典范不是「优秀」的最高级。优秀的论文在既定的问法里做到最好——它越好，那个问法就越牢固；典范论文做的是相反的事：它让原来的问法失效，逼后来的人换一个位置站着重新问。这两件事之间没有连续的坡道，只有一次断裂。所以「打磨到极致就会成为典范」是这个行业里最贵的一句安慰话：沿着那条路走到头，得到的是一篇无可挑剔、也无人需要绕开的论文。这个频道收的，是走完了那道断裂的长文。
A paradigm is not the superlative of "excellent". An excellent paper does its best inside an established way of asking — and the better it is, the firmer that way of asking becomes. A paradigm paper does the opposite: it voids the old question and forces those who follow to stand elsewhere and ask again. Between the two there is no continuous ramp, only a break. Which makes "polish it far enough and it becomes a paradigm" the most expensive consolation in this profession: walk that road to its end and you arrive at a paper that is flawless, and that no one needs to go around. This channel collects the essays that made the break.
三条都可判定，不需要等历史来评价——「有没有影响力」是结果，不是判准，用它判定等于说「等它成了典范我们就知道它是典范」。下面三条在稿子还没投出去的时候就能各自判死。
All three are decidable now; none waits on history. "Did it have impact" is an outcome, not a criterion — judging by it amounts to saying we will know it is a paradigm once it becomes one. Each of these three can kill a manuscript before it is ever submitted.
- 提问失效The Question Fails这篇论文出来以后，原来那个问法还能不能原样提出来？如果能，它只是一篇好论文——它把既定的问法擦得更亮了。典范做的是相反的事：它让原来的问法当场失效，逼后来的人换一个位置站着重新问。After this paper, can the old question still be asked in the old form? If yes, it is merely a good paper — it has polished the existing way of asking. A paradigm does the opposite: it voids the old question on the spot and forces those who follow to stand somewhere else to ask again.
- 删除测试The Deletion Test把它从这个领域里整个抽掉，地基会不会塌一块？不塌，它就是装饰——无论引用多少次。典范的标记不是被引，是被依赖：后来者的话必须先经过它才说得出口。Delete it from the field entirely: does a piece of the foundation collapse? If not, it was decoration — however often it is cited. The mark of a paradigm is not citation but load-bearing: what comes after cannot be said without passing through it.
- 接口留存A Joint Left Open后来的人能不能从它的裂缝钻进去，把它推成它原本不是的样子？1854年黎曼讲完，满座没几个人听懂——它活下来，是因为此后六十年里贝尔特拉米、里奇、爱因斯坦各自带着自己的问题从它的裂缝进去了。把话说尽的理论没有被驳倒，是自己做完了自己。Can those who follow enter through its cracks and push it into something it was not? In 1854 Riemann finished speaking and almost no one had followed; it survived because over the next sixty years Beltrami, Ricci and Einstein each entered through its cracks with problems of their own. A theory that closes every gap is not refuted — it finishes itself off.
三篇各约八千字，各自独立，撞的是同一件事：当一样东西可以不经过它原本的路径就被产出，那个靠路径来结算的制度会怎么样。教育里它叫作业，医学里它叫风险，企业里它叫方案——三处的病是同一种，而三处都把它当成了别的病在治。
Three essays of ~8,000 characters each, standing alone, striking the same thing: what happens to an institution that settles by path, once the thing it settles can be produced without taking the path. In education it is called homework; in medicine, risk; in business, a proposal. One disease in three places — and all three are treating it as something else.
寄食性裂变：AI 时代教育系统困境的制度病理学诊断Parasitic Fission: An Institutional-Pathological Diagnosis
"技术替代"解释不了一件事：为什么一个核心使命长期衰减的体系，面对替代方案时表现出的是主动"消化吸收"而非被动淘汰？本文把教育系统定义为确定性垄断体，给出三阶段机制链——功能寄生、免疫寄生、宿主瓦解与能量裂变——并立住一个反直觉判断：系统越成功地维持垄断，就越加速其核心使命的瓦解。关键在于它补全了微观基础：维持这个垄断的，是一群没有任何人想维持它的人——教师因专用性资本不敢先退、家长因损失厌恶不敢让孩子脱轨、评价产业链是唯一无需观望的群体。三方各自局部理性，聚合即垄断，而无人持有那个意图。由此得出一个比"渐进侵蚀"更强也更可测的预测：崩塌是雪崩，不是侵蚀。The system's own survival strategy is what dissolves its core mission — and nobody involved intends to maintain it.
读典范论文 · 网页 · PDF →Read → 王德生 著 · 教育 · 约 8000 字by Wang Desheng · Education · ~8k作业之死：教育评价对象的蒸发与发生过程的重新显影The Death of Homework: The Evaporation of What Education Evaluates
学校不是在防作弊。作弊是拿假证据冒充真过程，而这里根本没有过程被冒充——它压根没发生。作业两千年来一直是一根指针，指向那条看不见的路径；机器没有伪造指针，它让指针和被指的东西第一次脱钩。于是评价面对的不是欺骗，是对象的蒸发：那个东西已经不在了，分数照给，而且更高、更齐、更少争议——蒸发不制造警报，它制造捷报。
Schools are not fighting cheating. Cheating passes off false evidence for a real process; here no process is being impersonated — none occurred. For two millennia homework was a pointer to an invisible path. The machine did not forge the pointer; it decoupled pointer from path. What evaluation faces is not deception but evaporation — and the grades rise, converge, and draw fewer complaints. Evaporation raises no alarm; it files good news.
读全文 →Read → 王德生 著 · 健康 · 约 8000 字by Wang Desheng · Health · ~8k不可治愈的健康人：风险作为医学新对象的发生学解剖The Incurable Healthy: Risk as Medicine's New Object
你可以治好一场肺炎，你治不好一个 12% 的十年风险——因为那个数字根本不在这个人身上，它在队列里。医学换了对象却没换结算方式，于是它第一次接手了一件永远无法完成的事：每一次干预只把影子挪个位置，系统报告一次胜利，再开下一张单子。四十年，没有终点，而每一步都是对的。
You can cure pneumonia. You cannot cure a 12% ten-year risk — that number was never in the person; it is in the cohort. Medicine changed its object without changing how it settles, and so took on a task it can never complete: each intervention moves the shadow, the system files a victory and issues the next order. Forty years, no endpoint — and every single step correct.
读全文 →Read → 王德生 著 · 商业 · 约 8000 字by Wang Desheng · Business · ~8k判错权：当执行的成本归零，企业必须把流程变回判断The Right to Be Wrong: When Execution Costs Nothing
现代企业一百年的成就，是把判断从人身上剥下来冻成流程——这在执行昂贵的年代完全正确。而流程只能判它见过的错。当机器把候选变得免费，企业第一次获得了把没见过的错误规模化的能力：每一步都对，合起来是灾难。判错权是组织里唯一不可分割、不可积累、不可规模化的东西——因为权力的另一面是后果，而机器没有承受后果的位置。
A century of organisational achievement lay in stripping judgment off individuals and freezing it into process — entirely correct while execution was expensive. But process can only catch the errors it has already seen. Now that candidates are free, firms have gained, for the first time, the power to scale errors nobody has seen. Every step correct; the sum a disaster.
读全文 →Read → 三篇共用的那把刀The blade all three share同一个判准，三个战场One Criterion, Three Battlefields
三篇都没有回答本行业正在争论的那个问题——它们都在证明那个问题问错了。防作弊、早筛该不该做、AI 要不要参与决策：三场争论的双方共用同一个前提，而三篇文章各自把那个前提拆了。这正是上一节三条判准里的第一条：论文出来以后，原来的问法还站不站得住。
None of the three answers the question its field is arguing about; each shows the question was wrong. Anti-cheating, whether to screen, whether AI should decide — in all three debates both sides share one premise, and each essay takes that premise apart. Which is criterion one from the section above: after the paper, does the old question still stand?
看它们是怎么被撞出来的 →How they were struck out →它们不介绍任何框架，只把一个具体问题追问到它自己承认不住的地方。四篇都出自教育现场，但要处理的都不是教育独有的病。
They introduce no framework; each pursues one concrete question until it can no longer hold itself up. All four come from education, and none of them treats a disease peculiar to education.
临界点燃：教育发生学的不可还原单元Critical Ignition: Education's Irreducible Unit
在「传递」隐喻之下，有一种制度无法度量、也无法计划的教育发生——学习者由内部自发迸发的、不可逆的认知跃迁；本文追问制度化教育为何系统性地遗忘并排异它。
Beneath the 'transfer' metaphor lies an educational event institutions cannot measure or plan: the learner's self-organized, irreversible cognitive leap — and why schooling forgets it.
读全文 →Read → 教育专栏 · 典范文Education · Paradigm Essay制度性留白Institutional Reticence
AI 把「教育者知道你该长成什么样」这个两百年预设悄悄掏空了；学校真正不可替代的，是守护一片「自己去发现什么值得在乎」的、受规则保护的留白。
Both 'outsource to AI' and 'return to the classics' assume educators know what the educated should become. School's irreplaceable core: guarding space to discover what one cares about.
读全文 →Read → 教育专栏 · 典范文Education · Paradigm Essay在规则的字面碰撞处，让判断发生Letting Judgment Occur Where a Rule's Letter Collides
同一规则的两条指令字面冲突，把它诊断为「矛盾」而停机，本身是一个有发生路径的认知事件；在「谁有权解释」之前，先问一个人为何在碰撞处连判断都不做——为衡平传统补上前置一章。
Two instructions from one rule cannot both be obeyed. Before asking who may interpret, ask why one person sees a contradiction and stops — a prelude to the theory of equity.
读全文 →Read → 教育专栏 · 典范文Education · Paradigm Essay当困惑不再压舱When Confusion No Longer Ballasts
AI 即时消除困惑，瓦解了深层学习所需的「认知压舱」——那片让困惑合法滞留、张力得以蓄积的制度空间；出路不是回到过去，而是在新条件下重新把它造出来。
AI dissolves the institutional space where confusion is allowed to linger — the cognitive ballast deep learning needs. Not nostalgia, but re-inventing that space.
读全文 →Read →一个人很难自己跟自己碰撞——你的三条路径共用同一副脑子，误差同向，怎么撞都撞不出裂缝。所以要三台路径互锁的智能体。
One mind collides poorly with itself: three paths sharing one head err in the same direction, and no amount of collision opens a crack. Hence three agents on mutually locked paths.