that was reversed a few decades ago, which means it is not constrained by facts at all. The reverse holds as well. "That's a just-so story" can be just as cheap an accusation. If you say an account was made up after the fact, finish the sentence: which of its predictions could lose? If you cannot answer, your objection is made of the same material as its target.
1891 年，芬兰学者 Edvard Westermarck 提出了一个跟当时主流相反的说法。主流是：人天生想跟亲人交配，所以社会得立禁忌压住。他说反了——是先没有这个念头，禁忌才立得稳；而让念头消失的不是血缘本身，是小时候天天挤在一起。
这不只是换个说法。它伸出了一样可以被打断的东西：找两个没有血缘、但从小一起长大的人，他们成年后要是照样互相吸引，这个说法就完了。
刚好有两组数据能查，而且当初记录它们的人根本不为这个假说服务。
一组在以色列的集体农庄。那里的孩子按年龄编成同伴组，从小一起吃住长大，彼此没有血缘。后来统计下来，近三千桩婚姻里只有十四桩发生在同一个同伴组内部，而这十四桩里，没有一桩是在头六年一起养大的。
另一组在台湾。当地旧时有童养媳：女孩很小被抱进未来丈夫家，和他一起长大，成年后成婚。人类学家 Arthur Wolf 翻了日治时期留下的户籍册子，把这类婚姻和普通婚姻比——离婚更多，生下的孩子更少。
这两组数据的分量，不在于它们「支持」了什么，在于它们本来完全可以把这个说法打死。
争论还是得说清楚。集体农庄那边的数字被质疑过：同伴组内部本来就另有社会压力不让成婚。童养媳这边最硬的替代解释是，这种婚姻本就被看作穷人家的将就、受人轻视，过得差并不奇怪；Wolf 做过控制，但没有定案。
把标准又抬了一级的，是 2007 年发在《自然》上的一项研究。它的问法更狠：假如真有这么一套机制在估「这个人跟我有多近」，那它总得靠某个线索来估，而线索是可以去量的。研究去问人们小时候和手足同住了多少年，发现同住的年头越长，一个人对自己跟手足发生性关系的反感越强，对别人这么做的道德义愤也越强。而且对先出生的那个孩子来说，「有没有亲眼见过母亲照顾这个婴儿」是比同住年数更强的线索。
严格检验长的就是这个样子：不是回头把已经知道的事再解释一遍，是先说出某个量该往哪边动、动多少，然后去量。
EN In 1891 the Finnish scholar Edvard Westermarck proposed the opposite of the going view. The going view was that people naturally desire relatives, so society must impose a taboo. He reversed it: the desire is absent first, which is what lets the taboo hold — and what removes it is not shared blood but growing up in daily physical proximity. That is not merely a rephrasing. It sticks its neck out: find two unrelated people raised together from early childhood, and if they are still drawn to each other as adults, the claim is finished. Two bodies of data happened to be available, recorded by people with no stake in the hypothesis. One comes from Israeli collective farms, where children were raised in age-based peer groups, eating and sleeping together, unrelated. Of nearly three thousand marriages, only fourteen were within a peer group, and none of those fourteen involved people raised together through the first six years. The other comes from Taiwan, where a girl might be taken into her future husband's household very young, raised alongside him, and married to him as an adult. The anthropologist Arthur Wolf worked through household registers kept during the Japanese colonial period and compared these marriages with ordinary ones: more divorce, fewer children. What gives these numbers their weight is not that they support something, but that they could have killed the claim outright. The disputes are real. The kibbutz figures have been questioned, since peer groups carried other social pressures against marrying within them. For the Taiwanese case the strongest rival reading is that such marriages were seen as a poor family's makeshift and were looked down on, so unhappy outcomes are unsurprising; Wolf tried to control for this, and it is not settled. A 2007 paper in Nature raised the bar again with a sharper question: if machinery really estimates how closely related someone is, it must use some cue, and cues can be measured. Asking people how many years they had lived with a sibling, the study found that longer coresidence went with stronger personal aversion to sex with that sibling and stronger moral objection to others doing the same — and that for the elder child, having seen their mother care for the newborn was a stronger cue than years of coresidence. That is what rigorous testing looks like: not re-explaining what is already known, but stating in advance which way a quantity should move, and by how much, then going out to measure it.