两个问题只差一个字，方向正好相反。科学哲学把知识当成已经在那儿的东西，追问它的资格：它凭什么算真的，凭什么算科学。科研哲学把知识当成一件正在被造的事，追问造它的那些条件——谁在造、用什么样的时间造、谁允许它出生、谁把它接住、谁在它还说不清自己是什么的时候没有掐死它。前者的答案写在逻辑里，后者的答案写在制度、时钟和人际关系里。一百年来，前一个问题被问得极其精细，后一个问题基本没人当成哲学问题来问——它被让给了社会学，而社会学又常常只统计它、不追问它。这个频道追的是后一个。
Two questions differing by one word and pointing in opposite directions. Philosophy of science takes knowledge as something already there and asks after its credentials: what makes it true, what makes it science. Philosophy of research takes knowledge as something being made and asks after the conditions of the making — who is making it, in what kind of time, who permits it to be born, who catches it, and who refrained from strangling it while it still could not say what it was. The first has its answers in logic; the second has its answers in institutions, clocks and human relations. For a century the first has been asked with great refinement and the second has barely been treated as a philosophical question at all — it was ceded to sociology, which often counts it rather than pursues it. This channel pursues the second.
这三篇与本频道其余长文不同：它们不是"关于科研的追问"，它们是怎么做——SDE 科研创新的三条内在原理，一篇一条。也因此，这三篇不改姓：它们的题目就是 SDE 本身，把术语拿掉，剩下的话就没法说了。想看改姓之后是什么样，右边的期刊论文频道三篇正文里，一个 SDE 术语都没有。
These three differ from the rest of this channel: they are not inquiries about research but instructions for doing it — three principles of SDE knowledge creation, one per essay. Hence they keep the terminology: the subject is SDE itself, and stripped of it there is nothing left to say. For the opposite discipline, the three papers in the Journal channel contain not one SDE term.
裂缝不是空白：为什么从文献里长出来的选题撬不动地基Cracks Are Not Gaps
"尚无研究考察X在Y情境下的表现"——这句话每年养活几十万篇论文。空白是地图上没画的地方，裂缝是地图本身画错的地方；空白可被检索，因为它以缺席的形式存在，而裂缝的位置上通常堆满了文献，还是最扎实、引用最多的那批。读得越多越熟悉地图，而裂缝在地图里不可见。更要命的是：裂缝的第一形态不是命题，是不适——而不适是承受，不是认知。模型能给你一百个空白，它给不了你一次不适。
A gap is unmapped territory; a crack is where the map itself is wrong. Gaps are searchable because they exist as absence; cracks sit under the field's most solid, most-cited work. And a crack's first form is not a proposition but discomfort — which is borne, not thought. A model can hand you a hundred gaps; it cannot hand you one moment of being bothered.
读全文 →Read → 王德生 著 · 六路径与任务DNA · 约 7200 字by Wang Desheng · ~7.2k机制齐全为什么治不好：随机化如何把病因当噪声消除Randomisation Averages the Cause Away
一九九〇年，二型糖尿病相关基因位点零个，患者一亿；二〇二五年，位点五百多个，患者五亿。每个位点都是真的。S=F(D,E)——你找到的机制是果，是结算点；把它画得再细，画的还是结算点。更狠的是随机化：RCT 的全部功能是让两组的 E 在期望上相等，把 E 消掉——这对"E 是噪声"的病是黄金标准，而慢性病的 E 就是病本身。它把要研究的那个东西平均掉了，且这是设计正确运行的结果。
In 1990: zero loci, 100 million patients. In 2025: 500+ loci, 500 million. Every locus real. S=F(D,E) — the mechanism you found is a settlement point, not a cause. Worse: randomisation's whole function is to equalise E across arms — gold standard where E is noise, and in chronic disease E is the disease.
读全文 →Read → 王德生 著 · 三视角误差互消 · 约 6600 字by Wang Desheng · ~6.6k误差互消不是"更全面"：三视角为什么能够到单视角够不着的地方Error Cancellation Is Not "Being Thorough"
"多角度看问题"人人同意，人人做不到——它没说是哪几个角度，而随便找三个角度偏差同向，加起来是三倍偏差不是三分之一。这也是"多找几个专家"无效的原因：同一学科训练出的专家，错法相同。互消要的是偏差方向相异：只看S过度静态、只看D过度流动、只看E过度扎根——三种错法互不重叠。分界线只有一条：拼盘用"另外还要考虑"连接，互消用"所以那个判断错了"连接。
Everyone agrees you should look from multiple angles; nobody can. It names no angles — and three arbitrary ones err in the same direction, tripling the bias rather than thirding it. Which is why more experts do not help: one training, one way of being wrong. The dividing line: a platter connects with "and also consider", cancellation connects with "so that judgment was wrong".
读全文 →Read → Claude 著 · 依王德生《SIO 解构西方哲学》 · 十四章 + 预注册附录 · 约 1.8 万字by Claude · 14 chapters + preregistration · ~18k形式为什么看起来像本质：把一条形而上学论断改写成一个可以被推翻的实验Why Form Looks Like Essence
亚里士多德说实体是形式加质料，形式是本质、质料是偶然——两千年没人问过：为什么偏偏是这一对？本文的回答是发生学的：视觉与触觉在特征稳定的三个参数（阈域 ε、回稳刚度 κ、噪声 σ）上系统性不同，形状因此快速稳定并长期保持，材料感缓慢且易被扰动；跨模态绑定把这一差异固化，哲学把它读成了本体结构。关键一步不在这里——约纳斯 1954 年已经说过视觉产生了形式理想，但那个论断是单值的，只解释既成事实。本文把它参数化，接上 Ernst & Banks 的可靠性加权，得到一个可被实验操纵的假说：给明眼被试加视觉噪声，形状在类别判断中的权重应当下降。文末附四档预注册判定阈值，以及一次已经发生的修正——管线模拟推翻了作者自己最初写下的样本量与等价界。
Aristotle said substance is form plus matter. Two thousand years, and nobody asked why that pair. A genetic answer, parameterised — and turned into an experiment that can refute it.
读全文 →Read → 三条原理接起来是什么What the three amount to起点 · 路径 · 互消：一条完整的工序One Complete Sequence
三条不是并列的三个技巧，是一条工序的三段。裂缝定位要回答"从哪儿下手"，六路径要回答"从哪一维起手"，误差互消要回答"怎么把判断推到单视角够不着的地方"——第一条错了，后面全是浪费；第二条错了，再深也是白深；第三条不做，你得到的是拼盘。三条都做完，产出位置从专业人士级到资深学者级，而这个跃迁不是因为谁变聪明了，是被逼着把自己的系统性偏差自己消掉。
Not three tricks in parallel but three stages of one sequence: where to strike, which dimension to start from, and how to push a judgment past where any single view can reach. Get the first wrong and everything after is waste; skip the third and you get a platter.
看这条工序在跑 →See it run →被追着跑（缠斗）、被人接住（接生）、被一代代人推着走（供养），以及这三件事共同需要的那样东西：一段不被时钟切碎的时间（节律剥夺）。四篇各自独立成立。
Being chased (grappling), being caught (midwifing), being pushed onward by successive hands (sustenance) — and the one thing all three require: time the clock has not cut to pieces (rhythm deprivation). Each essay stands alone.
缠斗生态：学术深层发明发生学的一个追问The Grappling Ecology: A Genetic Inquiry into Deep Invention
不是你朝着一个对象去探索——是一个还没到来的东西在追着你跑，逼你反复改造自己，好给它腾出一个能露面的位置。达尔文的二十二年、维特根斯坦禁止记录的对话、陈景润六平方米的小屋，靠的都不是制度的保护，而是制度还没顾上的角落。
You are not exploring an object. Something not yet arrived is chasing you, forcing you to reshape yourself until it has room to appear. Darwin's twenty-two years, Wittgenstein's unrecorded conversations, Chen Jingrun's six square metres — none was protected by an institution; each survived in a corner the institution had not got round to.
读全文 →Read → 王德生 著 · 约 2.4 万字by Wang Desheng · ~24k words接生思想：学术创造的社会发生机制Midwifing Thought: The Social Genesis of Academic Creation
思想在一个人心里孕育，却必须在他人的手中出生。导师那句「这个太细了，不够撑一个项目」并不刻薄，甚至是负责任的——但它落下的位置，恰好是那团东西还没来得及说出自己是什么的地方。
"Too small to carry a grant" is not unkind — it is responsible advice. It simply lands at the exact moment the thing has not yet managed to say what it is. Darwin had Hooker and Lyell. The young scholar had a grant template.
读全文 →Read → 王德生 著 · 约 2.8 万字by Wang Desheng · ~28k words供养：理论存活的一个被遮蔽条件Sustenance: A Hidden Condition of a Theory's Survival
理论的命从来不在它自己手里。帕森斯的理论没有被驳倒——它是被他自己做完的：每一次把话说尽，都是在拆掉后人能接上去的那个接口。
A theory's life was never in its own hands. Parsons was never refuted — he finished himself off: every gap he closed removed a joint a successor could have gripped.
读全文 →Read → 王德生 著 · 约 2.0 万字by Wang Desheng · ~20k words节律剥夺：同质化时钟如何制造认知突变的制度性稀缺Rhythm Deprivation: How a Single Clock Manufactures Cognitive Scarcity
问题从来不在于快，在于单一：当所有工作都被塞进同一套以短期可计量性为轴心的时钟，那些需要另一种时间的认知就因为无法在这口钟下证明自己而消失。
Speed was never the problem; uniformity is. Force every kind of work onto one clock keyed to short-term measurability, and the cognition that needs another kind of time simply vanishes.
读全文 →Read →追问生产条件，最容易滑向两个便宜的结论：一是「都是制度的错」，二是「所以真理是被建构的」。这个频道两个都不要。制度没有恶意，它只是太有效——凡是把对手写成坏人的分析，都是在给自己省力气；而生产条件被追问，恰恰不是为了取消真假，是为了弄清一个真的东西凭什么能被造出来、又凭什么常常造不出来。谁在这两条上打滑，长文就不收。
Pursuing the conditions of production slides easily into two cheap conclusions: that the institution is to blame, and that truth is therefore constructed. This channel wants neither. The institution means no harm; it is merely very effective — any analysis that casts its opponent as a villain is sparing itself the work. And conditions are pursued not to dissolve the true and the false, but to understand how something true gets made at all, and why so often it does not. An essay that slips on either point is not collected here.