Back to Details

insurgent-campaign

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Insurgent Campaign

反叛式营销活动

A campaign-design skill for organizations and people who cannot win by outspending. It ideates campaign concepts across every major archetype, audits the user's spend asymmetry against their competition, assembles an insurgent channel stack, sets strict boost gates on any paid spend, and produces a lift-test plan so the user measures incremental impact rather than vanity metrics.
一款专为无法靠砸钱取胜的组织和个人打造的活动设计技能。它能针对各类主流活动原型构思活动概念，审核用户与竞争对手之间的预算差距，搭建一套反叛式渠道体系，对任何付费支出设置严格的推广门槛，并制定效果测试计划，让用户衡量增量影响而非虚荣指标。

Core Premise

核心前提

Paid media scales attention. Organic narrative and grassroots networks scale trust. When trust is the bottleneck — and in 2025–2026, for most underdogs, it is — additional ad spend hits diminishing and then negative returns. Saturation, inauthenticity, and narrative incoherence produce reactance, not persuasion. The leverage point is not reach; it is credibility and message-market fit.
Empirical spine:
  • Academic field experiments show digital ad ROI confidence intervals are so wide that most campaigns cannot be statistically distinguished from zero lift (Lewis & Rao, "The Unfavorable Economics of Measuring the Returns to Advertising," Quarterly Journal of Economics 2015). Meta and Google both ship Conversion Lift / Brand Lift tools precisely because they distinguish attributed conversions from incremental conversions — the platforms' own tooling is the tacit admission that dashboard ROAS is not causal.
  • Across 18,000+ brands, CPA rose in 13/14 industries in 2025; ROAS fell in 13/14; conversion rates fell in 13/14. More money is buying fewer results.
  • Facebook organic engagement sits near 0.15%, Instagram ~0.50%, X ~0.15%, TikTok ~2.50% — still the best organic window but compressing fast.
  • Meta and Google both ended political/issue ads in the EU in October 2025, removing paid channels entirely for large categories and forcing organic to carry the campaign.
  • Under the most extreme spending asymmetry imaginable (a grassroots party vs. a €4B state-backed communications apparatus with foreign support), the side with near-zero paid media won a parliamentary supermajority in Hungary in April 2026 — see 
    references/hungarian-case-study.md
    .
The same curve bends commercial advertising. The skill treats paid as an amplifier of proven organic winners, never as a standalone channel.
付费媒体扩大触达范围，而有机叙事和草根网络建立信任。 当信任成为瓶颈时——在2025-2026年，大多数弱势参与者都面临这一问题——额外的广告投入会先出现边际收益递减，而后产生负收益。信息饱和、不真实性和叙事不一致会引发受众抵触，而非说服。杠杆点不在于触达，而在于可信度和信息与市场的匹配度。
实证依据：
  • 学术实地实验表明，数字广告ROI的置信区间非常宽泛，大多数活动的效果在统计上与零提升无显著差异（Lewis & Rao，《衡量广告回报的不利经济学》，《经济学季刊》2015年）。Meta和Google推出Conversion Lift/Brand Lift工具，正是因为它们能区分归因转化和增量转化——平台自身的工具就默认承认了后台ROAS数据不具备因果性。
  • 2025年，在18000多个品牌中，14个行业里有13个行业的CPA上升、ROAS下降、转化率下降。投入更多资金，得到的结果却更少。
  • Facebook的自然互动率约为0.15%，Instagram约0.50%，X约0.15%，TikTok约2.50%——虽然仍是最佳自然触达窗口，但正在快速收窄。
  • 2025年10月，Meta和Google在欧盟停止了政治/议题广告，彻底取消了大类别的付费渠道，迫使活动只能依赖有机触达。
  • 在极端预算不对称的情况下（草根政党vs获得外国支持、预算达40亿欧元的官方宣传机构），几乎零付费媒体投入的一方在2026年4月匈牙利议会选举中赢得了绝对多数席位——详见
    references/hungarian-case-study.md
    。
同样的规律也适用于商业广告。该技能将付费视为已验证有机内容的放大器，而非独立渠道。

When to Apply This Skill

适用场景

Apply when the user:
  • Asks to plan any campaign (marketing, launch, fundraising, mobilization, awareness, turnout).
  • Describes being outspent by a competitor, incumbent, or adversary.
  • Is considering ad spend, boosts, influencer deals, or scaling a paid channel.
  • Is watching their organic reach collapse and wondering what to do.
  • Is preparing for a launch, election, fundraise, product push, or cause-led moment.
  • Is deciding between paid and organic allocation, or between channels.
Also apply when no explicit "campaign" is named but the user is debating where to invest time and money for distribution.
当用户出现以下情况时适用：
  • 要求规划任何活动（营销、发布、筹款、动员、品牌认知、投票动员）。
  • 描述自己被竞争对手、行业巨头或对手的预算碾压。
  • 考虑投入广告、推广帖子、网红合作或扩大付费渠道规模。
  • 发现自然触达率下降，不知如何应对。
  • 准备进行发布、选举、筹款、产品推广或主题活动。
  • 在付费和有机渠道之间分配资源，或选择渠道时犹豫不决。
即使未明确提及「活动」一词，但用户在讨论时间和资金的投放方向时，也可适用。

Workflow

工作流程

The skill runs in seven stages (1, 2, 3, 3a, 4, 5, 5b). Do not skip stages; later stages depend on the user's answers and selections from earlier ones. Stage 3a (MMF Gate) can refuse the full campaign plan and route the user to validation work instead. Stage 5 ends with a shape selection — Stage 5b only begins once the user has picked one of the three shapes.
This skill is designed to be portable across agent environments. Where the workflow says "ask the user," use whatever structured-question facility your environment provides (in Claude Code that is the 
AskUserQuestion
tool, batching up to 4 questions per call); if no structured tool exists, ask in plain text and wait for the user to reply before continuing. Either way, never invent answers.
Resume Protocol (idempotency). Before producing any output, scan the conversation for prior skill output. If a section's heading (
## N. <name>
) is already present in the conversation, do not regenerate it. Pick up at the first stage whose Contract preconditions are satisfied AND whose output section(s) are not yet in the conversation. If the user explicitly asks to restart, restart from Stage 1; otherwise resume rather than rerun. This makes the skill safely re-invokable: a second call in the same conversation produces only the next incremental section, never duplicates or regressions to earlier stages. Scope: idempotency holds within a single conversation; if prior output exists only in an external file or a previous session, paste it into the conversation first or restart from Stage 1.
Stage Contracts. Each stage below opens with a Contract block declaring inputs, outputs, preconditions, postconditions, and failure modes. The contract is the structural boundary — it is what makes the stage retry-safe, error- explicit, and self-describing. Treat the contract as binding: if the preconditions are not met, do not enter the stage; if a failure-mode trigger fires, take the declared action rather than improvising.
该技能分为七个阶段（1、2、3、3a、4、5、5b）。请勿跳过任何阶段，后续阶段依赖用户在前期阶段的回答和选择。Stage 3a（MMF门槛）可能拒绝完整活动计划，转而引导用户进行验证工作。Stage 5以选择活动形态结束——只有用户选定三种形态之一后，Stage 5b才会启动。
该技能设计为可在多Agent环境中使用。当流程要求「询问用户」时，使用所在环境的结构化提问工具（在Claude Code中为
AskUserQuestion
工具，每次最多批量提问4个）；如果没有结构化工具，则用普通文本提问，等待用户回复后再继续。无论哪种方式，切勿编造答案。
恢复协议（幂等性）。生成任何输出前，扫描对话历史查看是否有该技能的过往输出。如果对话中已存在某章节的标题（
## N. <名称>
），请勿重新生成。从满足前置条件且输出章节尚未在对话中出现的第一个阶段开始。如果用户明确要求重启，则从Stage 1开始；否则继续而非重复执行。这使该技能可安全重复调用：同一对话中的第二次调用仅生成下一个增量章节，不会重复或倒退到之前的阶段。范围：幂等性仅在单对话内有效；如果过往输出仅存在于外部文件或之前的会话中，请先将其粘贴到对话中，或从Stage 1重启。
阶段契约。每个阶段开头都有一个契约块，声明输入、输出、前置条件、后置条件和失败模式。契约是结构化边界——它使阶段具备重试安全性、错误明确性和自描述性。请严格遵守契约：如果前置条件未满足，请勿进入该阶段；如果触发失败模式，请执行指定操作而非自行发挥。

Stage 1 — Interview / Brief Capture

Stage 1 — 访谈/需求捕获

Contract
  • Inputs: user's initial brief (free text, any length).
  • Outputs: Section 0 (Assumptions table) under Mode C; mode-tag (A/B/C) and the 8 fields captured in working memory for downstream stages.
  • Preconditions: none (entry stage).
  • Postconditions: all 8 fields have a value (captured, defaulted, or flagged as 
    [ask]
    for a still-pending question); mode is determined.
  • Failure modes:
    • any always-capture field missing under Mode C → add it to the question batch; do not default silently.
    • a missing field would change Stage 3 asymmetry classification OR Stage 4 function choice → escalate it into the batch even if it would normally default.
    • user gives contradictory values for the same field → ask one clarifying question; do not pick one silently.
The skill has three intake modes. Pick the mode that matches the user's brief before asking any questions.
Mode A — Full-brief mode. The user has volunteered all 8 fields below up front (in a pasted brief, a prior plan, a detailed message). Skip the interview entirely, confirm understanding in one sentence, and proceed to Stage 2.
Mode B — Interview mode. The user explicitly asks to be scoped ("help me figure out what this campaign should be," "walk me through it," "interview me"). Ask the 8 questions below in batches of ≤4 per turn, using the environment's structured question facility if one exists (see the structured-tool note in the Workflow section above). Do not invent answers. Do not merge questions.
Mode C — Default mode (for terse briefs). The user gave a short brief (≤2 sentences) without asking for an interview. This is the common case. Do NOT run the full 8-question interview — it turns a 12-word prompt into an interrogation. Instead:
  1. Always-capture fields (ask once, batched). Four fields are too material to default silently; getting them wrong produces a generic plan. Ask the user in one batch of up to 4 questions covering:
    • Outcome — what specific action is the campaign driving (signups, ticket sales, votes, donations, attendance, pre-orders, qualified demos)?
    • Audience — who is the target? Who experiences the problem? Geography, seniority, community membership. "Everyone interested in AI" is not an audience; "London-based senior AI/ML engineers at Series A–C startups" is.
    • Sector — pick from the 6 riders in 
      references/sector-riders.md
      (
      cohort-education
      , 
      b2b-saas
      , 
      ngo
      , 
      consumer-brand
      , 
      political-civic
      , 
      personal-brand
      ), or 
      other
      with a one-sentence description. The sector rider materially changes Stage 4 channel weighting and Stage 5 archetype defaults.
    • Budget — what can you actually spend per month (€0 / <€1k / €1k–10k / €10k+) and what is the competitor's rough spend (unknown / similar / 5× ours / 50×+ ours)? Budget materially changes Stage 3 asymmetry classification and Stage 4 paid-channel availability. Never default this silently.
    If one of these four fields is already clearly present in the user's brief, drop it from the question batch. Only ask what is missing.
  2. Default-with-flag fields. The remaining fields default silently but are surfaced in an Assumptions table at the top of the final deliverable (Stage 5 output). The user can confirm or adjust inline after reading the plan:
    FieldDefault under terse brief
    Competition (specifics)Abstract — "cohort-based courses in the category," "mid-size SaaS competitors," etc. Use the heuristic questions in 
    references/asymmetry-audit-table.md
    to classify asymmetry qualitatively. Never invent specific competitor names.
    Existing channels / traction"Starting from zero" unless the user's brief, working directory, or CLAUDE.md clearly indicates an existing newsletter, community, or follower base.
    BottleneckTrust (the 2025–2026 default for almost every underdog campaign).
    Time window60 days to a single dated anchor moment (event, launch, election), then recurring cadence.
    Capacity6h/week.
  3. Escalation rule. If a missing field would materially change the Stage 3 asymmetry classification or the Stage 4 primary-function choice, do not default it — add it to the question batch. Example: if the user mentions "our whole team is posting already" but does not specify which channels, ask, because it changes the channel stack.
  4. Assumptions table is mandatory under Mode C. Open the final deliverable with a table listing every defaulted field and its assumed value. End the deliverable with: "Confirm or adjust any row in the Assumptions table and I will re-run the affected stages."
契约
  • 输入：用户的初始需求（自由文本，长度不限）。
  • 输出：Mode C下的Section 0（假设表）；模式标签（A/B/C），以及为后续阶段存储在工作内存中的8个字段。
  • 前置条件：无（入口阶段）。
  • 后置条件：所有8个字段都有值（已捕获、默认值或标记为
    [ask]
    表示待提问）；已确定模式。
  • 失败模式：
    • Mode C下任何必填字段缺失 → 将其加入提问批次；切勿默认填充。
    • 缺失字段会改变Stage 3的不对称分类或Stage 4的功能选择 → 将其加入提问批次，即使通常可默认填充。
    • 用户对同一字段给出矛盾值 → 提出一个澄清问题；切勿自行选择。
该技能有三种接入模式。在提问前，选择与用户需求匹配的模式。
Mode A — 完整需求模式。用户已主动提供以下全部8个字段（粘贴的需求、过往计划、详细消息）。完全跳过访谈，用一句话确认理解，然后进入Stage 2。
Mode B — 访谈模式。用户明确要求梳理需求（「帮我理清这个活动应该怎么做」「带我一步步来」「采访我」）。将以下8个问题分批提问，每批≤4个，使用环境的结构化提问工具（见工作流程部分的结构化工具说明）。请勿编造答案，请勿合并问题。
Mode C — 默认模式（适用于简短需求）。用户给出简短需求（≤2句话）且未要求访谈。这是常见情况。切勿进行完整的8个问题访谈——这会将12字的提示变成审问。取而代之：
  1. 必填字段（一次性批量提问）。有四个字段至关重要，不能默认填充；填错会导致方案泛化。在一个批次中向用户提问最多4个问题，涵盖：
    • 目标成果 — 活动要推动的具体行动是什么（注册、售票、投票、捐款、出席、预购、合格演示）？
    • 受众 — 目标群体是谁？谁面临相关问题？包括地域、层级、社区成员身份。「所有对AI感兴趣的人」不是明确受众；「伦敦地区A-C轮初创公司的资深AI/ML工程师」才是。
    • 行业领域 — 从
      references/sector-riders.md
      中的6个选项（
      cohort-education
      、
      b2b-saas
      、
      ngo
      、
      consumer-brand
      、
      political-civic
      、
      personal-brand
      ）中选择，或选择
      other
      并附上一句话描述。行业领域会显著影响Stage 4的渠道权重和Stage 5的活动原型默认值。
    • 预算 — 您每月实际可投入多少（€0 / <€1k / €1k–10k / €10k+），竞争对手的大致投入是多少（未知 / 相似 / 是我们的5倍 / 是我们的50倍以上）？预算会显著影响Stage 3的不对称分类和Stage 4的付费渠道可用性。切勿默认填充。
    如果用户的需求中已明确包含这四个字段之一，则从提问批次中移除该字段，仅询问缺失的部分。
  2. 带标记默认字段。其余字段默认填充，但会在最终交付物（Stage 5输出）顶部的假设表中列出。用户可在阅读方案后确认或调整：
    字段简短需求下的默认值
    竞争对手（详情）抽象描述——「同类 cohort-based 课程」「中型SaaS竞争对手」等。使用
    references/asymmetry-audit-table.md
    中的启发式问题定性分类预算不对称。切勿编造具体竞争对手名称。
    现有渠道/已有成果「从零开始」，除非用户需求、工作目录或CLAUDE.md明确表明已有通讯、社区或粉丝基础。
    瓶颈信任（2025-2026年几乎所有弱势活动的默认瓶颈）。
    时间窗口60天到一个特定锚点时刻（活动、发布、选举），之后为周期性节奏。
    可用时间每周6小时。
  3. 升级规则。如果缺失字段会显著改变Stage 3的不对称分类或Stage 4的主要功能选择，则不默认填充——将其加入提问批次。例如：如果用户提到「我们整个团队都在发帖」但未指定渠道，则需询问，因为这会改变渠道体系。
  4. Mode C下必须包含假设表。最终交付物开头需列出一个表格，包含所有默认字段及其假设值。交付物结尾需附上：「确认或调整假设表中的任何一行，我将重新运行受影响的阶段。」

The 8 Fields

8个字段

Regardless of mode, the campaign plan is grounded in these 8 fields:
  1. Sector and outcome — which of the 6 sector riders applies, and the specific action the campaign drives. (Sector is ask-once in Mode C; outcome is ask-once in Mode C.)
  2. Audience — who is the target, with enough specificity to picture them. (Ask-once in Mode C.)
  3. Competition / incumbent — name 3–5 specific competitors, or say "unknown" / "I'll describe them abstractly." If unknown, stay at category level and use 
    references/asymmetry-audit-table.md
    heuristics. Never invent names. (Defaults in Mode C; user can sharpen in the Assumptions confirmation.)
  4. Budget — user's and competitor's, in numbers or rough ratios. (Ask-once in Mode C.)
  5. Existing channels and traction — where does the user already reach the audience (newsletter size, community size, follower counts that matter, warm email list, prior press). (Defaults in Mode C.)
  6. Bottleneck — reach / trust / conversion / turnout / retention — pick one. (Defaults to trust in Mode C.)
  7. Time window — evergreen / 30 / 60 / 90 days / tied to a dated event or launch. (Defaults to 60 days in Mode C.)
  8. Capacity — realistic hours per week. Hard-constrains Stage 5's first-30-days action list; lowest-ROI actions get cut until total weekly effort fits. (Defaults to 6h/week in Mode C.)
无论采用哪种模式，活动计划都基于以下8个字段：
  1. 行业领域和目标成果 — 适用6个行业选项中的哪一个，以及活动推动的具体行动。（Mode C下行业领域需提问一次；目标成果需提问一次。）
  2. 受众 — 目标群体，需具体到可清晰描绘的程度。（Mode C下需提问一次。）
  3. 竞争对手/行业巨头 — 列出3-5个具体竞争对手，或说明「未知」/「我将抽象描述」。如果未知，保持类别层面，使用
    references/asymmetry-audit-table.md
    中的启发式方法。切勿编造名称。（Mode C下默认填充；用户可在假设表确认环节细化。）
  4. 预算 — 用户和竞争对手的预算，用数字或大致比例表示。（Mode C下需提问一次。）
  5. 现有渠道和已有成果 — 用户已通过哪些渠道触达受众（通讯订阅量、社区规模、有效粉丝数、精准邮件列表、过往媒体报道）。（Mode C下默认填充。）
  6. 瓶颈 — 触达 / 信任 / 转化 / 参与度 / 留存 — 选择其一。（Mode C下默认信任。）
  7. 时间窗口 — 长期 / 30 / 60 / 90天 / 绑定到特定事件或发布日期。（Mode C下默认60天。）
  8. 可用时间 — 每周实际可投入的时间。严格限制Stage 5前30天行动列表；优先剔除ROI最低的行动，直至总周工作量符合要求。（Mode C下默认每周6小时。）

Stage 2 — Ideation (generate 5+ campaign concepts across archetypes)

Stage 2 — 创意构思（生成5+跨原型活动概念）

Contract
  • Inputs: 8 fields from Stage 1 (especially sector, audience, competition descriptor).
  • Outputs: Section 1 (Campaign Ideas) with ≥5 concepts across distinct archetypes; an explicit ask to the user to pick one or more.
  • Preconditions: Stage 1 complete; sector field has a value (named or 
    other
    ).
  • Postconditions: ≥5 distinct-archetype concepts presented; user has been prompted to select.
  • Failure modes:
    • competitor unnamed in brief → use 
      [incumbent]
      placeholder or category descriptor in every concept; never invent a name.
    • cannot produce 5 genuinely distinct archetypes for this audience/sector → say so plainly, present what you have, do not pad with cosmetic variants of one archetype.
    • user's brief names one dominant player but expects industry peers → do not auto-name the obvious #2/#3; use the escape-hatch phrasings from the Industry-peer rule.
Before any audit or channel work, generate at least five distinct campaign concepts, drawn from different archetypes in 
references/campaign-archetypes.md
. Do not converge early. The point is to give the user a shaped menu to choose from.
For each concept, produce:
  • Name — short, memorable, in the user's voice.
  • One-line thesis — why this campaign works for this user against this competitor right now.
  • Archetype — which archetype it draws from (e.g., founder-story arc, counter-narrative, earned-media stunt, referral flywheel, community-build, coalition play).
  • Primary channel tier — which tier from 
    references/channel-tier-stack.md
    this concept leans on (Tier 1/2/3/4).
  • Authenticity hook — the specific real-world detail, person, story, or action that makes it credible and hard to fake.
  • Minimum viable version — the smallest possible execution that proves the concept works in two weeks or less.
Anti-fabrication rule for concept names, theses, and hooks. If the user named a specific competitor in the brief, use the name. If the user said "well-known incumbent," "market leader," "the dominant player," or otherwise did not name one, do not fill in the blank with your best guess. Stay at the level of abstraction the brief gave you. Use the category descriptor (e.g., "the $400-per-seat monitor"), the behavior ("the enterprise billing tax"), or a bracketed placeholder (
[incumbent]
) that the user will fill in. Concept names like "We Quit Sentry" or "DataDog Alternative" — invented from the category description rather than the brief — violate the Stage 1 anti-fabrication rule and must not appear in Stage 2 output. This applies equally to SEO keyword lists, earned-media pitch hooks, and any downstream section that inherits the concept name.
Industry-peer rule. The anti-fabrication rule covers any competitor name absent from the brief — not only invented names, but widely-known industry peers to a brief-named incumbent. If the brief names one dominant player (e.g., the user says "we're outspent 100x by [market leader]"), do not add the obvious #2 or #3 from the same category on your own ("LexisNexis and Westlaw," "Salesforce and HubSpot," "Datadog and New Relic") as if their presence were implied context. Use escape-hatch phrasing instead: "the other major [category] platforms," "[incumbent]-class tools," or "the dominant [category] incumbents." This applies in every section — Stage 2 concepts, Stage 4 SEO, Stage 5 competitor saturation, Stage 5b earned-media targets and ad-copy hooks, and any other downstream output. The discipline does not relax in any later stage. "Everyone in the industry knows it exists" is not a licence to name it.
After presenting the five concepts, ask the user to pick one (or more) to push through Stage 3–5.
契约
  • 输入：Stage 1的8个字段（尤其是行业领域、受众、竞争对手描述）。
  • 输出：Section 1（活动创意），包含≥5个不同原型的概念；明确要求用户选择一个或多个。
  • 前置条件：Stage 1完成；行业领域字段有值（指定名称或
    other
    ）。
  • 后置条件：呈现≥5个不同原型的概念；已提示用户选择。
  • 失败模式：
    • 需求中未提及竞争对手 → 在每个概念中使用
      [incumbent]
      占位符或类别描述；切勿编造名称。
    • 无法为该受众/行业生成5个真正不同的原型 → 如实说明，呈现现有内容，切勿用同一原型的变体填充。
    • 用户需求中提到一个主导玩家，但期望覆盖行业同行 → 切勿自动添加明显的第二名/第三名；使用行业同行规则中的替代表述。
在进行任何审核或渠道工作前，生成至少5个不同的活动概念，灵感来自
references/campaign-archetypes.md
中的不同原型。切勿过早收敛。目的是为用户提供一个结构化的选择菜单。
每个概念需包含：
  • 名称 — 简短易记，符合用户的语气。
  • 一句话核心论点 — 为什么这个活动对该用户、对抗该竞争对手、在当下有效。
  • 原型 — 源自哪个原型（例如：创始人故事弧、反叙事、 earned-media stunt、推荐飞轮、社区建设、联盟合作）。
  • 主要渠道层级 — 该概念依赖
    references/channel-tier-stack.md
    中的哪个层级（Tier 1/2/3/4）。
  • 真实性钩子 — 使其可信且难以模仿的具体现实细节、人物、故事或行动。
  • 最小可行版本 — 可在两周或更短时间内验证概念有效性的最小执行方案。
概念名称、论点和钩子的反编造规则。如果用户需求中提到了具体竞争对手名称，则使用该名称。如果用户说「知名行业巨头」「市场领导者」「主导玩家」或未提及名称，切勿自行猜测填充。保持需求中的抽象程度。使用类别描述（例如：「400欧元/席位的监控工具」）、行为描述（「企业账单税」）或括号占位符
[incumbent]
，由用户自行填充。像「We Quit Sentry」或「DataDog Alternative」这类从类别描述而非需求中编造的概念名称，违反了Stage 1的反编造规则，不得出现在Stage 2输出中。这同样适用于SEO关键词列表、 earned-media pitch钩子以及继承概念名称的任何下游章节。
行业同行规则。反编造规则涵盖需求中未提及的任何竞争对手名称——不仅是编造的名称，还包括需求中提到的行业巨头的知名同行。如果需求中提到一个主导玩家（例如：用户说「我们的预算是[市场领导者]的1/100」），切勿自行添加同一类别中明显的第二名或第三名（「LexisNexis和Westlaw」「Salesforce和HubSpot」「Datadog和New Relic」），仿佛它们是隐含的背景信息。取而代之，使用替代表述：「其他主要[类别]平台」「[incumbent]级工具」或「主导[类别]的行业巨头」。这适用于所有章节——Stage 2概念、Stage 4 SEO、Stage 5竞争对手饱和分析、Stage 5b earned-media目标和广告文案钩子，以及任何其他下游输出。该规则在后续阶段不会放宽。「行业内人人都知道它存在」不能成为编造名称的理由。
呈现五个概念后，要求用户选择一个（或多个）进入Stage 3-5。

Stage 3 — Asymmetry Audit

Stage 3 — 不对称性审核

Contract
  • Inputs: budget (user's + competitor's), or heuristic answers from 
    references/asymmetry-audit-table.md
    ; the concept(s) the user selected from Stage 2.
  • Outputs: Section 3 (Spend Asymmetry Verdict) — one-sentence classification + preconditions score (0–6).
  • Preconditions: Stage 2 complete; user has named one or more concepts to push forward.
  • Postconditions: asymmetry classified as mild / severe / categorical; preconditions count is in {0..6}; downstream stages know which playbook scale to run.
  • Failure modes:
    • 0–2 preconditions → name the missing ones, refuse the full playbook, route the user to building the missing preconditions as the campaign itself.
    • asymmetry numbers absent → ask the heuristic questions in the reference file; do not assume a level.
    • user gives competitor spend in a non-comparable form (e.g., revenue, headcount) → ask one clarifying question to convert to spend ratio; do not estimate silently.
Classify the user's spend asymmetry using the table in 
references/asymmetry-audit-table.md
.
  • If the user gave numbers: compute the ratio (competitor spend ÷ user spend) and map it to mild (1:2–1:5), severe (1:5–1:50), or categorical (1:50+).
  • If the user did not give numbers: ask the heuristic questions from the reference file (state-backing? can they afford billboards? do they dominate your category's search ads?) and classify qualitatively.
Report back one sentence: 
Your asymmetry is <level>. This means <what it means for your strategy>.
Do not hedge. Do not offer a "balanced" recommendation if the user is at categorical — it would be misleading.
Then run a Preconditions Check. The insurgent playbook wins when preconditions are present; asymmetry alone is not enough. Score the user's situation against the six factors that made the Hungarian case work (see 
references/hungarian-case-study.md
for the full mechanism). Ask or infer:
  1. Credible insider / founder-story / defector equivalent. Is there a real person whose authenticity the competitor cannot manufacture?
  2. Accumulated grievance or unmet demand. Is there anger the incumbent has ignored long enough that it only needs a vehicle?
  3. Consolidated challenger field. Is the user the one clear alternative, or one of many? Fragmented fields dilute organic narrative.
  4. Felt pain, not abstract pain. Does every target experience the cost directly (price, time, trust, service failure), or is the grievance distributed?
  5. Threshold-rewarding market / platform / system. Does the distribution system reward consolidation once a share crosses some threshold (network effects, category leadership, algorithm-bound attention, electoral math)?
  6. Incumbent overplaying a fear / saturation hand. Is the competitor running past the curve — more ads, more fear, more polished content — in a way the user can counter-position against?
Score 0–6. Tell the user the count plainly and what it implies:
  • 5–6 preconditions: run the insurgent playbook at full scale. Proceed to Stage 4.
  • 3–4 preconditions: proceed, but flag the missing ones as campaign sub-goals — the user will need to build them during the campaign (e.g., find a credible voice, surface the grievance) for organic to compound.
  • 0–2 preconditions: refuse to execute the playbook at full scale. Building the preconditions is the campaign. Recruit the credible voice, consolidate the coalition, surface and name the grievance, find the felt-pain story. Until those are present, the insurgent playbook will underperform and burn the volunteer/community energy it depends on. Say this out loud. Do not soften it.
契约
  • 输入：预算（用户+竞争对手），或
    references/asymmetry-audit-table.md
    中的启发式答案；用户从Stage 2选择的概念。
  • 输出：Section 3（预算不对称 verdict）——一句话分类+前置条件得分（0–6）。
  • 前置条件：Stage 2完成；用户已指定一个或多个推进的概念。
  • 后置条件：不对称性分类为轻度/重度/绝对；前置条件得分在{0..6}范围内；后续阶段知道要运行哪个规模的策略手册。
  • 失败模式：
    • 前置条件得分0–2 → 列出缺失的条件，拒绝完整策略手册，引导用户将缺失条件的构建作为活动本身。
    • 不对称性数据缺失 → 询问参考文件中的启发式问题；切勿假设级别。
    • 用户给出的竞争对手投入为不可比形式（例如：收入、员工数） → 提出一个澄清问题，转换为投入比例；切勿自行估算。
使用
references/asymmetry-audit-table.md
中的表格对用户的预算不对称性进行分类。
  • 如果用户提供了数字：计算比例（竞争对手投入÷用户投入），映射为轻度（1:2–1:5）、重度（1:5–1:50）或绝对（1:50+）。
  • 如果用户未提供数字：询问参考文件中的启发式问题（是否有官方支持？是否能负担广告牌？是否主导类别搜索广告？），进行定性分类。
用一句话反馈：
您的预算不对称性为<级别>。这意味着<对您策略的影响>。
切勿含糊其辞。如果用户处于绝对不对称级别，切勿提供「平衡」建议——这会产生误导。
然后进行前置条件检查。反叛式策略手册仅在满足前置条件时有效；仅靠不对称性不足以取胜。根据匈牙利案例成功的六个因素（见
references/hungarian-case-study.md
的完整机制），对用户的情况进行评分。询问或推断：
  1. 可信内部人士/创始人故事/叛逃者等价物。是否有一个真实人物，其真实性是竞争对手无法复制的？
  2. 累积不满或未满足需求。行业巨头是否长期忽视受众的愤怒，只需一个触发点即可爆发？
  3. 挑战者阵营整合。用户是明确的替代选择，还是众多选项之一？分散的阵营会削弱有机叙事效果。
  4. 切身痛点，而非抽象痛点。每个目标受众是否直接感受到成本（价格、时间、信任、服务失败），还是不满情绪分散？
  5. 达到阈值即获回报的市场/平台/系统。当份额超过某个阈值时，分发系统是否会奖励整合（网络效应、类别领导力、算法绑定的注意力、选举数学）？
  6. 行业巨头过度使用恐惧/饱和策略。竞争对手是否过度投放广告、渲染恐惧、制作过度 polished 的内容，以至于用户可以反其道而行之？
得分0–6。明确告知用户得分及其含义：
  • 5–6个前置条件：全面执行反叛式策略手册。进入Stage 4。
  • 3–4个前置条件：继续执行，但将缺失条件标记为活动子目标——用户需要在活动期间构建这些条件（例如：找到可信声音、挖掘不满情绪），以便有机内容产生复利效应。
  • 0–2个前置条件：拒绝全面执行策略手册。构建前置条件就是活动本身。招募可信声音、整合联盟、挖掘并命名不满情绪、找到切身痛点故事。在这些条件满足前，反叛式策略手册效果不佳，还会消耗其依赖的志愿者/社区精力。明确告知用户，切勿软化措辞。

Stage 3a — Message-Market-Fit Gate

Stage 3a — 信息-市场匹配（MMF）门槛

Contract
  • Inputs: three yes/no signals (revenue/commitment, audience-language, close) from the user.
  • Outputs: Section 3a (3 signals + 0–3 score + explicit verdict).
  • Preconditions: Stage 3 complete.
  • Postconditions: verdict ∈ {3/3 proceed, 2/3 validation-cycle, 0–1/3 refuse}; downstream stages know whether to proceed, insert a validation cycle, or stop.
  • Failure modes:
    • score 0–1/3 → produce only Section 3a + refusal text; do NOT generate Sections 4–11. Route the user to a discovery cycle.
    • score 2/3 → produce Section 3a + the validation-cycle pre-task description; allow Stage 4 only after the validation cycle is named.
    • user does not know the answer to a signal → treat that signal as a "no" (absence of evidence is evidence of absence for MMF); do not skip the question.
Before assembling a channel stack, verify the campaign is solving a distribution problem, not an MMF problem. Distribution amplifies signal; it cannot manufacture it. A founder's great LinkedIn post cannot sell a product nobody wants, and a movement's best volunteer network cannot turn out voters for a message that does not name their pain.
Ask the user three yes/no questions. Do not skip any. If the user does not know the answer to one, treat that as a "no" — absence of evidence is evidence of absence for MMF.
  1. Revenue/commitment signal: Have ≥10 people in the target audience paid, signed up, pre-registered, or directly asked (without your prompting) for what you are offering? For events: ≥10 past attendees or paid waitlist. For products: ≥10 customers or pre-orders. For movements: ≥10 signed volunteers/members.
  2. Language signal: Can you name a specific pain the audience articulates in their own words before you pitch them? Not a pain you infer — a sentence someone in the audience has actually said, written, or posted about.
  3. Close signal: If you asked 5 people in the target audience to commit today (at the intended price / format / ask level), would at least 3 say yes?
Scoring:
  • 3/3 — MMF confirmed. The campaign is a distribution problem. Proceed to Stage 4.
  • 2/3 — Borderline. Name the weak link and insert a 1–2 week validation cycle into the plan as a pre-campaign task before running the full Stage 4–5 playbook. Examples: if question 2 fails, do 5 customer discovery conversations and extract the language; if question 1 fails, run a paid waitlist or pre-order test. Revisit the MMF gate after the validation cycle.
  • 0–1/3 — Refuse the full campaign plan. MMF is the bottleneck, not distribution. Route the user to a discovery cycle: 5 structured customer conversations, a pre-order or paid-waitlist test, a small-room live demo. Explain plainly: "The insurgent playbook scales trust. It cannot manufacture demand for a product, event, or cause people do not already want. Running a full campaign now will burn the volunteer/community energy that Stage 5 depends on. Come back when at least 2/3 of these signals are present." Do not soften this. Do not produce the full Stage 4–5 deliverable.
Report the MMF verdict at the top of the Stage 3 section of the final output so the user sees it before the channel stack.
契约
  • 输入：用户给出的三个是/否信号（收入/承诺、受众语言、转化信号）。
  • 输出：Section 3a（3个信号+0–3得分+明确 verdict）。
  • 前置条件：Stage 3完成。
  • 后置条件： verdict ∈ {3/3 继续, 2/3 验证周期, 0–1/3 拒绝}；后续阶段知道是继续、插入验证周期还是停止。
  • 失败模式：
    • 得分0–1/3 → 仅生成Section 3a + 拒绝文本；切勿生成Sections 4–11。引导用户进入发现周期。
    • 得分2/3 → 生成Section 3a + 验证周期前置任务描述；仅在验证周期完成后允许进入Stage 4。
    • 用户不知道某个信号的答案 → 将该信号视为「否」（MMF中没有证据就是没有证据）；切勿跳过问题。
在搭建渠道体系前，验证活动解决的是分发问题，而非MMF问题。分发放大信号，但无法制造信号。创始人的优秀LinkedIn帖子无法销售没人想要的产品，运动的最佳志愿者网络无法为未提及受众痛点的信息动员选民。
向用户提出三个是/否问题，切勿跳过任何一个。如果用户不知道某个问题的答案，视为「否」——MMF中没有证据就是没有证据。
  1. 收入/承诺信号：是否有≥10个目标受众付费、注册、预注册，或主动（无需您提示）询问您提供的产品/服务？对于活动：≥10个过往参与者或付费候补名单。对于产品：≥10个客户或预订单。对于运动：≥10个已签署的志愿者/成员。
  2. 语言信号：您能否说出目标受众用自己的语言表达的具体痛点？不是您推断的痛点，而是受众实际说过、写过或发布过的句子。
  3. 转化信号：如果您邀请5个目标受众立即承诺（按预期价格/形式/要求），是否至少有3个会同意？
评分：
  • 3/3 — MMF已确认。活动是分发问题。进入Stage 4。
  • 2/3 — 临界状态。指出薄弱环节，并在计划中插入1–2周的验证周期作为活动前置任务，然后再运行完整的Stage 4–5策略手册。例如：如果问题2未通过，进行5次客户发现对话，提取受众语言；如果问题1未通过，运行付费候补名单或预订单测试。验证周期结束后重新检查MMF门槛。
  • 0–1/3 — 拒绝完整活动计划。瓶颈是MMF，而非分发。引导用户进入发现周期：5次结构化客户对话、预订单或付费候补名单测试、小型现场演示。明确解释：「反叛式策略手册建立信任。它无法为人们不想要的产品、活动或事业制造需求。现在运行完整活动会消耗Stage 5依赖的志愿者/社区精力。当至少2/3的信号满足时再回来。」 切勿软化措辞，切勿生成完整的Stage 4–5交付物。
在最终输出的Stage 3部分顶部报告MMF verdict，让用户在看到渠道体系前先看到它。

Stage 4 — Channel Tier Stack + 70/30 Allocation

Stage 4 — 渠道层级体系 + 70/30分配

Contract
  • Inputs: bottleneck (Stage 1 Q6), sector (Stage 1 Q1), asymmetry verdict (Stage 3), MMF verdict (Stage 3a).
  • Outputs: Section 4 (Channel Tier Stack) + Section 5 (70/30 or 80/20 Allocation).
  • Preconditions: Stage 3a verdict is 
    3/3 proceed
    OR 
    2/3 with validation cycle attached
    . If verdict is 
    0–1/3 refuse
    , this stage MUST NOT execute.
  • Postconditions: primary function chosen (demand-capture / paid-amplification / trust-compounding); sector rider applied; channels listed with weekly effort estimates; allocation split stated in plain numbers.
  • Failure modes:
    • asymmetry is categorical AND user requests broad cold-paid Tier 4 → refuse, route to Tier 1/2 + counter-positioning, explain the diminishing-returns curve from 
      references/authenticity-playbook.md
      .
    • sector = 
      other
      → flag rider mismatch in Assumptions table, proceed with the closest rider, name which rider assumptions do not apply.
    • function choice conflicts with sector rider (e.g., B2B SaaS rider biases founder-LinkedIn, function choice is demand-capture) → surface the conflict, present both paths, do not collapse silently.
Function before cost. Before picking channels, pick the campaign's primary function against the user's bottleneck (from Stage 1 Q6). Use the function table at the top of 
references/channel-tier-stack.md
:
  • Demand capture — bottleneck is "people already want this, they can't find us." Primary function for small or unknown challengers with addressable existing intent. Leads with non-brand Google Search, SEO, directory presence. Never brand-keyword bidding without a lift test (see Stage 5).
  • Paid amplification — bottleneck is "our organic content works but reaches too few people." Primary function only after organic winners exist (24–48h traction gate). Leads with retargeting, warm lookalikes, Spark Ads on proven TikTok content.
  • Trust compounding — bottleneck is "people find us, sometimes click, but don't convert, refer, or come back." Primary function for the majority of underdog campaigns. Leads with Tier 1: founder content, community nodes, newsletters, volunteer networks.
A mild-asymmetry user with a trust bottleneck should not be routed to paid amplification just because they can afford it. Route by function first; allocate within that function by cost/asymmetry second.
Apply the sector rider. After the function choice but before allocation, open 
references/sector-riders.md
and apply the rider matching the user's sector (captured in Stage 1). The rider adjusts:
  • Archetype defaults — which campaign archetypes fit this sector best (e.g., cohort-education biases toward community-build + referral flywheel; B2B SaaS biases toward founder-LinkedIn + demand capture).
  • Channel weighting — which Tier 1/2 channels compound fastest in this sector (e.g., consumer brands lean on UGC flywheels; political campaigns lean on ground game and counter-media; NGOs lean on volunteer networks + earned media on named-beneficiary stories).
  • Failure-mode warning — the sector-specific way the insurgent playbook underperforms when misapplied, which becomes a standing warning in Stage 5's anti-vanity dashboard.
Do not overwrite the function-first choice. The rider layers on top. If the rider and the function choice conflict (e.g., B2B SaaS rider biases toward founder-LinkedIn but the function choice is demand capture), surface the conflict and name both paths rather than collapsing to one.
If the user picked "other" for sector, flag the mismatch in the Assumptions table and proceed with the closest rider — naming which of the rider's structural assumptions do not apply.
Then assemble the channel stack using 
references/channel-tier-stack.md
. Allocation rules layer on top of the function choice and the sector rider:
  • Mild asymmetry → stack can include Tier 1, 2, 3, and selective Tier 4. Allocate ~70% effort to organic (Tier 1 + 2), 30% to paid amplification (Tier 3, with rare Tier 4).
  • Severe asymmetry → Tier 1, 2, and targeted Tier 3. Avoid broad Tier 4. 80/20 split toward organic is often safer.
  • Categorical asymmetry → Tier 1 and 2 only. Refuse to draft broad cold-paid Tier 4 creative — it will not work and it will burn runway. If the user insists, explain the diminishing-returns curve and the counter-positioning move (see 
    references/authenticity-playbook.md
    ) before reconsidering.
Output the stack as a prioritized list with: (a) the primary function chosen, (b) the channels mapped to that function, (c) an estimated weekly effort commitment per channel, (d) the 70/30 (or 80/20) split in plain numbers.
Anti-fabrication carries through. The Stage 2 anti-fabrication rule applies to every element of the Stage 4 output: SEO keyword examples, long-tail query lists, directory/review-site references, competitor saturation descriptions, and any sample copy shown inline with the stack. If the brief did not name the incumbent, the incumbent's proper noun must not appear in this stage in any casing — not title-case ("Sentry alternative"), not lowercase ("sentry alternative"), not hyphenated, not as part of a compound keyword or URL. Search-query examples that would otherwise require a brand name must either retain the 
[incumbent]
bracketed placeholder for the user to fill in, or be rewritten as non-brand equivalents ("error monitoring for small teams," "application monitoring under $100/seat", "lightweight APM for node.js"). This applies equally to Stage 5's competitor saturation map and to every Stage 5b output section (
## 8
–
## 11
): the banned-token discipline does not relax downstream of Stage 2.
契约
  • 输入：瓶颈（Stage 1 Q6）、行业领域（Stage 1 Q1）、不对称性 verdict（Stage 3）、MMF verdict（Stage 3a）。
  • 输出：Section 4（渠道层级体系） + Section 5（70/30或80/20分配）。
  • 前置条件：Stage 3a verdict为
    3/3 继续
    或
    2/3 附加验证周期
    。如果verdict为
    0–1/3 拒绝
    ，本阶段不得执行。
  • 后置条件：已选择主要功能（需求捕获/付费放大/信任复利）；已应用行业领域规则；已列出渠道及每周工作量估算；已明确分配比例。
  • 失败模式：
    • 绝对不对称性且用户要求广泛的冷付费Tier 4 → 拒绝，引导至Tier 1/2 + 反定位策略，解释
      references/authenticity-playbook.md
      中的边际收益递减曲线。
    • 行业领域 = 
      other
      → 在假设表中标记规则不匹配，使用最接近的规则继续，说明哪些规则假设不适用。
    • 功能选择与行业领域规则冲突（例如：B2B SaaS规则偏向创始人LinkedIn，功能选择是需求捕获） → 指出冲突，呈现两种路径，切勿自行合并。
功能优先于成本。在选择渠道前，根据用户的瓶颈（Stage 1 Q6）选择活动的主要功能。使用
references/channel-tier-stack.md
顶部的功能表：
  • 需求捕获 — 瓶颈是「人们已经想要这个，但找不到我们」。适用于有可触达现有意向的小型或不知名挑战者。以非品牌Google搜索、SEO、目录存在为核心。未经效果测试，切勿投放品牌关键词竞价（见Stage 5）。
  • 付费放大 — 瓶颈是「我们的有机内容有效，但触达人数太少」。仅在有机内容已验证有效（24–48h traction门槛）后使用。以重定向、相似受众、已验证TikTok内容的Spark Ads为核心。
  • 信任复利 — 瓶颈是「人们找到我们，有时点击，但不转化、推荐或回头」。适用于大多数弱势活动。以Tier 1为核心：创始人内容、社区节点、通讯、志愿者网络。
即使轻度不对称且预算充足的用户面临信任瓶颈，也不应引导至付费放大。先按功能选择；再根据成本/不对称性在该功能内分配资源。
应用行业领域规则。选择功能后、分配前，打开
references/sector-riders.md
，应用与用户行业领域（Stage 1捕获）匹配的规则。规则会调整：
  • 原型默认值 — 哪些活动原型最适合该行业（例如：cohort-education偏向社区建设+推荐飞轮；B2B SaaS偏向创始人LinkedIn+需求捕获）。
  • 渠道权重 — 哪些Tier 1/2渠道在该行业复利最快（例如：消费品牌依赖UGC飞轮；政治活动依赖线下行动和反媒体；NGO依赖志愿者网络+基于受益人的 earned media）。
  • 失败模式警告 — 反叛式策略手册在该行业误用后的具体表现，会成为Stage 5反虚荣仪表盘的常设警告。
切勿覆盖功能优先选择。规则是叠加在功能选择之上。如果规则与功能选择冲突（例如：B2B SaaS规则偏向创始人LinkedIn，但功能选择是需求捕获），指出冲突并呈现两种路径，而非合并为一种。
如果用户选择行业领域为「other」，在假设表中标记不匹配，使用最接近的规则继续——说明该规则的哪些结构假设不适用。
然后使用
references/channel-tier-stack.md
搭建渠道体系。分配规则叠加在功能选择和行业领域规则之上：
  • 轻度不对称 → 体系可包含Tier 1、2、3和选择性Tier 4。约70%的精力投入有机渠道（Tier 1 + 2），30%投入付费放大（Tier 3，偶尔Tier 4）。
  • 重度不对称 → Tier 1、2和定向Tier 3。避免广泛Tier 4。通常80/20分配给有机渠道更安全。
  • 绝对不对称 → 仅Tier 1和2。拒绝起草广泛的冷付费Tier 4创意——无效且会消耗预算。如果用户坚持，先解释边际收益递减曲线和反定位策略（见
    references/authenticity-playbook.md
    ），再重新考虑。
输出优先级排序的渠道体系，包含：(a) 选择的主要功能，(b) 映射到该功能的渠道，(c) 每个渠道的每周工作量估算，(d) 明确的70/30（或80/20）分配比例。
反编造规则延续。Stage 2的反编造规则适用于Stage 4输出的所有元素：SEO关键词示例、长尾查询列表、目录/评论网站参考、竞争对手饱和描述、以及体系中展示的任何样本文案。如果需求中未提及行业巨头，本阶段不得出现行业巨头的专有名词任何形式——包括标题大小写（"Sentry alternative"）、小写（"sentry alternative"）、连字符形式，或作为复合关键词或URL的一部分。原本需要品牌名称的搜索查询示例，要么保留
[incumbent]
括号占位符由用户填充，要么改写为非品牌等效表述（"小型团队错误监控"、"100欧元/席位以下的应用监控"、"轻量级Node.js APM"）。这同样适用于Stage 5的竞争对手饱和图和所有Stage 5b输出章节（
## 8
–
## 11
）：禁用专有名词的规则在Stage 2之后不会放宽。

Stage 5 — Shape Menu + Competitor Saturation Map

Stage 5 — 形态菜单 + 竞争对手饱和图

Contract
  • Inputs: selected concept (Stage 2), channel stack + allocation (Stage 4).
  • Outputs: Section 6 (Competitor Saturation Map) + Section 7 (Three Alternative Campaign Shapes). No other sections.
  • Preconditions: Stage 4 complete; allocation stated.
  • Postconditions: 3 shapes presented with tradeoffs; user has been explicitly asked to pick one; turn ends.
  • Failure modes:
    • end of stage reached without a shape selection from the user → end the turn with the selection ask; do NOT produce Sections 8–11.
    • user cannot tell a Self-story (Marshall Ganz framework) → drop the founder-led shape, present community-first / earned-media-first / search-capture-first instead.
    • brief already pre-selects a shape (rare) → confirm in one sentence, proceed to Stage 5b.
Given the selected concept + channel stack, produce only the two items below. Stage 5 ends with the user picking one of the three shapes; do not generate any of the Stage 5b output (
## 8
–
## 11
) until the selection has been made.
  1. Competitor saturation map. Before shapes, before ad copy, before anything: for each of the top 2 named competitors (or top 2 competitor categories if the user did not name specific ones — see Principles), produce:
    (a) What they saturate — the channels, visual style, message tropes, production value, and emotional register the competitor is flooding. Be specific: "paid-heavy LinkedIn carousels with stock illustrations and growth-hack CTAs," not "social media ads."
    (b) The absence that becomes your signal. What is the competitor doing that your refusal to do becomes the positioning? Worked examples:
    • "Fidesz saturated billboards → Tisza's absence from billboards was the message."
    • "Cohort bootcamps saturate paid LinkedIn funnels + affiliate links → our refusal to advertise and our free open curriculum is the message."
    • "SaaS competitors saturate agency-produced demo videos → our terminal-only raw-footage weekly changelog is the message."
    (c) One-sentence positioning line the user commits to holding across the campaign. This is the single sentence every piece of content must reinforce.
  2. Three alternative campaign shapes for executing the concept, with tradeoffs. Examples of shapes: community-first (start with 100 real people, grow through word of mouth), earned-media-first (one newsworthy action drives press + organic amplification), search-capture-first (dominate long-tail high-intent queries where demand already exists). Use the 
    alternative-generator
    pattern — do not collapse to one recommendation prematurely; let the user choose. Each shape must include at least one flagship piece of content structured as Self / Us / Now (Marshall Ganz's organizing framework — see 
    references/authenticity-playbook.md
    ): the leader's lived experience, the shared community reality, and the specific time-bound ask. If the user cannot tell their Self story, drop the founder-led shape and route to community-first, earned-media-first, or search-capture-first instead.
Stop point. Output only sections 
## 6
(Competitor Saturation Map) and 
## 7
(Three Alternative Campaign Shapes), then end the turn with an explicit request: "Pick one of the three shapes (or ask me to revise them) and I will produce the First-30-Days plan, ad copy, lift-test, and anti-vanity dashboard." Do not generate 
## 8
–
## 11
until the user names a shape. If the user's initial brief already pre-selected a shape (rare), confirm that selection in one sentence and proceed to Stage 5b.
契约
  • 输入：选定的概念（Stage 2）、渠道体系+分配（Stage 4）。
  • 输出：Section 6（竞争对手饱和图） + Section 7（三种可选活动形态）。无其他章节。
  • 前置条件：Stage 4完成；已明确分配比例。
  • 后置条件：呈现3种带权衡的形态；已明确要求用户选择一个；回合结束。
  • 失败模式：
    • 阶段结束但用户未选择形态 → 以选择请求结束回合；切勿生成Sections 8–11。
    • 用户无法讲述Self故事（Marshall Ganz框架） → 移除创始人主导形态，改为社区优先/ earned-media优先/搜索捕获优先。
    • 需求已预先选定形态（罕见） → 用一句话确认，进入Stage 5b。
基于选定的概念+渠道体系，仅生成以下两项内容。Stage 5以用户选择三种形态之一结束；在用户做出选择前，切勿生成任何Stage 5b输出（
## 8
–
## 11
）。
  1. 竞争对手饱和图。在形态、广告文案之前，先为前2个指定竞争对手（或用户未指定时的前2个竞争对手类别——见原则）生成：
    (a) 他们的饱和点 — 竞争对手大量投放的渠道、视觉风格、信息 tropes、制作水准和情感基调。要具体：「付费为主的LinkedIn轮播图，配库存插图和增长黑客CTA」，而非「社交媒体广告」。
    (b) 成为您信号的空白点。竞争对手正在做的事情，您拒绝做的部分将成为您的定位。示例：
    • "Fidesz 饱和了广告牌 → Tisza 不在广告牌上出现本身就是信息。"
    • "Cohort训练营饱和了付费LinkedIn漏斗+联盟链接 → 我们拒绝投放广告并提供免费开放课程本身就是信息。"
    • "SaaS竞争对手饱和了代理制作的演示视频 → 我们纯终端的原始每周更新日志本身就是信息。"
    (c) 活动期间需坚守的一句话定位。这是所有内容必须强化的核心句子。
  2. 三种可选活动形态，用于执行概念，各有权衡。形态示例：社区优先（从100个真实用户开始，通过口碑增长）、earned-media优先（一个有新闻价值的行动驱动媒体报道+有机放大）、搜索捕获优先（主导需求已存在的长尾高意向查询）。使用
    alternative-generator
    模式——切勿过早收敛到单一推荐；让用户选择。每种形态必须包含至少一个结构为Self / Us / Now（Marshall Ganz的组织框架——见
    references/authenticity-playbook.md
    ）的旗舰内容：领导者的亲身经历、社区的共同现实、以及具体的限时行动请求。如果用户无法讲述Self故事，移除创始人主导形态，改为社区优先、earned-media优先或搜索捕获优先。
停止点。仅输出
## 6
（竞争对手饱和图）和
## 7
（三种可选活动形态）章节，然后以明确请求结束回合：「选择三种形态之一（或要求我修改），我将生成前30天计划、广告文案、效果测试和反虚荣仪表盘。」 在用户指定形态前，切勿生成
## 8
–
## 11
。如果用户初始需求已预先选定形态（罕见），用一句话确认选择，然后进入Stage 5b。

Stage 5b — Selected-Shape Plan

Stage 5b — 选定形态计划

Contract
  • Inputs: the one shape the user selected from Stage 5; capacity (Stage 1 Q8); paid-channel role from Stage 4 allocation.
  • Outputs: Section 8 (First-30-Days Action List) + Section 9 (Ad Copy + Boost Rules, only if paid applies) + Section 10 (Lift-Test / Measurement Plan) + Section 11 (Anti-Vanity Metric Dashboard).
  • Preconditions: Stage 5 complete AND a specific shape has been named by the user (by index or name).
  • Postconditions: 30-day plan fits within stated capacity ceiling (cuts named if not); a single named lift-test template selected; track/ignore metrics enumerated.
  • Failure modes:
    • entered without a named shape from Stage 5 → ask for the selection; do NOT produce any Section 8–11 content.
    • drafted weekly effort exceeds capacity ceiling → cut lowest-ROI actions until total fits; name the cuts explicitly.
    • earned-media targets unknown / not in brief → flag week-1 research as the action with explicit research method; do not invent outlet names.
    • paid does not apply in the channel stack → omit Section 9 entirely (do not produce empty/placeholder ad copy).
    • asymmetry categorical AND paid in stack → produce a counter-positioning refusal note in Section 9, not cold-paid creative.
Begin only after the user has picked one of the three shapes from Stage 5. If no selection has been received, ask for it and stop. The four items below are produced against the one selected shape — not all three.
  1. First-30-days action list — concrete weekly actions for weeks 1–4, mapped to the chosen shape. Each action has an owner (if multiple people), an effort estimate, and a clear success signal.
    Capacity is a hard constraint. After drafting the week-by-week list, sum total weekly effort. If it exceeds the user's stated capacity (Stage 1 Q8, default 6h/week), cut the lowest-ROI actions until total effort fits inside the ceiling. Name the cuts explicitly: "I am cutting X and Y because the draft came to 14h/week and you said 6h/week. These are the actions to re-add if you can carve out more time later." Do not ship a plan the user cannot execute.
    Earned-media actions must be specific or flagged. Every earned-media action in the list must include:
    • (a) A named target (e.g., "Latent Space podcast, The Pragmatic Engineer newsletter"), or an explicit "target TBD — research is the week-1 action" flag. Never pitch "5 podcasts in the niche" without naming them; that is not an action, that is a wish.
    • (b) A one-sentence pitch hook matched to the target's recent content — not a generic bio blast.
    • (c) Success criteria — reply, mention, guest spot, cross-post, podcast booking, or newsletter feature.
    • (d) Outreach day and channel — Tuesday morning via email, Thursday via LinkedIn DM, etc.
    If the skill does not know specific targets in the user's niche (because the user did not name them and the skill cannot invent names — see Principles), assign target research as the week-1 action and set success criteria for the research itself (e.g., "produce a ranked list of 15 targets with RSS + contact channel by Friday").
    Community-build is a multi-week sub-campaign, not a line item. If a Slack / Discord / WhatsApp / Circle community node is in the channel stack, it gets its own block in the first-30-days list, not one line at 1h/week:
    • Week 0 (before public announcement): pick the platform, write the rules, seed with 10 personal invites from the user's existing network. Dead rooms are worse than no room.
    • Weeks 1–2: founder posts daily for 14 consecutive days. Non-negotiable. Without the founder's daily presence, the community never reaches escape velocity.
    • Weeks 3–4: hand off three recurring rituals (weekly thread, AMA cadence, member spotlight) to 2–3 engaged early members. If no members step up, the community will die when the founder stops; surface this as a validation failure, not a staffing problem.
    • Month 2+: budget 3–4h/week sustained — moderation, weekly post, member welcomes, pruning dead accounts. Underinvest and the community dies.
  2. Ad copy + boost rules — only if paid has a role in the chosen stack:
    • Creative direction in the user's authentic voice (see 
      references/authenticity-playbook.md
      ).
    • The explicit 24–48h organic traction gate: do not boost a post until it has demonstrated genuine organic signal (saves, shares, sustained watch time, thread-depth comments — not raw likes). Likes are cheap and lie.
    • Audience definition: warm retargeting first, lookalikes second, cold audiences only for proven winners with a lift-test plan attached.
    • Frequency cap and creative refresh cadence to avoid fatigue (see ad fatigue section in 
      references/channel-tier-stack.md
      ).
    • Do not bid on your own brand-name keywords without a lift test. Blake, Nosko & Tadelis's eBay field experiment (2015) found weak or no incremental lift from brand- keyword bidding — the traffic arrives organically anyway. Platform-reported ROAS on brand keywords is always excellent because the traffic would have converted regardless. This is one of the most reliable ways established brands waste paid budget. If the user is already bidding on their own brand, require a brand-keyword holdout test (Template 5 in 
      references/lift-test-templates.md
      ) before continued spend.
    • If asymmetry is categorical, refuse to produce broad cold-paid copy. Offer Tier 1/2 content instead and explain why.
  3. Lift-test / measurement plan — mandatory, no exceptions. One concrete experiment using the templates in 
    references/lift-test-templates.md
    . Template selection:
    • Budget exists: Template 1 (geo-holdout) or Template 2/3 (conversion-lift).
    • Tiny budget (<€5k), formal templates underpowered: Template 4 (micro-lift).
    • Brand-keyword bidding already in play: Template 5 (brand-keyword holdout).
    • Zero budget: Template 6 (organic-source attribution) — directional, UTM-tagged, 30-day window, per-channel conversion-rate ranking, cut-the-bottom-20%-reinvest- in-the-top decision rule. Specify:
    • The hypothesis (paid channel X drives incremental action Y on top of organic baseline — or, for Template 6, "channel X outperforms channel Y on conversion rate per unique visitor").
    • Control vs. test group definition (or channel-comparison definition for Template 6).
    • Holdout percentage, duration, and minimum sample size.
    • The incremental metric (not attributed; not platform-reported ROAS). For Template 6, per-channel conversion rate, with the explicit caveat that it is directional only.
    • The decision threshold: what lift level justifies continued spend, what level means stop. For Template 6: top channel → double effort; bottom <20% → drop.
  4. Anti-vanity metric dashboard — the short list of metrics the user should track and the longer list of metrics they should explicitly ignore. Examples:
    • Track: saves, shares, sustained watch time, signed-up volunteers/subscribers, incremental conversions from the lift test, word-of-mouth referrals.
    • Ignore: impressions, CPM, follower count, raw likes, platform-reported attributed ROAS, vanity engagement rate without segmentation.
契约
  • 输入：用户从Stage 5选定的一种形态；可用时间（Stage 1 Q8）；Stage 4分配中的付费渠道角色。
  • 输出：Section 8（前30天行动列表） + Section 9（广告文案+推广规则，仅当付费适用时） + Section 10（效果测试/衡量计划） + Section 11（反虚荣指标仪表盘）。
  • 前置条件：Stage 5完成且用户已指定具体形态（按索引或名称）。
  • 后置条件：30天计划符合指定可用时间上限（如不符合，需明确列出删减内容）；已选择一个具体效果测试模板；已列出需跟踪/忽略的指标。
  • 失败模式：
    • 未从Stage 5获得指定形态就进入本阶段 → 请求用户选择；切勿生成任何Section 8–11内容。
    • 起草的每周工作量超过可用时间上限 → 剔除ROI最低的行动，直至总工作量符合上限；明确列出删减内容。
    • earned-media目标未知/未在需求中提及 → 将第1周的研究列为行动，并明确研究方法；切勿编造媒体名称。
    • 渠道体系中无付费渠道 → 完全省略Section 9（切勿生成空/占位广告文案）。
    • 绝对不对称且体系中有付费渠道 → 在Section 9中生成反定位拒绝说明，而非冷付费创意。
仅在用户从Stage 5选定三种形态之一后开始。如果未收到选择，请求用户选择并停止。以下四项内容针对选定的一种形态生成——而非全部三种。
  1. 前30天行动列表 — 第1-4周的具体每周行动，与选定形态匹配。每个行动有负责人（如有多人）、工作量估算和明确的成功信号。
    可用时间是硬约束。起草完每周列表后，汇总总周工作量。如果超过用户指定的可用时间（Stage 1 Q8，默认每周6小时），剔除ROI最低的行动，直至总工作量符合上限。明确列出删减内容：「我删减了X和Y，因为草稿总工作量为每周14小时，而您指定为每周6小时。如果您能腾出更多时间，可重新添加这些行动。」切勿交付用户无法执行的计划。
    earned-media行动必须具体或标记。列表中的每个earned-media行动必须包含：
    • (a) 指定目标（例如："Latent Space播客，The Pragmatic Engineer通讯"），或明确的「目标待定——第1周行动为研究」标记。切勿笼统地说「向 niche 中的5个播客投稿」而不指定名称；这不是行动，而是愿望。
    • (b) 匹配目标近期内容的一句话投稿钩子——而非通用自我介绍。
    • (c) 成功标准——回复、提及、嘉宾席位、交叉发布、播客预约或通讯报道。
    • (d) ** outreach日期和渠道**——周二上午邮件、周四LinkedIn私信等。
    如果技能不知道用户 niche 中的具体目标（因为用户未提及且技能无法编造名称——见原则），将目标研究列为第1周行动，并为研究本身设定成功标准（例如：「周五前生成包含15个目标的排名列表，附RSS和联系渠道」）。
    社区建设是多周子活动，而非单项任务。如果渠道体系包含Slack/Discord/WhatsApp/Circle社区节点，它在前30天列表中会有单独的模块，而非每周1小时的单项任务：
    • 第0周（公开宣布前）：选择平台，撰写规则，从用户现有网络邀请10个人作为种子用户。死寂的社区不如没有社区。
    • 第1-2周：创始人连续14天每天发帖。这是必须的。没有创始人的日常参与，社区永远无法达到爆发点。
    • 第3-4周：将三个 recurring 仪式（每周主题帖、AMA节奏、成员 spotlight）交给2-3个活跃的早期成员。如果没有成员站出来，创始人停止参与后社区就会消亡；将此视为验证失败，而非人员问题。
    • 第2个月及以后：每周持续投入3-4小时——审核、每周发帖、欢迎新成员、清理僵尸账号。投入不足，社区就会消亡。
  2. 广告文案+推广规则 — 仅当选定体系中有付费渠道时生成：
    • 符合用户真实语气的创意方向（见
      references/authenticity-playbook.md
      ）。
    • 明确的24–48h有机 traction门槛：帖子在展示出真实有机信号（收藏、分享、持续观看时间、深度评论——而非原始点赞）前，切勿推广。点赞廉价且不可靠。
    • 受众定义：优先重定向受众，其次是相似受众，仅为已验证的有效内容附带效果测试计划时才使用冷受众。
    • 频率上限和创意更新节奏，避免受众疲劳（见
      references/channel-tier-stack.md
      中的广告疲劳章节）。
    • 未经效果测试，切勿投放自有品牌关键词竞价。Blake、Nosko & Tadelis的eBay实地实验（2015）发现，品牌关键词竞价的增量提升微弱或无——流量无论如何都会自然到来。平台报告的品牌关键词ROAS总是很高，因为流量无论如何都会转化。这是成熟品牌浪费付费预算最常见的方式之一。如果用户已在投放自有品牌关键词，要求进行品牌关键词保留测试（
      references/lift-test-templates.md
      中的Template 5）后再继续投入。
    • 如果是绝对不对称，拒绝生成广泛的冷付费文案。取而代之，提供Tier 1/2内容并解释原因。
  3. 效果测试/衡量计划 — 强制要求，无例外。使用
    references/lift-test-templates.md
    中的模板进行一项具体实验。模板选择：
    • 有预算：Template 1（地理保留）或Template 2/3（转化提升）。
    • 预算极小(<€5k)，正式模板效果不足：Template 4（微提升）。
    • 已在投放品牌关键词竞价：Template 5（品牌关键词保留）。
    • 零预算：Template 6（有机来源归因）——方向性，带UTM标记，30天窗口，按渠道转化率排名，遵循「砍掉后20%，重新投入前20%」的决策规则。 需明确：
    • 假设（付费渠道X在有机基线之上驱动增量行动Y——或对于Template 6，「渠道X的每独立访客转化率优于渠道Y」）。
    • 对照组 vs 测试组定义（或Template 6的渠道比较定义）。
    • 保留比例、持续时间和最小样本量。
    • 增量指标（非归因；非平台报告ROAS）。对于Template 6，为按渠道转化率，需明确说明仅为方向性指标。
    • 决策阈值：什么提升水平证明继续投入合理，什么水平意味着停止。对于Template 6：顶级渠道 → 加倍投入；底部<20% → 放弃。
  4. 反虚荣指标仪表盘 — 用户应跟踪的短指标列表，以及应明确忽略的长指标列表。示例：
    • 跟踪：收藏、分享、持续观看时间、已注册志愿者/订阅者、效果测试的增量转化、口碑推荐。
    • 忽略：曝光量、CPM、粉丝数、原始点赞、平台报告的归因ROAS、未细分的虚荣互动率。

Output Template

输出模板

Produce the final deliverable in this exact order so the user can scan it and act:
undefined
按以下精确顺序生成最终交付物，方便用户扫描并采取行动：
undefined

Insurgent Campaign Plan — <user / project name>

反叛式营销活动计划 — <用户/项目名称>

0. Assumptions (required under Mode C — default mode; omit under Mode A/B)

0. 假设 (Mode C默认模式下必填；Mode A/B下省略)

<table: field → assumed value, flagging every default applied from Stage 1 so the user can confirm or adjust inline at the end>
<table: 字段 → 假设值，标记Stage 1中应用的所有默认值，方便用户在末尾确认或调整>

1. Campaign Ideas (5+ concepts across archetypes)

1. 活动创意（5+跨原型概念）

<concepts with thesis, archetype, primary tier, authenticity hook, MVP>
<包含核心论点、原型、主要层级、真实性钩子、MVP的概念>

2. Selected Concept

2. 选定概念

<the one (or more) the user picked>
<用户选择的一个（或多个）概念>

3. Spend Asymmetry Verdict

3. 预算不对称 verdict

<mild / severe / categorical, one sentence explaining what it means>
<轻度/重度/绝对，一句话解释含义>

3a. Message-Market-Fit Gate

3a. 信息-市场匹配（MMF）门槛

<3-question score, verdict (confirmed / borderline / failed), and — if borderline or failed — the validation cycle the user must run before proceeding>
<3个问题得分，verdict（已确认/临界/失败），以及——如果临界或失败——用户在继续前必须完成的验证周期>

4. Channel Tier Stack

4. 渠道层级体系

<prioritized channel list with weekly effort>
<带每周工作量的优先级渠道列表>

5. 70/30 (or 80/20) Allocation

5. 70/30（或80/20）分配

<organic % / paid %, with rationale>
<有机/付费比例，附理由>

6. Competitor Saturation Map (Stage 5 output)

6. 竞争对手饱和图 (Stage 5输出)

<per competitor: what they saturate, the absence that becomes your signal, one-sentence positioning line>
<每个竞争对手：他们的饱和点、成为您信号的空白点、一句话定位>

7. Three Alternative Campaign Shapes (Stage 5 output — end the turn here; await shape selection)

7. 三种可选活动形态 (Stage 5输出 — 在此结束回合；等待形态选择)

<three shapes with tradeoffs>
<三种带权衡的形态>

8. First-30-Days Action List (Stage 5b output — only after shape selection)

8. 前30天行动列表 (Stage 5b输出 — 仅在形态选择后)

<week 1–4 concrete actions, scaled to stated capacity with cuts named>
<第1-4周具体行动，按指定可用时间调整，明确列出删减内容>

9. Ad Copy + Boost Rules (Stage 5b output, if paid applies)

9. 广告文案+推广规则 (Stage 5b输出，如付费适用)

<creative direction + 24–48h gate + frequency cap + refusal note if categorical>
<创意方向+24–48h门槛+频率上限+绝对不对称时的拒绝说明>

10. Lift-Test / Measurement Plan (Stage 5b output)

10. 效果测试/衡量计划 (Stage 5b输出)

<one concrete experiment with threshold>
<一项带阈值的具体实验>

11. Anti-Vanity Metric Dashboard (Stage 5b output)

11. 反虚荣指标仪表盘 (Stage 5b输出)

<track list / ignore list> ```
Completeness — do not compress the final deliverable. This is the payload, not a summary of it; the user should never have to run follow-up rounds to pull out the full plan. Fill every applicable section in full prose, not headline fragments. For each recommendation include the why — the mechanism, the tradeoff, the diminishing-returns or authenticity logic behind it — not just the conclusion; the reasoning is what lets the user hold the discipline after the run ends. Concretely: each campaign concept carries its full thesis + archetype + primary tier + authenticity hook + MVP; the channel stack names weekly effort per channel; the action list spells out week-by-week actions with the cuts named; the lift test states hypothesis, control/test groups, duration, incremental metric, and decision threshold. Brevity in structure (scannable headings, tables) is the goal; brevity in content is a failure. If a section does not apply (e.g. no paid in the stack), state why in one line rather than dropping it silently.
<跟踪列表/忽略列表>

**完整性 — 切勿压缩最终交付物**。这是核心内容，而非摘要；用户无需后续跟进即可获取完整计划。每个适用章节都要用完整文字填充，而非标题片段。每个建议都要包含*原因*——背后的机制、权衡、边际收益递减或真实性逻辑——而非仅结论；推理能让用户在技能执行结束后仍坚持原则。具体来说：每个活动概念包含完整的核心论点+原型+主要层级+真实性钩子+MVP；渠道体系列出每个渠道的每周工作量；行动列表明确每周行动及删减内容；效果测试明确假设、对照组/测试组、持续时间、增量指标和决策阈值。目标是*结构*简洁（可扫描的标题、表格）；*内容*简洁是失败。如果某章节不适用（例如：体系中无付费渠道），用一句话说明原因，而非默默删除。

Principles to Hold Throughout

全程遵循的原则

  • Do not sell reach as persuasion. Reach above the first 5–6 impressions does not persuade; it annoys. Say this out loud when recommending frequency caps.
  • Do not propose broad cold-paid as a primary channel for severe or categorical asymmetry. It will not work. Refuse and explain the alternative.
  • Do not generate content that impersonates authenticity the user does not have. If there is no real founder, no real volunteer network, no real earned-media hook, say so and propose how to build one — do not fake it with AI-generated "real-looking" content.
  • Do not accept platform-reported ROAS as proof. Insist on an incremental lift test. Platforms are graded on attributed conversions; the user is graded on actual lift. Academic field experiments (Lewis & Rao 2015, QJE) show digital ad ROI confidence intervals are so wide most campaigns cannot be distinguished from zero. Meta and Google ship Conversion Lift / Brand Lift precisely because they admit this. Cite the work when a user pushes back.
  • Always produce alternatives, not a single recommendation. The user has information you do not; give them a shaped menu and let them choose.
  • Explain the why. When refusing a paid push or a channel, explain the diminishing-returns curve, the counter-positioning move, or the fatigue dynamic. A user who understands the mechanism will hold the discipline after the skill run ends.
  • Do not invent specifics the user did not give. No made-up competitor names, no fabricated budgets, no invented past-campaign references, no hallucinated podcasts or newsletters in the user's niche. If the user did not name them, stay abstract ("cohort- based courses in the category," "mid-size SaaS competitors with paid-growth teams") and say what you are doing: "I am describing competitors at the category level because you did not name specific ones — name them if you want sharper positioning."
  • Vague earned-media targets produce vague results. Force specificity (named target, matched hook, success criteria, outreach day) or flag research as a week-1 action. "Pitch 5 podcasts" is not a plan.
  • A plan the user cannot execute is not a plan. Respect stated weekly capacity; name the cuts required to fit inside it. A 6h/week plan that succeeds beats a 14h/week plan that collapses in week 3.
  • If MMF is failing, the campaign is the wrong problem to solve. Say so. Distribution amplifies signal; it cannot manufacture it. If the MMF gate (Stage 3a) returns 0–1 / 3, refuse the full campaign plan and route the user to a discovery / validation cycle. Running a full insurgent campaign against a broken offer burns the volunteer, community, and founder-attention capital the playbook depends on.
  • Do not over-generalize the Hungarian case. Organic beats paid saturation when preconditions are present: credible insider, accumulated grievance, consolidated challenger, felt pain, threshold-rewarding system, overplayed incumbent. Absent most of these, the playbook alone will not win — name the missing preconditions and recommend building them first.
  • Propaganda and paid advertising sit on the same curve. Troll farms, state disinformation, and commercial ad buys all operate on one diminishing-returns curve and all face the same authenticity collapse at saturation. You can buy reach; you cannot buy belief; above a threshold, buying more reach makes belief harder. The troll farms have not gone away — they have learned this lesson too and will adapt (smaller networks, embedded authenticity, parasocial mimicry). Design for the adapted adversary, not the 2020-era one: lean on verifiable authenticity (real people, real places, real time), narrative coherence, and provenance signals the adversary cannot manufacture without being caught.
  • 切勿将触达等同于说服。超过前5-6次曝光的触达无法说服受众；只会惹恼他们。在推荐频率上限时明确说明这一点。
  • 切勿为重度或绝对不对称的用户将广泛冷付费作为主要渠道。无效。拒绝并解释替代方案。
  • 切勿生成用户不具备的虚假真实性内容。如果没有真实创始人、真实志愿者网络、真实earned-media钩子，如实说明并提出构建方案——切勿用AI生成「看起来真实」的内容造假。
  • 切勿接受平台报告的ROAS作为证据。坚持进行增量效果测试。平台的考核指标是归因转化；用户的考核指标是实际提升。学术实地实验（Lewis & Rao 2015，QJE）表明，数字广告ROI的置信区间非常宽泛，大多数活动的效果在统计上与零无显著差异。Meta和Google推出Conversion Lift/Brand Lift工具，正是因为他们承认这一点。当用户反驳时，引用相关研究。
  • 始终提供替代方案，而非单一推荐。用户掌握您不知道的信息；提供结构化菜单让用户选择。
  • 解释原因。当拒绝付费推广或某个渠道时，解释边际收益递减曲线、反定位策略或受众疲劳动态。理解机制的用户在技能执行结束后仍会坚持原则。
  • 切勿编造用户未提供的细节。不要编造竞争对手名称、预算、过往活动参考、用户 niche 中不存在的播客或通讯。如果用户未提及，保持抽象（「同类 cohort-based 课程」「有付费增长团队的中型SaaS竞争对手」）并说明：「我正在类别层面描述竞争对手，因为您未指定具体名称——如果您想要更精准的定位，请提供名称。」
  • 模糊的earned-media目标产生模糊的结果。强制要求具体（指定目标、匹配钩子、成功标准、 outreach日期），或标记研究为第1周行动。「向5个播客投稿」不是计划。
  • 用户无法执行的计划不是计划。尊重指定的每周可用时间；明确列出为符合时间要求而删减的内容。每周6小时且成功的计划，胜过每周14小时但第3周就崩溃的计划。
  • 如果MMF失败，活动是错误的问题解决方案。明确说明。分发放大信号，但无法制造信号。如果MMF门槛（Stage 3a）得分0–1/3，拒绝完整活动计划，引导用户进入发现/验证周期。针对有缺陷的产品/活动/事业运行完整反叛式活动，会消耗策略手册依赖的志愿者、社区和创始人注意力资源。
  • 切勿过度泛化匈牙利案例。有机触达击败付费饱和仅当满足前置条件时：可信内部人士、累积不满、挑战者阵营整合、切身痛点、达到阈值即获回报的系统、行业巨头过度策略。如果大多数条件缺失，仅靠策略手册无法取胜——列出缺失的前置条件并建议先构建它们。
  • 宣传和付费广告遵循同一曲线。 troll farms、官方虚假信息和商业广告都遵循同一条边际收益递减曲线，在饱和时都会面临真实性崩塌。您可以购买触达，但无法购买信任；超过阈值后，购买更多触达会让信任更难建立。 troll farms并未消失——他们也吸取了教训并会调整（更小的网络、嵌入式真实性、类社交模仿）。针对调整后的对手设计策略：依赖可验证的真实性（真实人物、真实地点、真实时间）、叙事一致性和对手无法伪造而不被发现的来源信号。

Failure Modes Quick Reference

失败模式速查

One-page summary of the per-stage Contracts above. Use this when reviewing an output to check whether a declared failure-mode action was actually taken.
StageTriggerRequired action
1Always-capture field missing under Mode CAsk in the question batch; do not default silently
1Missing field changes Stage 3 / Stage 4 outcomeEscalate into the batch even if normally defaulted
2Competitor unnamed in briefUse 
[incumbent]
/ category descriptor; never invent
2Cannot generate 5 distinct archetypesSay so; present what you have; do not pad
3Preconditions score 0–2Refuse full playbook; route to building preconditions
3Asymmetry numbers absentAsk heuristic questions; do not assume a level
3aMMF score 0–1/3Produce only Section 3a + refusal; STOP — do not generate Sections 4–11
3aMMF score 2/3Insert validation cycle before Section 4
3aUser does not know a signalTreat as a "no"
4Categorical asymmetry + broad cold-paid requestRefuse; route to Tier 1/2 + counter-positioning
4Sector = 
other
Flag rider mismatch; proceed with closest rider
4Function ↔ sector-rider conflictSurface both paths; do not collapse silently
5End of stage without shape selectionRepeat the ask; STOP — do not produce Sections 8–11
5User cannot tell Self-storyDrop founder-led shape; route to community / earned-media / search
5bEntered without named shapeAsk for selection; do not produce any Section 8–11
5bDrafted effort exceeds capacityCut lowest-ROI actions; name the cuts
5bEarned-media targets unknownFlag week-1 research with method; do not invent outlets
5bPaid does not apply in stackOmit Section 9; do not produce placeholder copy
5bCategorical asymmetry + paid in stackProduce counter-positioning refusal in Section 9, not cold-paid creative
上述各阶段契约的单页摘要。用于审核输出时检查是否执行了指定的失败模式操作。
阶段触发条件要求操作
1Mode C下必填字段缺失加入提问批次；切勿默认填充
1缺失字段改变Stage 3 / Stage 4结果加入提问批次，即使通常可默认填充
2需求中未提及竞争对手使用
[incumbent]
/类别描述；切勿编造
2无法生成5个不同原型如实说明；呈现现有内容；切勿填充
3前置条件得分0–2拒绝完整策略手册；引导构建前置条件
3不对称性数据缺失询问启发式问题；切勿假设级别
3aMMF得分0–1/3仅生成Section 3a + 拒绝文本；停止——切勿生成Sections 4–11
3aMMF得分2/3在Section 4前插入验证周期
3a用户不知道某个信号答案视为「否」
4绝对不对称且要求广泛冷付费拒绝；引导至Tier 1/2 + 反定位
4行业领域 = 
other
标记规则不匹配；使用最接近的规则
4功能↔行业规则冲突呈现两种路径；切勿自行合并
5阶段结束未选择形态重复选择请求；停止——切勿生成Sections 8–11
5用户无法讲述Self故事移除创始人主导形态；引导至社区/ earned-media/搜索
5b未获得指定形态就进入请求选择；切勿生成任何Section 8–11内容
5b起草工作量超过可用时间剔除ROI最低的行动；明确列出删减内容
5bearned-media目标未知标记第1周研究并明确方法；切勿编造媒体
5b体系中无付费渠道省略Section 9；切勿生成占位文案
5b绝对不对称且体系中有付费渠道在Section 9中生成反定位拒绝说明，而非冷付费创意

References (read when relevant)

参考资料（相关时阅读）

  • references/campaign-archetypes.md
    — 15+ archetypes the ideation engine draws from.
  • references/asymmetry-audit-table.md
    — decision table for classifying spend asymmetry.
  • references/channel-tier-stack.md
    — Tier 1–4 channels with 2025–2026 benchmark data.
  • references/authenticity-playbook.md
    — founder voice, kitchen-table framing, counter-positioning, narrative coherence.
  • references/lift-test-templates.md
    — six lift-test templates (geo-holdout, conversion-lift, synthetic-control, micro-lift, brand-keyword holdout, zero-budget UTM).
  • references/sector-riders.md
    — six sector-specific overlays applied in Stage 4.
  • references/hungarian-case-study.md
    — Tisza vs. Fidesz 2026 worked example.
  • references/campaign-archetypes.md
    — 创意引擎参考的15+活动原型。
  • references/asymmetry-audit-table.md
    — 预算不对称性分类决策表。
  • references/channel-tier-stack.md
    — Tier 1–4渠道，附2025–2026基准数据。
  • references/authenticity-playbook.md
    — 创始人语气、生活化框架、反定位、叙事一致性。
  • references/lift-test-templates.md
    — 六种效果测试模板（地理保留、转化提升、合成控制、微提升、品牌关键词保留、零预算UTM）。
  • references/sector-riders.md
    — Stage 4中应用的六种行业专属覆盖规则。
  • references/hungarian-case-study.md
    — 2026年Tisza vs Fidesz案例分析。