Insight
形跟随兴趣
从一句建筑格言，到 POI，到商圈
建筑界有一句老话：form follows function，形跟随功能。Sullivan 在 1896 年写下的，后来成了现代主义的信条。意思很直白——你先知道这栋楼要干什么，形自然就长出来了。
但这句老话有一个漏洞。function 是图纸上的声明：这层是办公，那层是机房。可交付之后呢？办公室改成了仓储，机房改成了直播间。形从来没有真正追上过功能，它追的是别的东西。
动作即是形
2012 年，耶鲁的 Keller Easterling 写了一本小册子，标题很长：《The Action Is the Form: Victor Hugo's TED Talk》。她把 McLuhan 那句 the medium is the message 搬到了空间上：形不是装动作的容器，形就是动作被记录下来的样子。
走进一个商场，你会看到哪面墙被磨得发亮，哪个转角永远有人停，哪段走廊没人走。这些都是证据，不是设计。空间是动作留下的痕迹。
这句话比 Sullivan 那句更狠，因为主语换了。但它也留下一个问题：动作那么多，为什么被复制的总是其中某几个？
我们的下一步：形跟随兴趣
动作不会自己复制。复制是被选择的，而选择者是这个位置上的 interest。
这里要停一下。interest 这个词的好处，只有在英文里才成立。它在英文里同时是三个东西：
- 兴趣、注意力——人往哪看、往哪去；
- 利害——谁在这里有份，谁会去守它；
- 利息——钱的时间价格，租金、回报、资本化率。
中文没有一个词能同时装下这三样。写成"兴趣"，丢掉了利害和利息；写成"利益"，丢掉了注意力。所以这一篇里我保留 interest 原词：三个意思共用同一个词，本身就是这个论点的一部分——注意力、利害、资金价格，在底层是同一个量，只是在不同的时间尺度上表现出来。
注意力决定人流，利害决定谁留下来，利息决定谁付得起。
于是"形跟随兴趣"不是一句漂亮话。它把"形是被什么生成的"从设计意图、从实际动作，推到了需求分布这一层。
POI 是兴趣的计数器
POI，point of interest，兴趣点。我们天天在画它：做选址、做商圈、做竞品。但很少有人认真对待它的字面意思。
它的单位不是地点，是被归类过的兴趣。一个坐标、一个类别、几项属性——一条兴趣被固化下来，于是可以被数。
如果形跟随兴趣，那 POI 就不是地图上的装饰点，而是生成层的采样。我们其实一直在给"兴趣"做人口普查，只是没这么叫它。
POI 的第一个字母是 point，不是 place。
商圈，是兴趣在距离上的分布
如果 POI 是兴趣的一次采样，商圈就是兴趣在空间里怎么衰减的一张图。而这件事，一百年前就有人开始算了。
1931 年，Reilly 提出零售引力定律：两个商业中心抢同一片腹地，分界点落在哪里？他把零售中心的吸引力类比成质量，放在分子；把距离的平方放在分母；两边引力相等的地方，就是分界线。这是商圈切分最简单的近似，也是后来所有模型的思想源头。
1963 年，Huff 把它从"硬边界"改成了"概率面"：
A 是吸引力，通常用营业面积、店铺数、停车位合成；D 是距离；α 是吸引力参数，β 是距离衰减参数。一个顾客去哪家店，不再是一个"属于哪个商圈"的判断题，而是一个概率。把概率面按等值线切开，再按人口加权，就得到市场画像和营收估算；如果手上有真实客流，还能反过来标定 α 和 β，做开店、关店的推演。
这两个参数是这套语言里最值得琢磨的地方。α 说的是"规模有多重要"，β 说的是"距离有多快把兴趣磨掉"。同一个城市里，β 大，说明人懒，就近解决；β 小，说明人愿意为了一件事跨区。它们不是技术参数，是一地生活方式的两个刻度。
所以到了这一步，"形跟随兴趣"第一次有了可以算的形式：形沿着这张概率面最高的地方生长，而概率面的形状，由 α 和 β 决定。
再往后是等时圈。把概率面换成时间面——三十分钟能到的地方，就是今天多数人心里的"商圈"。等时圈不预测概率，它直接画出一个人愿意为兴趣付出的时间预算。
单点数不出组合
但抱着一堆概率，仍然会犯错。
数一个片区有多少家咖啡，说不清这些咖啡挨着什么。前者是边际分布，后者才是联合分布。Jacobs 在 1961 年就写过，城市的多样性来自四个条件的组合，缺一即废。六十多年过去，我们的报表还在数"多少家咖啡"。
Huff 模型也有同样的盲区：它默认各家店在竞争里彼此独立，谁离得近、谁体量大，谁就赢。可真实世界里，一家咖啡是因为隔壁有书店才活得好，一家健身房是因为楼下有超市才有人顺路。店与店之间的关系，不是简单的此消彼长。
GIS 里其实有现成的方法处理这件事：spatial co-location pattern mining。它不数单类，数的是"哪几类总是一起出现"。用邻近邻域替代购物篮，用 participation index 替代频次。
密度不是变量，组合才是变量。
留一把尺
最后还有一个问题：是不是所有被验证的兴趣，都值得被复制？
Illich 在 1973 年的《Tools for Conviviality》里给过一个区分。有一种垄断不是品牌的垄断，而是类型的垄断——汽车不是被哪家公司垄断，它是把步行和骑车从城市里请了出去。他管这叫 radical monopoly。
所以真正要问的不是"这个业态火不火"，而是：它扩散之后，这个地方变强了还是变弱了？是更能自己长出下一个东西，还是从此离不开它？
回到那条街
一条街不是被设计一次的。它是兴趣结晶过一次、并且持续结晶的地方。
设计者的工作不是发明兴趣，是让这个位置接得住兴趣：留下旧的、压住租金、打通步行、允许混杂、不急着统一立面。
剩下的，交给兴趣自己去找它要的形。
方法说明：本文的框架来自公开文献与通行方法——Reilly 的零售引力定律、Huff 的概率模型、Jacobs 的城市多样性条件、Illich 的工具判据，以及空间共位分析。文中不引用任何具体项目数据。
Form Follows Interest
From an architectural maxim to the point of interest — and to the trade area
There is an old line in architecture: form follows function. Sullivan wrote it in 1896 and modernism turned it into doctrine. The meaning is plain enough — know what a building is for, and its shape will follow.
The line has a hole in it. Function is a declaration made on a drawing: this floor is office, that one is plant. But after handover? The office becomes storage; the plant room becomes a livestream studio. Form never quite caught up with function. It was following something else.
The action is the form
In 2012, Keller Easterling, who teaches at Yale, published a short book with a long title: The Action Is the Form: Victor Hugo's TED Talk. She took McLuhan's the medium is the message and moved it into space. Form is not a container for action; form is what action looks like once it has been recorded.
Walk into a mall and you can see it. Which wall has been worn shiny. Which corner always has someone standing in it. Which corridor nobody uses. These are evidence, not design. Space is the residue of what people did in it.
It is a harder claim than Sullivan's, because the subject has changed. But it leaves a question open: if there is so much action, why are only a few patterns ever copied?
The next step: form follows interest
Action does not replicate on its own. Replication is selected — and the selector is the interest held at that location.
The word is doing a lot of work. It carries three meanings at once, and all three happen to be variables of commercial space:
- Interest as attention — where people look, where they go;
- Interest as stake — who holds a position here, and will defend it;
- Interest as the price of money — rent, return, capitalisation rate.
Attention decides footfall. Stake decides who stays. The price of money decides who can afford to. Three meanings, one word: underneath, they are the same quantity seen at different time scales.
So form follows interest is not a slogan. It moves the question of what generates form — past declared intent, past observed action — down to the distribution of demand.
A POI is a counter of interest
POI stands for point of interest. We draw them constantly — for site selection, catchment work, competitor mapping — and we rarely take the name seriously.
Its unit is not a place. It is an interest that has been classified: a coordinate, a category, a handful of attributes. An interest, fixed in place, and therefore countable.
If form follows interest, then a POI is not decoration on a map. It is a sample of the generative layer. We have been running a census of interest for years without calling it that.
The first letter of POI is point, not place.
A trade area is interest, distributed over distance
If a POI is a single sample of interest, a trade area is the map of how that interest decays across space. People have been computing it for a century.
In 1931, Reilly proposed a law of retail gravitation. Two centres compete for the same hinterland; where does the dividing line fall? He treated each centre's attraction as mass in the numerator and the square of distance as the denominator, and put the boundary where the two pulls are equal. It is the simplest approximation of trade-area division, and the source of everything that followed.
In 1963, Huff turned the hard boundary into a probability surface:
A is attraction, usually built from floor area, store count, parking; D is distance; α is the attraction parameter and β the distance-decay parameter. Where a customer goes is no longer a question of which catchment they belong to, but a probability. Slice the surface into contours, weight it by population, and you have a market profile and a revenue estimate. With real customer-origin data, you can calibrate α and β and run openings and closures as scenarios.
The two parameters are the interesting part. α says how much size matters; β says how fast distance wears interest away. In a city with a high β, people solve things nearby; with a low β, they will cross the city for one good reason. These are not technical settings. They are two dials describing how a place lives.
Which is where form follows interest first becomes computable: form grows along the highest ground of that probability surface, and the shape of the surface is set by α and β.
Then comes the isochrone. Swap the probability surface for a surface of time — what you can reach in thirty minutes is what most people now call a trade area. An isochrone does not predict probability; it draws the time budget a person is willing to spend on interest.
Points cannot count combinations
But a stack of probabilities still gets things wrong.
Ask how many cafés a district has, and you still do not know what those cafés sit next to. The first is a marginal distribution; the second is a joint distribution. Jacobs made the argument in 1961: urban diversity comes from four conditions in combination, and the absence of any one frustrates the rest. Sixty years on, our spreadsheets still count cafés.
Huff has the same blind spot: it treats stores as independent competitors, where the nearest and the largest win. In reality a café does well because there is a bookshop next door; a gym fills because there is a supermarket underneath. The relationship between shops is not simply zero-sum.
GIS already has the method for this: spatial co-location pattern mining. It does not count categories; it counts which categories are always found together. Proximity neighbourhoods replace market baskets, and a participation index replaces raw frequency.
Density is not the variable. Combination is.
Keep a ruler
One question remains: is every validated interest worth copying?
Illich drew a distinction in Tools for Conviviality in 1973. There is a monopoly of brands, and there is a monopoly of a type. Cars were never monopolised by one company; they pushed walking and cycling out of the city. He called it radical monopoly.
So the question is not whether a format is hot. It is what happens to the place after the format spreads: does it become more able to grow the next thing on its own, or does it now need the format in order to function?
Back to the street
A street is not designed once. It is a place where interest crystallised, and then kept crystallising.
The designer's job is not to invent interest. It is to keep the site able to receive it: keep the old buildings, hold the rent down, open the walking routes, allow the mixture, resist the urge to unify the façade.
Interest will find its own form.
Method: the framework draws on published methods — Reilly's law of retail gravitation, Huff's probabilistic model, Jacobs on urban diversity, Illich on tools, and spatial co-location analysis. No project-specific data is cited.