故事 · 高尔顿和他的「豆子机」
Origin Story · Galton and the Bean Machine
19 世纪的英国博学家 弗朗西斯·高尔顿（Francis Galton）痴迷于测量一切 —— 身高、指纹、甚至祈祷的效果。
他想给人讲清楚「为什么这么多东西都呈钟形」，于是造了一台木盒装置：顶部一个入口，里面钉满交错的小钉子，底部排着一列收集格。
倒进一把豆子，每颗豆子碰到钉子都随机偏左或偏右，落到底部完全靠运气。可当几百颗豆子落定，
它们无一例外地堆成一座对称的小山 —— 正态曲线。高尔顿用这台机器告诉所有人：
钟形不是被谁画出来的，而是无数次独立随机选择累加后的必然结果。后人把它叫做高尔顿板，也叫 bean machine（豆子机）。
The 19th-century English polymath Francis Galton was obsessed with measuring everything — heights, fingerprints, even the efficacy of prayer.
To show people why so many things come out shaped like a bell, he built a wooden box: a single inlet at the top, a staggered field of pegs inside, and a row of collecting bins along the bottom.
Pour in a handful of beans; each one kicks randomly left or right at every peg and lands wherever luck takes it. Yet once a few hundred beans have settled,
they always pile up into a symmetric little mountain — a normal curve. Galton's machine made one thing visible to everyone:
the bell isn't drawn by a designer — it's the inevitable result of summing countless independent random choices. Posterity calls the device a Galton board, or Quincunx, or simply the bean machine.
1 放球：随机左右弹，底部堆成钟形 Drop the Beads: Random Left/Right Bounces Pile Into a Bell
点「放球」开始hit "Drop beads" to start每颗球在每排钉子上五五开地选左或选右。单看一颗，落点全凭运气； 可一旦累积起来，底部的直方图就自动逼近正态曲线（白线）。多放几波，看它越来越像钟形。 At every peg each bead picks left or right with a 50/50 coin flip. One bead is pure luck; accumulate them and the bottom histogram automatically converges onto the normal curve (the white line). Drop a few more batches and watch the bell tighten.
2 高尔顿板告诉我们的事 What the Galton Board Is Really Saying
正态的起源：正态曲线不是凭空规定的，而是大量随机小事件叠加后「涌现」出来的形状。这台机器就是它的出生现场。
Where the bell comes from: the normal curve isn't an arbitrary shape — it emerges when many small random events stack up. This machine is its birthplace.
中心极限定理的直觉：把每排「左/右」看作一次独立随机变量，n 排之和的分布趋于正态。CLT 抽象，高尔顿板把它变成看得见的豆子山。
Central limit theorem, made tangible: treat each left/right bounce as an independent random variable; the sum across n rows converges to a normal. The CLT is abstract — the Galton board turns it into a pile of beads you can see.
随机误差的叠加：一个零件的最终尺寸 = 设计值 + 机床抖动 + 材料波动 + 温度漂移……众多小误差相加，结果自然鼓成钟形。
Stacking random errors: a finished part's dimension = nominal + machine jitter + material variation + thermal drift + … . Many small errors summed together swell naturally into a bell.
二项 → 正态：每排左右各 50%，落格服从二项分布；当层数 n 变大，二项分布越来越像正态。这正是棣莫弗-拉普拉斯定理（往后会讲）。
Binomial → normal: with 50/50 splits per row, the bin counts follow a binomial distribution. As n grows, the binomial increasingly resembles the normal — this is the de Moivre–Laplace theorem (we'll get to it later).
一句话In One Line
高尔顿板回答了一个根本问题：为什么正态分布无处不在？
答案是只要一个结果由许多独立、微小、随机的因素累加而成，无论每个因素本身长什么样，它们的总和都会趋向正态。
这就是中心极限定理，也是六西格玛能放心地拿正态去建模制程波动、误差、产出的底层依据 ——
因为制程中的波动，本质就是无数小扰动的叠加。
但也要警惕它的前提：因素得「独立」且「足够多」。若存在一个主导性的大因素，或因素之间强相关，结果就不再是漂亮的钟形了。
The Galton board answers a fundamental question: why is the normal distribution everywhere?
Whenever an outcome is the sum of many independent, tiny, random factors, the total tends toward a normal — regardless of how each individual factor is distributed.
That's the central limit theorem, and it's exactly why Six Sigma can comfortably model process jitter, error, and yield with a normal distribution —
because process variation is, at heart, the sum of countless small perturbations.
Just respect the preconditions: the factors must be independent and there must be enough of them. If one factor dominates, or the factors are strongly correlated, the result is no longer a clean bell.
常见误用Common Mistakes
以为「随机」就一定均匀，看到中间高两边低觉得「不随机」。单步五五开，但多步累加后中间路径最多，所以中间最高，这才是随机叠加的真相。
Assuming "random" must mean "uniform" and concluding that a tall middle and short tails means "not random". Each single step is 50/50, but across many steps the most paths lead to the middle — that's what truly random summing looks like.
把高尔顿板当成「任何分布都会变正态」。它演示的是「独立随机变量之和」趋于正态，前提是因素独立且数量足够多。
Reading the Galton board as "any distribution becomes normal". It demonstrates that the sum of independent random variables tends to normal — and only when the factors are independent and there are enough of them.
用一颗球的落点去预测下一颗。单次结果不可预测，正态只在大量样本的整体分布上才显现。
Using one bead's landing spot to predict the next one. A single outcome is unpredictable. The normal shape only shows up in the aggregate distribution over many samples.