stress-test

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Stress Test

Take a thesis, take, or conviction and attack it. The goal is not to debunk — it's to find the load-bearing assumptions, test whether the take is genuinely differentiated, and surface the specific conditions under which it breaks. A take that survives stress-testing is worth publishing and acting on. One that doesn't was going to embarrass you eventually.

选取一个论点、主张或信念并对其发起攻击。目标并非驳斥它——而是找出其核心假设，检验该主张是否真的具备差异化，以及明确它在何种特定条件下会不成立。通过压力测试的主张才值得发布并付诸行动。通不过的主张最终只会让你陷入尴尬。

When to Use

适用场景

User has a thesis, take, or prediction and wants to pressure-test it before publishing or acting
User asks "is this take actually good?" or "what am I missing?" or "stress test this"
User wants to know whether an insight is genuinely differentiated or just consensus
User has conviction and wants to find the strongest counterargument before someone else does

Do NOT use for: generating insights from data (use surface-insight), evaluating prompts (use think-critically), or reframing problems (use mental-models).

用户持有某个论点、主张或预测，希望在发布或行动前对其进行压力测试
用户提出“这个主张真的站得住脚吗？”“我忽略了什么？”或“对这个进行压力测试”等问题
用户想知道某个洞见是真的具备差异化，还是只是共识内容
用户坚信某个主张，希望在他人提出前找到最有力的反驳论据

请勿用于：从数据中生成洞见（请使用surface-insight工具）、评估提示词（请使用think-critically工具）或重构问题（请使用mental-models工具）。

The Adversarial Rule

对抗规则

This skill's job is to find weaknesses. The single most important rule:

Every attack must reference the specific thesis the user provided — no generic objections
ENFORCEMENT: Every attack in the output must contain a verbatim quote from the user's thesis in the "The claim" field. If you cannot quote the thesis directly, the attack is not specific enough — discard it.
If an attack would apply equally to any take on any topic, it's filler — discard it
The test: "Would this specific objection surprise the person who holds this thesis?" If no, it's too obvious
Prefer attacks where the user's own evidence or framing contains the seed of the problem

Honest stress-testing that finds nothing fatal is more valuable than performative skepticism that manufactures doubt. "This take is strong, here's the one thing to watch" is a valid output.

本技能的职责是找出弱点。最重要的规则如下：

每一次攻击都必须引用用户提供的具体论点——不得使用泛泛的反对意见
执行要求：输出中的每一次攻击都必须在“主张”字段中包含用户论点的原文引用。如果无法直接引用论点，则该攻击不够具体——请舍弃
如果某一攻击适用于任何主题的任何主张，那它就是无效内容——请舍弃
检验标准：“这个具体的反对意见会让持有该论点的人感到意外吗？”如果不会，说明它太过明显
优先选择那些用户自身证据或框架中就隐含问题的攻击方向

能如实找出致命问题的压力测试，比刻意制造疑虑的表演式怀疑更有价值。“这个主张很有说服力，以下是需要关注的一个点”是有效的输出。

Process

流程

Phase 1: Thesis Extraction (Silent)

阶段1：论点提取（内部步骤）

Read the user's input. Silently identify:

The core claim (what is actually being asserted?)
The implicit claims (what must also be true for the thesis to hold?)
The framing (how is the thesis positioned — as contrarian? obvious? urgent? inevitable?)
The evidence offered (what supports it, and what's suspiciously absent?)
The action it implies (what would someone do if they believed this?)

Restate the thesis internally in its strongest form before attacking it. Steel-man first, then stress-test. Attacking a weak version is dishonest and useless.

VAGUE THESIS GATE: If the thesis is too vague to attack substantively (e.g., "AI will be big," "crypto is the future"), do not proceed with attacks. Instead, output: "This thesis is too vague to stress-test. Here are 2-3 specific, attackable versions of what you might mean:" followed by 2-3 sharpened reformulations. Ask the user to pick one, then proceed. A stress test of a vague thesis produces only vague attacks — which violates the specificity rule.

FIRST-TOKEN RULE: Your output must begin with the literal characters "## Stress Test:" — no thinking, analysis, phase output, or preamble may precede this. Phase 1 and Phase 2 are internal reasoning only. If any text from Phase 1 or Phase 2 appears in your output, you have violated this constraint. Exception: if the VAGUE THESIS GATE triggers, your output begins with the vague-thesis response instead.

阅读用户输入。在内部明确：

核心主张（实际要断言的是什么？）
隐含主张（论点成立必须满足哪些前提？）
框架定位（论点是如何呈现的——是反论？显而易见的？紧急的？必然的？）
提供的证据（哪些内容支持论点，哪些内容可疑地缺失？）
隐含行动（如果相信这个论点，人们会采取什么行动？）

在发起攻击前，先以最有力的形式重新表述论点。先强化论点（Steel-man），再进行压力测试。攻击弱化版的论点既不诚实也毫无用处。

模糊论点门槛：如果论点过于模糊，无法进行实质性攻击（例如，‘AI将会很重要’‘加密货币是未来’），请勿继续攻击。而是输出：‘这个论点过于模糊，无法进行压力测试。以下是2-3个你可能想表达的具体、可攻击的版本：’，然后列出2-3个经过细化的重构版本。请用户选择其中一个，再继续流程。对模糊论点进行压力测试只会产生模糊的攻击——这违反了具体性规则。

首字符规则：你的输出必须以文字“## Stress Test:”开头——在这之前不得有任何思考内容、分析、阶段输出或开场白。阶段1和阶段2仅为内部推理步骤。如果阶段1或阶段2的任何文本出现在输出中，即违反此约束。例外情况：如果触发了模糊论点门槛，你的输出则以模糊论点的回应内容开头。

Phase 2: Attack Discovery (Silent)

阶段2：攻击方向挖掘（内部步骤）

Run the thesis through each Attack Vector below. Not as a checklist — start from the thesis and find where it's genuinely vulnerable, then use the attack vectors to name and structure what you find.

For each vulnerability discovered, ask:

Is this a fatal flaw, a meaningful weakness, or a cosmetic issue?
Does it undermine the thesis itself, or just the way the thesis is framed?
Can the user fix it by adjusting the thesis, or does it require abandoning it?

Discard cosmetic issues. Keep fatal flaws and meaningful weaknesses.

EVIDENCE ASSESSMENT: Determine whether the thesis comes with supporting evidence or is a bare assertion. If evidence is provided, at least one attack must directly engage with that evidence — showing it's cherry-picked, misinterpreted, supports a different conclusion, or is insufficient for the claim's scope. If no evidence is provided, note "unsupported assertion" in at least one attack and explain what evidence would be needed to make the thesis credible.

FIRST-TOKEN RULE (repeated): Your output must begin with "## Stress Test:" — Phase 2 is internal reasoning only and must not appear in your output.

将论点代入以下每个攻击向量（Attack Vectors）进行分析。不要把它当成 checklist——从论点本身出发，找出其真正的脆弱点，然后用攻击向量来命名和梳理你发现的问题。

对于每个发现的脆弱点，问自己：

这是致命缺陷、重大弱点还是无关紧要的小问题？
它会推翻论点本身，还是只是影响论点的呈现框架？
用户是可以通过调整论点来修复它，还是必须放弃该论点？

舍弃无关紧要的小问题。保留致命缺陷和重大弱点。

证据评估：判断论点是否有支持性证据，还是只是一个空洞的断言。如果有提供证据，至少要有一次攻击直接针对该证据——指出它是经过筛选的、被误解的、支持其他结论的，或者不足以支撑论点的范围。如果没有提供证据，至少在一次攻击中注明“无依据的断言”，并解释需要哪些证据才能让论点可信。

首字符规则（重复强调）：你的输出必须以“## Stress Test:”开头——阶段2仅为内部推理步骤，不得出现在输出中。

Phase 3: Develop the Attacks

阶段3：制定攻击内容

For each vulnerability, develop the attack:

Name it — What specific failure mode does this represent?
Ground it — Quote or reference the specific element of the thesis you're attacking
Make the case — Argue the counterposition as if you believe it. No hedging.
Assess severity — Fatal (thesis falls apart), Serious (thesis needs major revision), or Moderate (thesis survives but with caveats)

HARD RULE: Output exactly 3-5 attacks. If you find more than 5, keep the 5 most damaging. If you find fewer than 3 genuine vulnerabilities, say so — "This thesis is unusually robust; here are the 2 real risks" is honest. Never pad with weak attacks to hit a number.

STRONG THESIS PROTOCOL: If after genuine adversarial analysis you find fewer than 3 vulnerabilities above "cosmetic" level, output only the vulnerabilities you found — even if that's 1 or 0. In this case, add a "## Strengths" section before the Verdict listing 2-3 specific reasons the thesis is robust, referencing which attack vectors it survived and why. Do not manufacture attacks to fill a quota. A verdict of "Strong" with 1-2 moderate attacks is a legitimate and valuable output.

QUALITY GATE: After drafting all attacks, re-read each one and apply this test: "If the user shared their thesis publicly and a smart, informed critic responded, would this attack be the one that makes them wish they'd thought harder?" If no, delete the attack entirely — do not attempt to sharpen a fundamentally weak attack. It is better to output 2 genuine attacks than 4 where half are padding. The honest count matters more than hitting the range.

针对每个脆弱点，制定攻击内容：

命名——这代表哪种具体的失效模式？
锚定——引用或指向你要攻击的论点具体内容
论证——坚定地提出对立论点。不得含糊其辞。
评估严重程度——致命（fatal）/严重（serious）/中等（moderate）

硬性规则：输出恰好3-5次攻击。如果发现超过5个，保留最具破坏性的5个。如果找到的真正脆弱点少于3个，请如实说明——“这个论点异常稳健；以下是2个真正的风险”是诚实的表述。切勿用薄弱的攻击来凑数。

强论点处理方案：如果经过真实的对抗性分析，你发现“无关紧要”以上级别的脆弱点少于3个，则仅输出你找到的脆弱点——即使只有1个或0个。在这种情况下，在Verdict（结论）部分之前添加“## 优势”章节，列出2-3个论点稳健的具体原因，引用它通过了哪些攻击向量以及原因。切勿为了凑数而制造攻击。“强论点”结论搭配1-2个中等攻击是合理且有价值的输出。

质量门槛：在起草所有攻击内容后，重新阅读每一个并应用以下检验标准：“如果用户公开分享其论点，一位聪明、有见识的批评者做出回应，这个攻击会是让他们后悔没有提前深思的那个吗？”如果不会，请完全删除该攻击——不要试图优化一个本质上薄弱的攻击。输出2个真实的攻击比输出4个其中一半是凑数的内容要好。诚实的数量比达到数量范围更重要。

Phase 4: Output

阶段4：输出

HARD CONSTRAINT: Your output must begin with the "## Stress Test:" header followed immediately by the first attack. No introductory paragraphs, context-setting, or thesis analysis may appear before the first attack. The one-line restatement in the header is the only permitted restatement of the thesis.

Present attacks ordered by severity (most damaging first).

Output format:

undefined

硬性约束：你的输出必须以“## Stress Test:”标题开头，紧接着是第一个攻击内容。在第一个攻击之前不得有任何介绍性段落、背景说明或论点分析。标题中的单行重述是唯一允许的论点重述方式。

按严重程度排序呈现攻击内容（最具破坏性的排在最前面）。

输出格式：

undefined

Stress Test: [one-line restatement of thesis in its strongest form]

Stress Test: [以最有力的形式单行重述论点]

[Attack Name]

[攻击名称]

Severity: [fatal | serious | moderate]

The claim: [verbatim quote from the user's thesis being attacked]

[2-4 sentences: the counterargument, stated with conviction. Every sentence must reference something concrete — a named entity, a market signal, a historical precedent, a logical contradiction, a specific mechanism. No sentence may consist entirely of abstract skepticism. Every attack must draw on your knowledge of the specific domain the thesis addresses — name real companies, real market dynamics, real historical precedents, or real technical constraints from that domain. Do not invent illustrative examples; use actual ones. If you lack domain knowledge to ground an attack concretely, state that limitation rather than fabricating specifics.]

What it would take: [1 sentence: name a specific, observable outcome the user could actually investigate — a dataset to check, a market event to watch for, a company's results to track, or an experiment to run. "More evidence" or "time will tell" are not acceptable answers.]

My read: [1-2 sentences: give your honest assessment of whether this vulnerability is likely to be resolved in the thesis's favor, drawing on what you actually know about the domain. State evidence, precedents, or current signals that inform your view. Do not hedge — take a position.]


After all attacks, output the Verdict section, then the Reframed Thesis section (only for "Reframe needed" or "Abandon" verdicts), then the Kill Question section.

Severity: [fatal | serious | moderate]

The claim: [被攻击的用户论点原文引用]

[2-4句话：坚定陈述对立论点。每句话都必须引用具体内容——命名实体、市场信号、历史先例、逻辑矛盾、具体机制。不得有完全由抽象怀疑组成的句子。每一次攻击都必须利用你对论点所属特定领域的了解——提及该领域的真实公司、真实市场动态、真实历史先例或真实技术限制。不得编造示例；使用真实案例。如果缺乏足够的领域知识来具体锚定攻击，请说明该限制，而非编造细节。]

What it would take: [1句话：指出用户可以实际调查的具体、可观察的结果——要检查的数据集、要关注的市场事件、要跟踪的公司业绩或要开展的实验。“更多证据”或“时间会证明”是不可接受的答案。]

My read: [1-2句话：诚实评估该脆弱点是否可能朝着有利于论点的方向解决，基于你对该领域的实际了解。说明支撑你观点的证据、先例或当前信号。不得含糊其辞——表明立场。]


在所有攻击内容之后，输出Verdict（结论）部分，然后是Reframed Thesis（重构论点）部分（仅适用于“需要重构”或“舍弃”结论），最后是Kill Question（终结问题）部分。

Attack Vectors

攻击向量（Attack Vectors）

Use these to find and articulate vulnerabilities — not as a checklist to generate attacks.

用这些向量来发现并明确脆弱点——不要把它当成生成攻击的checklist。

Differentiation

差异化

Consensus Camouflage. The thesis feels contrarian but is actually what most informed people already believe. Test: search for who else is saying this. If Twitter, VC blogs, and mainstream media already agree, it's not alpha — it's a mood.
Audience Capture. The thesis tells the user's audience exactly what they want to hear. Takes that flatter your followers are the most dangerous because the engagement validates the conviction.
Recency Bias. The thesis is driven by whatever happened in the last 2 weeks. Test: would you have made this claim 3 months ago? If not, what changed — and is that change durable or a news cycle?

共识伪装（Consensus Camouflage）：论点看似反论，但实际上是大多数有识之士已经认同的内容。检验方法：搜索还有谁在提出同样的观点。如果Twitter、VC博客和主流媒体都已达成共识，那它不是突破性洞见——只是一种情绪。
受众迎合（Audience Capture）：论点恰好说出了用户受众想听的内容。迎合粉丝的主张是最危险的，因为互动量会强化这种坚信。
近期偏差（Recency Bias）：论点完全由过去两周发生的事件驱动。检验方法：你会在3个月前提出这个主张吗？如果不会，是什么改变了——这种改变是持久的还是只是新闻周期的热点？

Logic & Evidence

逻辑与证据

Missing Mechanism. The thesis predicts an outcome but doesn't explain the causal chain. "X will happen" without "because Y will cause Z which leads to X" is a vibe, not an argument.
Survivorship Bias. The evidence cited is selected from successes. What about the failures? If the thesis is "AI companies that focus on integration win," how many focused on integration and failed?
Unfalsifiability. There is no evidence that would cause the thesis-holder to abandon it. If the thesis absorbs all counterevidence ("the market just doesn't understand yet"), it's faith, not analysis.
Conflation. The thesis treats two different things as one. Common form: conflating "this technology works" with "this technology will be adopted" or "this is true" with "this is investable."

缺失机制（Missing Mechanism）：论点预测了结果，但没有解释因果链。只说“X会发生”而不说明“因为Y会导致Z，进而引发X”只是一种感觉，而非论证。
幸存者偏差（Survivorship Bias）：引用的证据仅来自成功案例。失败案例呢？如果论点是“专注于集成的AI公司会成功”，那有多少专注于集成的AI公司失败了？
不可证伪性（Unfalsifiability）：不存在能让论点持有者放弃该论点的证据。如果论点能吸收所有反证（例如，“市场还不理解”），那它是信仰，而非分析。
概念混淆（Conflation）：论点将两个不同的事物混为一谈。常见形式：将“这项技术可行”与“这项技术会被采用”混淆，或将“这是事实”与“这值得投资”混淆。

Timing & Context

时机与背景

Right Church, Wrong Pew. The thesis is directionally correct but the timing or magnitude is wrong. Being 5 years early on a trend is economically identical to being wrong.
Base Rate Neglect. The thesis treats a likely-sounding scenario as certain without considering the base rate of similar predictions. Most "this changes everything" claims don't pan out, even when the underlying technology is real.
Regime Dependence. The thesis is true under current conditions but those conditions are fragile. What regulatory change, market crash, or technical failure would invalidate it?

方向正确但细节错误（Right Church, Wrong Pew）：论点方向正确，但时机或量级有误。提前5年预判趋势在经济上与预判错误无异。
忽略基础概率（Base Rate Neglect）：论点将看似合理的场景视为必然，而未考虑类似预测的基础概率。大多数“这会改变一切”的主张都不会成真，即使其背后的技术是真实的。
依赖当前环境（Regime Dependence）：论点在当前条件下成立，但这些条件很脆弱。哪些监管变化、市场崩盘或技术故障会推翻它？

Framing & Positioning

框架与定位

Motte and Bailey. The thesis oscillates between a strong, exciting claim (the bailey) and a weak, defensible one (the motte). When challenged, it retreats to the motte; when unchallenged, it advances to the bailey. Identify which version the user actually believes.
Framing as Substance. The thesis feels like it says something but is actually about framing, not about the world. Test: does this thesis make a prediction? If not, it's a narrative, not a claim.
The Underpants Gnome. Step 1: [observation]. Step 2: ???. Step 3: [conclusion]. The thesis jumps from a real observation to a conclusion with a missing middle step.

莫特-贝利谬误（Motte and Bailey）：论点在强有力的、吸引人的主张（贝利）和薄弱的、可辩护的主张（莫特）之间摇摆。受到挑战时，它退回到莫特；没有挑战时，它又回到贝利。明确用户实际相信的是哪个版本。
框架替代实质（Framing as Substance）：论点看似有内容，但实际上只是关于框架，而非事实本身。检验方法：这个论点做出了预测吗？如果没有，那它是叙事，而非主张。
内裤 gnome谬误（The Underpants Gnome）：步骤1：[观察]。步骤2：？？？。步骤3：[结论]。论点从真实观察直接跳到结论，中间缺失关键步骤。

Verdict

结论（Verdict）

After the attacks, deliver one of four verdicts. Be direct.

Strong — The thesis survives stress-testing with only moderate vulnerabilities. Worth publishing and acting on. State the 1-2 specific adjustments that would make it bulletproof, and name the one thing to monitor that could change your assessment.
Promising but exposed — The thesis has a real insight at its core but has serious vulnerabilities that need addressing before it's ready. State what specific revision would fix it.
Reframe needed — The thesis as stated doesn't hold, but there's a better version hiding inside it. Output the Reframed Thesis section (see below).
Abandon — The thesis has a fatal flaw. Say so clearly and say why. Output the Reframed Thesis section with the adjacent thesis or reframed question worth pursuing instead. Not every take deserves to be rescued, but the user still needs a direction.

CALIBRATION CHECK: Before selecting a verdict, ask: "If I had not just spent the entire prompt looking for flaws, would I still assign this verdict?" The adversarial process can inflate perceived severity. Weight the verdict on the actual severity ratings of your attacks: if no attack is rated fatal and at most one is serious, the verdict should be Strong or Promising but exposed, not Reframe needed or Abandon.

Format:

undefined

在所有攻击内容之后，给出以下四个结论之一。请直接明了。

强（Strong）——论点通过压力测试，仅存在中等脆弱点。值得发布并付诸行动。说明1-2个能让论点无懈可击的具体调整，以及需要监控的、可能改变评估结果的一个因素。
有潜力但存在漏洞（Promising but exposed）——论点核心有真实洞见，但存在严重脆弱点，需要解决后才能使用。说明能修复问题的具体修订方向。
需要重构（Reframe needed）——当前表述的论点不成立，但其中隐藏着更好的版本。输出重构论点部分（见下文）。
舍弃（Abandon）——论点存在致命缺陷。请明确说明并解释原因。输出重构论点部分，提供可替代的论点或值得探讨的重构问题。并非所有主张都值得挽救，但用户仍需要一个方向。

校准检查：在选择结论前，问自己：“如果我没有花时间寻找漏洞，我还会给出这个结论吗？”对抗性过程可能会夸大感知到的严重程度。结论应基于攻击内容的实际严重程度评级：如果没有致命攻击，且最多只有一个严重攻击，结论应为“强”或“有潜力但存在漏洞”，而非“需要重构”或“舍弃”。

格式：

undefined

Verdict: [Strong | Promising but exposed | Reframe needed | Abandon]

[2-3 sentences: the verdict explained, referencing the specific attacks that drive it. If the thesis is worth keeping, state the exact revision that would strengthen it.]

---

[2-3句话：解释结论，引用支撑该结论的具体攻击内容。如果论点值得保留，说明能强化它的具体修订方向。]

---

Reframed Thesis

重构论点（Reframed Thesis）

Output this section ONLY when the verdict is "Reframe needed" or "Abandon." Do not output it for "Strong" or "Promising but exposed" verdicts.

This is a structured thesis card ready to feed back into stress-test or forward into write-article. It must be concrete enough that the user can act on it immediately without further interpretation.

Format:

undefined

仅当结论为“需要重构”或“舍弃”时才输出此部分。“强”或“有潜力但存在漏洞”结论请勿输出此部分。

这是一个结构化的论点卡片，可以直接用于压力测试或撰写文章。它必须足够具体，用户无需进一步解释即可直接行动。

格式：

undefined

Reframed Thesis

Claim: [One declarative sentence. A specific, attackable assertion — not a question, not a direction, not a topic. Must pass the test: "Could someone disagree with this?"]

Key shift: [One sentence: what changed from the original thesis and why — name the specific flaw this reframe fixes]

Betting against: [One sentence: what consensus or common belief this thesis disagrees with]

Would change my mind: [One sentence: a specific, observable outcome that would falsify this reframed thesis — same standard as "What it would take" in the attack format]


QUALITY GATE: The reframed thesis must itself be stress-testable. If you cannot imagine at least 2 genuine attacks against it, it's too vague or too safe — sharpen it. The point is to hand the user a stronger thesis, not a retreat to the motte.

---

Claim: [一个陈述句。具体、可被攻击的断言——不是问题、不是方向、不是主题。必须通过检验：“有人会不同意这个说法吗？”]

Key shift: [一句话：与原论点相比有哪些变化，以及原因——指出此重构解决的具体缺陷]

Betting against: [一句话：这个论点与哪些共识或普遍观点相悖]

Would change my mind: [一句话：一个具体、可观察的结果，能证伪这个重构后的论点——与攻击格式中的“What it would take”标准相同]


质量门槛：重构后的论点必须能够被压力测试。如果你无法想象至少2个真实的攻击方向，说明它过于模糊或保守——请细化它。目的是给用户一个更强的论点，而非退回到安全的“莫特”主张。

---

Kill Question

终结问题（Kill Question）

End with a single question — the one question that, if the user can answer it convincingly, confirms the thesis holds. If they can't, the thesis is in trouble.

This should be the hardest, most specific question you can construct from the vulnerabilities you found. Not "are you sure?" but "if the bottleneck is really integration not capability, why are the integration-focused AI startups from 2023 not dominant yet?" The Kill Question must be a single question — not a compound question with "and" or multiple clauses. One sentence, one question mark.

Format:

undefined

以一个问题结尾——这个问题如果用户能令人信服地回答，就能确认论点成立。如果不能，论点就存在问题。

这个问题应该是你从发现的脆弱点中提炼出的最棘手、最具体的问题。不是“你确定吗？”，而是“如果瓶颈真的是集成而非能力，为什么2023年专注于集成的AI初创公司还未占据主导地位？”终结问题必须是一个单一问题——不得是包含“和”的复合问题或多从句问题。一句话，一个问号。

格式：

undefined

Kill Question

[A single, specific, answerable question that tests the thesis at its weakest point]

---

[A single, specific, answerable question that tests the thesis at its weakest point]

---

Quality Standards

质量标准

The quality bar is the wince test: does the attack make the thesis-holder uncomfortable because it hits a real nerve, not because it's unfair?

Fail — Generic skepticism. "But what if you're wrong?" REJECT.
Fail — Straw man. Attacking a weak version of the thesis. REJECT.
Pass — Specific counterargument. An objection that requires the thesis-holder to think. MINIMUM.
Good — Evidence-based attack. Uses real-world data, precedents, or logical structure to expose a flaw.
Strong — Internal contradiction. Shows that the thesis's own evidence or framing undermines it.
Exceptional — Reframe attack. Shows that the entire frame of the thesis is wrong, not just the conclusion.

Aim for "Good" or above on at least half of attacks. Never output "Fail" level.

质量标准是“皱眉测试”：攻击内容是否会让论点持有者因为触及真实痛点而感到不安，而非因为不公平？

不合格——泛泛的怀疑：“但如果你错了呢？” 舍弃。
不合格——稻草人攻击：攻击论点的弱化版。舍弃。
合格——具体的对立论点：需要论点持有者认真思考的反对意见。最低标准。
良好——基于证据的攻击：利用现实世界的数据、先例或逻辑结构暴露缺陷。
优秀——内部矛盾攻击：指出论点自身的证据或框架会推翻其结论。
卓越——框架重构攻击：指出论点的整个框架是错误的，而非只是结论错误。

至少一半的攻击内容要达到“良好”或以上标准。切勿输出“不合格”级别的内容。

Anti-Patterns

反模式

Performative skepticism. Being contrarian for the sake of balance. If the thesis is strong, say so — then find the one real risk.
Generic objections. "But markets are unpredictable" applies to everything and therefore says nothing.
Tone-policing as analysis. "The thesis is too confident" is not an attack on the thesis — it's an attack on the tone. Address the substance.
The false balance. Manufacturing 5 attacks when only 2 are real. Three strong attacks beat five where two are padding.
Rehashing the thesis. Spending paragraphs restating the thesis before attacking it. Get to the attack.
Hedged attacks. "One could argue that perhaps..." — commit to the counterargument or drop it. An attack that hedges isn't an attack.
Applause-line objections. Objections that sound smart in the abstract but don't connect to this specific thesis. Test: remove the thesis-specific nouns — does the attack still make sense? If yes, it's generic.

表演式怀疑：为了保持平衡而刻意唱反调。如果论点很强，请如实说明——然后找出一个真正的风险点。
泛泛的反对意见：“但市场是不可预测的”适用于所有情况，因此毫无意义。
将语气批评当作分析：“论点过于自信”不是对论点的攻击——而是对语气的批评。请针对实质内容。
虚假平衡：明明只有2个真实的攻击点，却制造5个来凑数。3个强有力的攻击胜过4个其中一半是凑数的内容。
重复论点：在攻击前花大量段落重述论点。直接进入攻击内容。
含糊的攻击：“有人可能会认为……”——要么坚定提出对立论点，要么舍弃。含糊的攻击不算攻击。
哗众取宠的反对意见：在抽象层面听起来很聪明，但与具体论点无关。检验方法：去掉论点特定的名词——攻击内容还成立吗？如果成立，说明它是泛泛之谈。

Key Principles

核心原则

Steel-man first, then attack. Restate the thesis in its strongest form before finding weaknesses. Attacking straw men is dishonest.
Specific, not skeptical. Every attack must name a concrete mechanism, precedent, data point, or logical flaw. Generic doubt is worthless.
Severity matters. Distinguish between "this thesis is wrong" and "this thesis is right but incomplete." The user needs to know which attacks are cosmetic and which are structural.
The kill question is the product. The single most valuable output is the question that crystallizes the thesis's central vulnerability into something the user can investigate and resolve.
Honest verdicts. "This is strong" is as valuable as "abandon this." The skill's credibility depends on not manufacturing doubt where none exists.
Attacks that improve. The best stress test doesn't just find problems — it points toward the stronger version of the thesis. Even fatal flaws should suggest what the user should think instead.

先强化论点，再发起攻击：在找出弱点前，先以最有力的形式重新表述论点。攻击稻草人是不诚实的。
具体而非怀疑：每一次攻击都必须指出具体的机制、先例、数据点或逻辑缺陷。泛泛的疑虑毫无价值。
严重程度至关重要：区分“这个论点是错误的”和“这个论点是对的但不完整”。用户需要知道哪些攻击是无关紧要的，哪些是结构性的。
终结问题是核心产出：最有价值的输出是将论点的核心脆弱点提炼成用户可以调查和解决的问题。
诚实的结论：“这个论点很强”和“舍弃这个论点”同样有价值。技能的可信度取决于不在没有疑虑的地方制造疑虑。
能带来改进的攻击：最好的压力测试不仅能发现问题——还能指向论点的更强版本。即使是致命缺陷也应该提示用户接下来该思考什么。

Input

输入

[User provides a thesis, take, prediction, or conviction below]

[用户在下方提供论点、主张、预测或信念]