asb-carol-define

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Define Carol: The Ideal Customer Everything Aims At

定义Carol：所有工作的核心理想客户

Carol may not exist in her purest form — she is an ideal, artificially constructed from the most favorable characteristics. But she is not a fantasy: every marker in her definition is derived from what the previous steps established about who needs an extreme version of your strengths, who is disqualified by your weaknesses, and what moment turns fit into purchase. This skill assembles that definition, holds every line to an actionability bar, and writes the one file the rest of the company works from.

Carol可能并非以纯粹的形态存在——她是一个理想模型，由最有利的特征人工构建而成。但她并非空想：定义中的每一项特征都源自前期步骤所确定的内容，即谁需要你的极致优势、谁会因你的劣势被排除、哪个时刻会让潜在客户转化为实际购买者。本工具会整合这些定义，确保每一条内容都具备可操作性，并生成一份全公司后续工作的核心文件。

The mental model

心智模型

The five-step derivation, completed here

五步推导法，在此完成最后一步

Strengths and weaknesses — an honest accounting of who you are.
Keystones — customer circumstances that require an extreme version of a strength, with the segments that typify them.
Deal-breakers — the disqualifiers, the anti-market, and the honing qualifiers they force onto keystone segments.
Inciting events — what moves a keystone-fit customer from could-buy to buying-today.
Carol — this step: the composite of keystones, deal-breakers, and inciting events, expressed as behavioral and attitudinal characteristics a team can act on.

优势与劣势——对自身的诚实评估。
核心要素——需要极致优势的客户场景，以及代表这些场景的细分群体。
否决项——排除标准、非目标市场，以及由此对核心细分群体产生的细化筛选条件。
触发事件——促使符合核心要素的客户从“可能购买”转向“立即购买”的契机。
Carol——本步骤：整合核心要素、否决项和触发事件，以团队可落地执行的行为和态度特征呈现。

Beyond demographics

超越人口统计学特征

Traditional market descriptions — company size, geography, age, B2B-vs-B2C — usually misdescribe the target. The sitcom whose demographic profile said "women over forty-five" was meanwhile becoming a cult hit with young men who loved sharp, character-driven insult comedy; the demographic was true of the average viewer and useless for finding the next one. An email client's real target isn't a company size or a country — it's people who get hundreds of mails a day, or people with assistants who process mail, or people running systematic outbound. What defines a target market is what the best customers actually share: the shape of the keystones, deal-breakers, and inciting events. Demographics enter only when genuinely determinative — back-office software for dentists in Norway may honestly say "dentists in Norway." The test is predictive power: does the marker separate the best customers from the worst, or does it merely describe everyone politely?

传统的市场描述——公司规模、地域、年龄、B2B或B2C——通常会错误描述目标客户。某情景喜剧的人口统计特征显示“面向45岁以上女性”，但实际上它深受喜爱尖锐、角色驱动型讽刺喜剧的年轻男性追捧；人口统计特征描述的是平均受众，但对寻找下一批受众毫无用处。某邮件客户端的真正目标并非某一公司规模或国家——而是每天收到数百封邮件的人、有助理处理邮件的人，或是开展系统性 outbound 业务的人。定义目标市场的关键是最佳客户真正共有的特征：核心要素、否决项和触发事件的具体形态。只有当人口统计学特征具有真正的决定性时才会纳入——例如挪威牙医使用的后台软件，确实可以描述为“挪威的牙医”。测试标准是预测能力：该特征能否区分最佳客户与最差客户，还是仅仅是对所有人的礼貌描述？

A menu of non-demographic dimensions

非人口统计学维度参考清单

Demographics are the lazy axes; the dimensions below are the kind that actually separate the best customers from the worst. Use the menu two ways: to test a candidate marker — "speed" is not a segment, so ask which of these it really is (an emergency-vs-deliberate trigger? a decision-cycle difference? a problem-ownership one?) — and to name the axis on which two segments diverge before deciding whether one Carol can honestly span them. These are examples, not the full universe. Treat them two ways at once: use the ones that fit the context directly, and take the rest as models for generating others of the same kind that this specific business demands — the sharpest marker is frequently a dimension no list contains. This is a starter set to prime the thinking, never a checklist to complete and never the boundary of what counts; in any given case most won't apply, and the honest answer is usually "don't know" or "don't care." (Drawn from Choosing a target market, which offers these as a primer, not an inventory.)

The person and their role

Individual vs team — one product can face two markets; sometimes "both" (a free single-player mode hooks people, teams are where the money is — Asana, Notion).
Role inside the company — customer-facing or inward-facing; what others expect of them.
Title(s) — usually a variety, especially across industries or company sizes.
Maker vs administrator — creating/making/delivering, or managing/monitoring/analyzing/reporting?
Problem ownership — who "owns" the problem you solve; is it one role or distributed; is the problem-owner also the solution-owner?
Technical acumen — from "engineers who argue about algorithms" to "technophobes reluctantly using devices they hate."
Learning style — video, docs, 1:1 training, or figure-it-out; shapes onboarding, support, and marketing.

What success means to them

Job-to-be-done success — what objective numbers they're held to; which are vital vs operational, satisfied vs maximized; what number changing would get them promoted or fired.
Personal success — what fulfills them, what they'll tolerate friction for, what makes them advocate for you internally (their Needs Stack).
Professional success — how they're evaluated and promoted; early- vs mid- vs late-career priorities differ regardless of company size.
Corporate goals — revenue growth rate, revenue size, profit dollars, profit percent, market share, Rule-of-40, cost savings, GPM.

The organization

Organization size — SMB is often bought by larger but not vice-versa; matters most selling top-down or when the org itself is the subject (e.g. HR software).
Growth trajectory — hyper-growth, steady-state, or declining; each creates different priorities and constraints.
Regulatory environment — how heavily regulated, and by what; drives feature needs, compliance, and risk tolerance.
Cultural attributes — productivity vs work-life balance; ethical posture; external innovation vs internal efficiency.
Crossing-the-Chasm phase — innovator (tinkerer), early adopter (risk for advantage), early majority (wary, needs the whole product), late majority (dragged in by pressure).
Risk tolerance — new markets require risk-accepting customers; uncorrelated with company size.
Tech-stack philosophy — open-source, proprietary, cloud-native, or on-prem; usually a values stance, not just a technical need.

How they buy and decide

Budget type — fixed/annual, flexible, or find-money-when-needed; is there a threshold where the sale changes shape?
Sales process — top-down vs bottom-up/PLG/freemium (top-down often correlates with size; bottom-up often doesn't).
Decision cycle — hours, weeks, or quarters.
Emergency vs deliberate — an urgent trigger (a security breach) vs a researched, compared, piloted decision.
Business model — subscription, one-time, transaction, usage, freemium, ad-supported; dictates the metrics they care about and how they can pay.

Their tools and workflow

Integrations — what your product must connect to (workflow, AI, to-do, project management, chat).
Other software used daily — what characterizes their expectations and experience.
Processes — the workflows they use or aspire to, especially where they intersect your product.
Work style — remote/office/hybrid; time-zone-split; hours worked; synchronous vs asynchronous.
Update cadence — frequent-incremental vs infrequent-substantial; shapes product rhythm.
Communication culture — email-heavy, Slack-dependent, or meeting-oriented.

人口统计学特征是偷懒的划分方式；以下维度才是真正能区分最佳客户与最差客户的类型。使用这份清单的两种方式：一是测试候选特征——“速度”并非细分群体，因此要明确它实际属于哪一类（是紧急vs审慎的触发因素？决策周期差异？还是问题归属差异？）；二是命名两个细分群体的分歧维度，再判断单个Carol能否真正覆盖这两个群体。这些只是示例，并非全部。 同时以两种方式使用：直接采用符合场景的维度，其余则作为模板，生成适用于特定业务的其他同类维度——最精准的特征往往不在任何清单中。这只是启动思考的入门工具，绝非需要完成的检查清单，也绝非边界限制；在任何特定场景下，大多数维度都不适用，诚实的答案通常是“不知道”或“不关心”。（摘自*选择目标市场*，该文章将这些内容作为入门介绍，而非完整清单。）

个人及其角色

个人vs团队——一款产品可能面向两个市场；有时是“两者兼顾”（免费单人模式吸引用户，团队版本创造营收——如Asana、Notion）。
公司内部角色——面向客户或内部服务；他人对其的期望是什么。
职位头衔——通常多种多样，尤其是跨行业或不同公司规模的情况。
创造者vs管理者——负责创造/制作/交付，还是管理/监控/分析/汇报？
问题归属——谁“负责”你要解决的问题；是单个角色还是多个角色共同负责；问题负责人是否也是解决方案负责人？
技术能力——从“争论算法的工程师”到“勉强使用讨厌设备的技术恐惧症患者”。
学习风格——视频、文档、一对一培训，还是自行摸索；影响入职培训、支持服务和营销方式。

他们眼中的成功

待办任务的成功标准——他们需要达成的客观指标；哪些是关键指标vs运营指标，满足即可vs需要最大化；哪些指标的变化会让他们获得晋升或被解雇。
个人成功——什么能满足他们，他们愿意为了什么忍受摩擦，什么会让他们在内部为你宣传（他们的需求层级）。
职业成功——他们的评估和晋升标准；无论公司规模如何，职业生涯早期、中期、晚期的优先级都有所不同。
企业目标——营收增长率、营收规模、利润额、利润率、市场份额、Rule-of-40、成本节约、毛利率（GPM）。

组织特征

组织规模——中小企业（SMB）的客户通常会购买更大型的产品，但反之则不然；当采用自上而下销售模式，或产品本身以组织为服务对象时（如HR软件），组织规模最为重要。
增长轨迹——高速增长、稳定状态或衰退；每种状态都会产生不同的优先级和约束条件。
监管环境——监管力度如何，由谁监管；驱动功能需求、合规要求和风险承受能力。
文化属性——生产力vs工作与生活平衡；道德立场；外部创新vs内部效率。
跨越鸿沟阶段——创新者（探索者）、早期采用者（为获取优势愿意承担风险）、早期大众（谨慎，需要完整产品）、晚期大众（因压力被迫接受）。
风险承受能力——新市场需要愿意承担风险的客户；与公司规模无关。
技术栈理念——开源、专有、云原生或本地部署；通常是一种价值立场，而非单纯的技术需求。

他们的购买与决策方式

预算类型——固定/年度预算、灵活预算，或按需找预算；是否存在改变销售模式的阈值？
销售流程——自上而下vs自下而上/产品主导增长（PLG）/免费增值模式（自上而下通常与规模相关；自下而上则通常无关）。
决策周期——数小时、数周或数月。
紧急vs审慎——紧急触发因素（如安全漏洞）vs经过研究、对比、试点的决策。
商业模式——订阅制、一次性付费、交易制、按使用量付费、免费增值、广告支持；决定他们关心的指标和支付方式。

他们的工具与工作流程

集成需求——你的产品必须对接的工具（工作流、AI、待办事项、项目管理、聊天工具）。
日常使用的其他软件——决定他们的期望和使用体验。
流程规范——他们当前使用或期望采用的工作流程，尤其是与你的产品相关的部分。
工作方式——远程/办公室/混合模式；跨时区；工作时长；同步vs异步。
更新频率——频繁增量更新vs不频繁大幅更新；影响产品迭代节奏。
沟通文化——重度依赖邮件、Slack，或频繁开会。

The keep-asking-why chain

持续追问“为什么”的链条

The first answer is rarely the marker. "Our best customers worry about speed" — speed is a feature, not a segment. Ask why, and keep asking: speed → because for e-commerce, checkout speed is revenue; speed → because for media, pages-per-session is ad revenue; speed → because for B2B landing pages, a whole budget converges on a few thousand visits. The chain ends at a person: someone for whom a faster website directly increases revenue, and who can calculate the dollar value of a hundred milliseconds. That's a definition Marketing can target, Sales can qualify, and Product can build features to thrill — the bar every marker in CAROL.md must clear.

第一个答案很少是真正的特征。“我们的最佳客户关心速度”——速度是功能，而非细分群体。要追问为什么，并持续追问：速度→因为对于电商来说，结账速度直接影响营收；速度→因为对于媒体来说，每会话页面数影响广告收入；速度→因为对于B2B着陆页来说，全部预算都集中在数千次访问上。追问链条最终指向具体的人：那些网站速度直接影响营收，且能计算出100毫秒价值的人。 这才是营销团队可定位、销售团队可筛选、产品团队可打造功能让其满意的定义——CAROL.md中的每一项特征都必须达到这个标准。

The bullseye: why only-Carol wins everyone

靶心：为何只瞄准Carol就能赢得所有客户

Targeting Carol alone feels like shrinking the market; it is the opposite. Around Carol sits a ring of customers who value most of the same trade-offs and are merely indifferent to the rest — roughly ten times as many. Around them, a far larger ring who weigh the trade-offs and, because you stated yours clearly and confidently, conclude you're their best option — up to a hundred times as many. Clear trade-offs, confidently stated, are how people actually buy; generic positioning aimed at everyone is what loses all three rings at once. If the message can't excite even Carol, it certainly won't convince anyone else. This is the strategic case the definition file states, briefly, so every reader knows why the aim is so narrow.

只瞄准Carol看似缩小了市场范围，实则相反。Carol周围是一圈客户，他们重视大部分相同的取舍，对其余部分则漠不关心——数量约为Carol群体的10倍。再往外是更庞大的一圈客户，他们会权衡取舍，由于你清晰自信地表明了自身的取舍，他们会认定你是最佳选择——数量可达Carol群体的100倍。清晰明确且自信的取舍才是客户实际的购买依据；面向所有人的通用定位会同时失去这三圈客户。如果信息连Carol都无法打动，那肯定无法说服其他人。定义文件会简要阐述这一战略逻辑，让每位读者都明白为何目标如此精准。

Diverging segments force a choice

分歧群体必须做出选择

When the honed keystones name genuinely different segments — solo shops and multi-location groups, say, wanting different things at different prices — Carol cannot be their average: a definition set at the midpoint describes nobody who exists. Sometimes the why-chain reveals a shared circumstance underneath and one Carol spans the segments honestly. When it doesn't, the user must choose the primary — informed by where the keystones concentrate, where the observed inciting events are, which segment the profit evidence favors — and the definition says plainly who was chosen and who remains a served-but-secondary market. Choosing is the user's call; refusing to choose (or hiding the choice in vague wording) defeats the exercise.

当打磨后的核心要素指向真正不同的细分群体时——比如个体店铺和多地点连锁店，需求和价格都不同——Carol不能是这些群体的平均值：处于中间点的定义无法描述任何真实存在的客户。有时追问“为什么”的链条会揭示群体背后的共同场景，单个Carol就能真正覆盖这些群体。如果没有共同场景，用户必须选择一个主要群体——依据核心要素的集中度、观察到的触发事件、利润数据倾向的群体——定义中会明确说明选择了谁，以及哪些是次要服务市场。选择由用户决定；拒绝选择（或用模糊措辞掩盖选择）会使整个练习失去意义。

Vocabulary

术语定义

Carol — the ideal customer: a composite, possibly not existing in pure form, whose markers are all derived from the working files.
Marker — one behavioral, attitudinal, or circumstantial characteristic in the definition; each must be targetable, qualifiable, and buildable-for.
Qualifiers / disqualifiers — the questions and signals Sales uses to sort a prospect toward or away from Carol, drawn from keystones and deal-breakers.
Primary segment — when segments diverge, the one Carol is defined for; the others are named as secondary, not blended in.

Carol——理想客户：一个复合模型，可能并非以纯粹形态存在，其所有特征均源自工作文件。
特征（Marker）——定义中的一项行为、态度或场景特征；每一项都必须可定位、可筛选、可围绕其打造功能。
筛选条件/排除条件——销售团队用于判断潜在客户是否符合Carol标准的问题和信号，源自核心要素和否决项。
主要细分群体——当群体出现分歧时，Carol所定义的目标群体；其他群体被列为次要群体，而非混合纳入。

The definer's posture

定义者的原则

Be clear, not clever

清晰直白，而非卖弄技巧

Write to be understood, not admired. The work here wrestles with hard concepts, and clever metaphors, wordplay, or cute turns of phrase make them harder to grasp, not easier. Say plainly what you mean. If a sentence reads more clearly without a flourish, cut the flourish. State the actual point rather than gesturing wittily at it.

写作的目的是让人理解，而非让人赞赏。这里的工作涉及复杂概念，巧妙的隐喻、文字游戏或俏皮表达会让概念更难理解，而非更容易。直白地说出你的意思。如果去掉修饰后句子更清晰，就删掉修饰。直接陈述实际观点，而非巧妙地暗示。

Restate references; never cite a bare token

重述参考内容；切勿仅引用标记

When you mention a numbered or lettered item to the user — K4, W2, O17, H3, and the like — add a few plain words on what it actually is ("K4 — the owner whose career rides on the site"). A bare token is unreadable to a human who saw it defined hours or days ago: the tag is for traceability, the gloss is for comprehension. Keep the tag for accuracy; always add the gloss.

当向用户提及编号或字母标记——如K4、W2、O17、H3等——时，要添加几句直白的解释（“K4——职业生涯依赖网站表现的所有者”）。仅引用标记对于数小时或数天前看过定义的人来说难以理解：标记用于追溯来源，解释用于帮助理解。保留标记以确保准确性；始终添加解释。

Two readers, one file — definition first

两类读者，一份文件——定义优先

CAROL.md serves a skimming human (the definition, crisp, up top) and the downstream workers and tools (the reference sections, cited, below). The definition section is register-gated by exclusion: each marker is a bare, specific characteristic — no citations, no derivation story, no "which comes from our strength in…", no hedging. Wrong: "~ Values support quality (derived from S2, though segments differ)." Right: "Runs payroll personally, nights and weekends, with no office staff." The derivation lives in the reference sections, where every keystone, deal-breaker, and event carries its [K]/[D]/[E] numbers back to the working files.

CAROL.md服务两类读者：快速浏览的人（位于顶部的清晰定义）和下游工作人员及工具（下方的参考部分）。定义部分采用排除法筛选内容：每一项特征都是明确具体的特征——无引用、无推导过程、无“源自我们的某项优势”、无含糊其辞。错误示例：“~重视支持质量（源自S2，但群体存在差异）。” 正确示例：“亲自处理薪资，在夜间和周末完成，无办公室员工协助。” 推导过程位于参考部分，每一项核心要素、否决项和事件都带有[K]/[D]/[E]编号，可追溯至工作文件。

Synthesize; don't re-litigate

整合内容；勿重新争论

The working files already survived their own gates — don't reopen settled classifications or re-press honed segments. The work here is assembly, compression through the why-chain, and the actionability bar. Exception: when synthesis exposes a genuine contradiction between files (a keystone segment the anti-market swallowed; an inciting event pointing at nobody), surface it as a finding and, if it requires upstream rework, say which file and stop short of editing it silently.

工作文件已经过审核——无需重新讨论已确定的分类或重新打磨细分群体。此处的工作是整合、通过追问链条压缩内容，以及确保可操作性。例外情况：当整合过程中发现文件之间存在真正的矛盾（核心细分群体被非目标市场覆盖；触发事件指向无对应群体），需将其作为发现点提出，如果需要上游文件修改，要说明涉及哪个文件，切勿悄悄编辑。

Press every marker through the bar

确保每一项特征都达到标准

For each candidate marker, check aloud when in doubt: could Marketing target this (an ad channel, a search phrase, a list)? Could Sales qualify it in one or two questions? Could Product name a feature that thrills this person specifically? A marker failing all three is description, not definition — press it through the why-chain (testing it against the dimensions menu) or drop it. A predefined way to press harder: if a devil's-advocate interrogation skill is installed in the environment (for example Rude Q&A /

asb-rude-qa

, from the same author as this method), invoke it against the draft definition with this brief: attack this ideal-customer definition — find every marker that is demographic laziness, every one too generic to separate the best customers from the worst, every one that no ad could target and no salesperson could qualify, and any place two diverging segments were averaged instead of chosen. Don't accept wishful or hand-wavy defenses. If no such skill is available, run that interrogation yourself, visibly, as a review pass.

对于每一项候选特征，如有疑问可大声检查：营销团队能否针对其开展定位（广告渠道、搜索关键词、客户列表）？销售团队能否通过1-2个问题进行筛选？产品团队能否命名一个让这类客户满意的功能？如果一项特征全部不满足，那只是描述，而非定义——通过追问链条（对照维度清单测试）优化，或直接删除。一种更严格的预定义方式：如果环境中安装了“魔鬼代言人”审核工具（例如同一作者开发的Rude Q&A /

asb-rude-qa

），可针对 draft 定义发起审核，要求如下：攻击这份理想客户定义——找出所有偷懒的人口统计学特征、所有过于宽泛无法区分最佳与最差客户的特征、所有无法通过广告定位或销售筛选的特征，以及所有将两个分歧群体取平均值的地方。不接受一厢情愿或含糊其辞的辩解。如果没有此类工具，可自行进行审核，作为可见的审查环节。

The choice is theirs; the refusal to average is not

选择由用户决定；但绝不允许取平均值

Where segments diverge, present the evidence for each candidate primary (keystone concentration, observed events, profit signals from the strengths file's differentials) and make the user choose — one honest press if the choice contradicts their evidence, then their call stands, recorded with its reasoning. What never stands: a definition worded vaguely enough to cover both ("growing businesses that value efficiency") — that's the averaging this method exists to prevent, and it doesn't get written, however phrased.

当群体出现分歧时，需向用户展示每个候选主要群体的依据（核心要素集中度、观察到的事件、优势文件中的利润差异信号），并让用户做出选择——如果选择与依据矛盾，可诚恳提出一次意见，之后用户的决定即为最终结果，并记录其理由。绝不允许：使用模糊措辞覆盖两个群体的定义（“重视效率的成长型企业”）——这正是本方法要避免的取平均值行为，无论措辞如何，都不得写入。

How to use this skill

如何使用本工具

Phase A — Ingest

阶段A——导入

Read the four working files (defaults, in the current directory:

STRENGTHS-WEAKNESSES.md

KEYSTONES.md

DEALBREAKERS.md

INCITING-EVENTS.md

). The minimum viable input is a keystones file with honed segments plus deal-breakers; missing inciting events weaken the definition's buying-moment section (say so; the section notes the gap); missing strengths/weaknesses weakens the trade-offs framing. If even the minimum is absent, don't define from vibes: name the upstream steps invoke-if-installed style (the first is Observations /

asb-carol-observations

) and offer only a provisional sketch, clearly labeled, if the user insists. Check the keystones change log for honing; unhoned segments mean the definition inherits over-broad markers — flag it. If a CAROL.md exists: in-progress header means resume at the reviewed-through pointer; complete means ask whether to revise.

Output:

CAROL.md

in the same directory. If the input was pasted and no path is known, ask where the method's files should live before creating anything (default: the current directory) — never scatter files silently.

读取四个工作文件（默认位于当前目录：

STRENGTHS-WEAKNESSES.md

、

KEYSTONES.md

、

DEALBREAKERS.md

、

INCITING-EVENTS.md

）。最低要求输入是带有打磨后细分群体的核心要素文件，加上否决项；缺少触发事件会削弱定义中的购买契机部分（需说明这一点；该部分会标注缺口）；缺少优势/劣势文件会削弱取舍框架。如果连最低要求都不满足，请勿凭感觉定义：指明上游步骤（第一步是Observations /

asb-carol-observations

），如果用户坚持，仅提供明确标注为临时草稿的内容。检查核心要素文件的变更日志，确认是否经过打磨；未打磨的细分群体意味着定义会继承过于宽泛的特征——需标记出来。如果已存在CAROL.md：文件头部标注“进行中”意味着从已审核到的部分继续；标注“完成”意味着询问是否需要修订。

输出：在同一目录生成

CAROL.md

。如果输入是粘贴的内容且未知路径，需先询问方法文件应存储在何处（默认：当前目录）——切勿随意分散文件。

Phase B — Synthesize and draft whole

阶段B——整合并完整起草

Sweep silently: cluster the honed keystone segments (do they share a circumstance, or diverge?); run the why-chain on each candidate marker; harvest qualifiers/disqualifiers from keystones and deal-breakers; pull the buying-moment material from the events file; note contradictions between files. If segments diverge, STOP drafting and run the choice with the user first — the primary decision shapes every line, so it can't wait for review — and record the decision to disk the moment it's made (a one-line CAROL.md stub with the header is enough), so a died session doesn't lose it. Once recorded, the choice doesn't reopen through vague re-wording ("can't it just cover both?") — refuse the averaging; offer to rerun the choice against the evidence if the user genuinely wants to switch primaries.

Then write the complete draft to disk with the in-progress header — the file is the memory — and present it whole for review, a section or two per exchange, top-to-bottom in file order (the reviewed-through pointer assumes that order; a resumed session relies on it). Keep the pointer current on disk after each pass. Before finalization, run the devil's-advocate press (delegated or self-run, per the posture) as a visible review pass over the definition — not silently. Corrections of fact go against the working files (the files win over memory — a resumed session re-reads them for exactly this reason); wording and emphasis are the user's; markers move only through the bar. When a review comment implies an upstream file is wrong (an anti-market segment recast as secondary, an event's status remembered differently), the same rule as synthesis contradictions applies: say which file it belongs to and stop short of editing it silently — CAROL.md never quietly overrides its sources.

先进行静默梳理：将打磨后的核心细分群体聚类（是否共享场景，还是存在分歧？）；对每个候选特征执行追问链条；从核心要素和否决项中提取筛选/排除条件；从事件文件中提取购买契机内容；标记文件之间的矛盾。如果群体存在分歧，停止起草，先让用户做出选择——主要群体的决定会影响每一条内容，因此不能等到审查阶段再做——并在做出决定后立即记录到磁盘（只需在CAROL.md中写入一行头部内容作为草稿即可），避免会话中断导致丢失。记录后，不得通过模糊措辞重新开启选择（“能不能同时覆盖两个群体？”）——拒绝取平均值；如果用户确实想更换主要群体，可重新依据证据进行选择。

然后将完整草稿写入磁盘，并标注“进行中”头部——文件即为记忆载体——然后完整呈现给用户审查，每次交流展示1-2个部分，按照文件顺序从上到下（已审核到的部分指针基于此顺序；恢复会话时依赖此顺序）。每次审查后更新磁盘上的指针。定稿前，对定义进行“魔鬼代言人”审核（委托工具或自行执行，遵循上述原则），作为可见的审查环节——切勿静默进行。事实修正需对照工作文件（文件优先于记忆——恢复会话时会重新读取文件正是出于此原因）；措辞和侧重点由用户决定；特征仅能通过标准测试进行调整。当审查意见暗示上游文件存在错误（非目标市场群体被重新归类为次要群体，事件状态记忆有误），需遵循整合矛盾时的规则：说明涉及哪个文件，切勿悄悄编辑——CAROL.md绝不悄悄覆盖其来源文件。

The file structure

文件结构

markdown

undefined

markdown

undefined

Carol — <company / project name>

Carol — <公司/项目名称>

⚠️ IN PROGRESS — review not started; begin at the definition. (Later: "draft under review; reviewed through <section>." Removed at finalization.)

⚠️ 进行中 — 尚未开始审查；从定义部分开始。（后续：“草稿审查中；已审核至<section>部分。”定稿时移除。）

<一行内容：整合自哪些文件，截至何时。如果选择了主要群体而非其他群体，用一行说明选择结果。>

Carol

<Marker: circumstance — "runs 3+ locations with staff floating between them">
<Marker: behavior — "does payroll personally, Sunday nights">
<Marker: attitude — "treats compliance as an existential risk, not paperwork">
<Marker: economics — "values one prevented violation above a year of fees">

<定义部分。一段简短的文字描述（2-4句话），随后列出各项特征——每一项都是明确具体的行为/态度/场景特征。此处无引用、无推导、无含糊其辞。当数字可量化预测场景时可使用（“20名及以上拿小费的员工”——该数字即为核心要素阈值）；但不得将数字作为替代指标（“营收200万美元以上”——这是列表构建的启发式方法，而非特征；需追问其代表的场景）。>

<特征：场景——“运营3家及以上门店，员工在门店间流动”>
<特征：行为——“亲自处理薪资，在周日夜间完成”>
<特征：态度——“将合规视为生存风险，而非例行文书工作”>
<特征：经济因素——“重视避免一次违规，胜过一年的费用”>

How to recognize Carol (qualifiers and disqualifiers)

如何识别Carol（筛选与排除条件）

Qualify toward: <the one-or-two-question sorts, from the keystones — "do you take the tip credit, and in which states?"> Disqualify away: <the anti-market signals, from the deal-breakers.>

筛选条件：<1-2个问题，源自核心要素——“你是否享受小费抵扣，所在州是哪些？”> 排除条件：<非目标市场信号，源自否决项。>

Why aiming only at Carol wins the whole market

为何只瞄准Carol就能赢得整个市场

<3-4句话：靶心逻辑——约10倍规模的群体共享她的取舍，约100倍规模的群体权衡取舍后因你清晰的选择而选择你；如果信息无法打动Carol，就无法说服任何人。如果选择了主要群体，需说明仍作为次要服务市场的群体。>

Keystones (reference)

核心要素（参考）

<每行一条，带有[K-编号]——哪些场景需要哪些优势，以及打磨后的细分群体。>

Deal-breakers and anti-market (reference)

否决项与非目标市场（参考）

<每行一条，带有[D-编号]；保留非目标市场说明——明确排除、劝阻性广告语言、不为长尾群体开发功能。>

Inciting events (reference)

触发事件（参考）

<每行一条，带有[E-编号]和状态（已观察/假设）——购买契机，用一两个词说明可识别性。如果事件文件缺失或内容单薄：标注缺口，并说明验证路径。>

Provenance

来源

undefined

<哪些文件，截至哪些日期/变更日志状态；整合过程中发现的矛盾及处理方向。>

undefined

Phase C — Finalize

阶段C——定稿

Finalize when every section is reviewed or the user explicitly waives the remainder. Remove the header and read the definition back one final time — it is the part everyone will quote. Close with the handoff: this file is the input to what follows — positioning speaks Carol's language, marketing buys her attention, sales qualifies against her markers, product builds what thrills her — and each reference section is where that work starts. If the definition leans on hypothesized inciting events or unvalidated hypotheses, say so: the companion customer-interview method (its first skill is

asb-interview-goals

, if installed) is the systematic way to test the definition against real customers — its first canonical goal is literally "What does Carol look like?"

当所有部分都已审查，或用户明确表示跳过剩余部分时，即可定稿。移除头部内容，最后重读一遍定义——这是所有人都会引用的部分。最后交接说明：这份文件是后续工作的输入——定位文案要使用Carol的语言，营销要吸引她的注意力，销售要依据她的特征筛选客户，产品要打造让她满意的功能——每个参考部分都是后续工作的起点。如果定义依赖假设的触发事件或未验证的假设，需说明这一点：配套的客户访谈方法（第一步是

asb-interview-goals

，如果已安装）是系统地将定义与真实客户进行验证的方式——其首要标准目标正是“Carol的真实形态是什么？”

Refusal conditions

拒绝执行的情况

No upstream files. A Carol defined from vibes is a persona poster, not an instrument. Point at the method's first step; offer only a clearly-labeled provisional sketch if pressed.
Demographic laziness. "SMBs," "millennials," "companies over $1M" don't enter the definition unless genuinely determinative — press through the why-chain to the circumstance that predicts buying, or drop the marker.
Averaged segments. When keystone segments diverge, wording vague enough to cover both is never written. The choice is made, named, and reasoned — or the divergence is recorded as an open decision and the definition waits.
Markers that fail the bar. A characteristic no ad can target, no salesperson can qualify, and no feature can thrill is not recorded, however true it feels.
Doing the downstream work. Positioning statements, ad copy, homepage language, feature roadmaps — declined; the reference sections are their input, and each deserves its own session.
Skipping the synthesis. "Just polish the persona we already have" without the working files gets the honest answer: this method derives Carol from evidence; a polish of an underived persona inherits its guesses. Offer to run the derivation against whatever files exist.

无上游文件。凭感觉定义的Carol只是一个人物海报，而非实用工具。指明方法的第一步；如果用户坚持，仅提供明确标注为临时草稿的内容。
偷懒的人口统计学特征。“中小企业”“千禧一代”“营收100万美元以上的公司”等内容，除非具有真正的决定性，否则不得纳入定义——通过追问链条找到预测购买行为的场景，或删除该特征。
群体取平均值。当核心细分群体出现分歧时，绝不允许写入模糊到覆盖两个群体的措辞。必须做出选择、明确说明并记录理由——或记录分歧为未决决策，暂停定义工作。
未达标准的特征。无论感觉多么真实，只要无法通过广告定位、无法通过销售筛选、无法围绕其打造功能的特征，都不得记录。
执行下游工作。定位文案、广告文案、首页内容、功能路线图——均拒绝执行；参考部分是这些工作的输入，每项工作都应单独开展会话。
跳过整合步骤。“只需优化我们已有的客户画像”但无工作文件时，需如实告知：本方法基于证据推导Carol；优化无依据的客户画像会继承其猜测内容。可提议依据现有文件执行推导步骤。