git:20260706.bab1535 to git:20260706.6c09d25

100 added, 101 removed. Audit A to A.

---

name: deep-company-series

category: analysis

description: "Write a publication-grade 8-part deep-dive series on a single company (~120k words total): cognitive reset / moat / profit engine / hidden assets / era variable (e.g. AI) / financials Buffett-style / management / valuation+redlines. The core IP is NOT writing but REVISING — a strict fact-check checklist catches pseudo-precision (probability-weighted expectations, third-party MAU discrepancies, linear extrapolation), absolute language, and cross-article number inconsistencies that most finance long-forms violate. Each piece stands alone but shares one valuation/management/price framework. Use when the user wants textbook-level depth on one company for public publishing (a single research report or earnings note is NOT this — use investment-research / earnings-review instead)."

---

- # 深度公司系列:8 篇长文拆一家公司

+ # Deep-Company Series: An 8-Part Deep Dive on One Company

- 为一家公司撰写 8 篇深度长文系列(约 12 万字),从认知重置到决策框架完整闭环。**核心 IP 不是"会写",而是"会改"——大多数财经文章违反本 skill 的事实核查标准**。

+ Write an 8-part deep-dive series (~120k words total) on a single company, from cognitive reset to a decision framework. **The core IP is not "writing well" but "revising strictly" — most finance long-form violates this skill's fact-check standard.**

- ## 一、触发场景

+ ## 1. When to Use

- 用户希望为一家公司做"教科书级别"的深度研究,以**系列长文**形式公开发布。区别于单篇研报:

- - 8 篇约 12 万字,从认知重置到决策框架完整闭环

- - 每篇独立成文(适合单篇分享),但贯穿一套估值/管理层/价格判断

- - 写给"愿意花 90 分钟读懂一家公司"的读者

+ The user wants "textbook-level" deep research on a company, published as a **series of long-form articles**. Distinct from a single research report:

+ - 8 parts, ~120k words, full loop from cognitive reset to a decision framework

+ - Each part stands alone (shareable singly) but shares one valuation / management / price framework

+ - Written for readers willing to spend 90 minutes understanding one company

- **不适合**:单篇研报、季报点评、行业研究——那些用别的技能(基本面/财报/行业)。

+ **Not for**: a single research report, earnings note, sector study — use other skills (fundamentals / earnings / sector).

- ## 二、系列篇目模板(8 篇)

+ ## 2. Series Template (8 Parts)

- | # | 篇名模板 | 核心问题 | 字数 |

- |---|---------|---------|------|

- | 01 | 你以为你看懂了 X,其实没有 | 认知重置:破 3 个常见错觉 | 4,000-5,000 |

- | 02 | X 的护城河——`<生意本质一句话>` | 护城河深不深、未来 5/10 年还在不在 | 6,000-8,000 |

- | 03 | X 的最大利润引擎——`<最赚钱业务>` | 主业是什么、为什么能持续 | 6,000-8,000 |

- | 04 | X 藏在账上的另一家公司——`<隐藏资产>` | 投资组合 / 子公司 / 隐藏价值 | 8,000-10,000 |

- | 05 | AI(或当下叙事)时代,X 是赢家还是输家 | 时代变量:分业务拆影响 | 8,000-10,000 |

- | 06 | 用巴菲特方式拆 X 的财报 | 财务深度:毛利率/FCF/ROE/SBC | 8,000-10,000 |

- | 07 | `<管理层金句>`——X 的管理层值不值得托付 | 资本配置纪律 + 诚信检验 + 接班人 | 8,000-10,000 |

- | 08 | 多少钱值得买,什么信号必须卖(终章) | DCF 三情景 + 红线清单 + 仓位框架 | 10,000-12,000 |

+ |---|----------------|---------------|-------|

+ | 01 | You think you understand X — you don't | Cognitive reset: break 3 common illusions | 4,000-5,000 |

+ | 02 | X's moat — `{one-line business essence}` | Is the moat deep; will it be there in 5/10 years | 6,000-8,000 |

+ | 03 | X's biggest profit engine — `{most profitable business}` | What is the core business; why it persists | 6,000-8,000 |

+ | 04 | The other company hidden on X's balance sheet — `{hidden asset}` | Investment portfolio / subsidiary / hidden value | 8,000-10,000 |

+ | 05 | In the AI (or current narrative) era, is X a winner or loser | Era variable: decompose the impact by business | 8,000-10,000 |

+ | 06 | Reading X's financials the Buffett way | Financial depth: gross margin / FCF / ROE / SBC | 8,000-10,000 |

+ | 07 | `{management quote}` — is X's management worth entrusting | Capital-allocation discipline + integrity test + succession | 8,000-10,000 |

+ | 08 | At what price to buy, what signal to sell (finale) | DCF 3-scenario + red lines + position framework | 10,000-12,000 |

- 加一篇 `00-系列说明.md` 作目录索引(不发表)。

+ Plus `00-series-overview.md` as an index (unpublished).

- ## 三、写作风格规范

+ ## 3. Writing Style

- ### 语气

- - **直接、犀利、不说废话**——第一句就给数字或反常识结论

- - **价值投资框架**——巴菲特/芒格/段永平/李录视角穿插(不堆砌名言)

- - **不预设立场**——先摆数据、再推逻辑、最后得结论

- - **呈现正反两面**——每个核心判断都附"但另一方面..."的反方

+ ### Voice

+ - **Direct, sharp, no filler** — open with a number or a counterintuitive claim

+ - **Value-investing frame** — Buffett/Munger/Duan Yongping/Li Lu lenses woven in (no name-dropping)

+ - **No preset stance** — data first, logic next, conclusion last

+ - **Show both sides** — every core judgment carries a "but on the other hand..."

+ - **Mobile preview** — the first 18-20 characters must stand alone

- ### 禁用词

+ ### Banned Words

- | 禁用 | 原因 | 替代 |

- |------|------|------|

- | 显然 / 必然 / 一定 | 主观绝对化 | 数据显示 / 证据表明 |

- | 我认为 / 我觉得 | 主观腔调 | 删除或改为"按本框架" |

- | 教科书级别 / 神来之笔 | 流量党褒奖 | 描述具体事实 |

- | 严重不匹配 / 严重低估 | 强主观词 | 给具体折让百分比 |

- | 完美 / 无可挑剔 | 单边判断 | 加上反方观察 |

+ | Banned | Why | Replace with |

+ |--------|-----|--------------|

+ | obviously / inevitably / certainly | Subjective absolutism | "the data shows" / "evidence suggests" |

+ | I think / I feel | Subjective tone | cut, or "under this framework" |

+ | textbook-level / brilliant | Hype adjectives | describe the concrete fact |

+ | severely mismatched / severely undervalued | Strong subjective | give the specific discount % |

+ | perfect / flawless | One-sided | add the counter-observation |

- ### 标题风格

- - 用**反差数字**或**反共识结论**做钩子("15 年 7 次挑战全失败"、"年薪 4292 万占利润 0.0017%")

- - 副标题中性、概括内容

- - **避免流量党比喻**:"小巴菲特"、"中国版 X"、"YYDS" 一律避开

+ ### Title Style

+ - Hook with a **contrast number** or a **counter-consensus claim** ("15 years, 7 failed challenges"; "salary 42.92M = 0.0017% of profit")

+ - Neutral subtitle summarizing content

+ - **Avoid hype metaphors**: "the next Buffett", "the X of China", "GOAT" — all banned

- ## 四、严苛事实核查 Checklist(核心 IP)

+ ## 4. Strict Fact-Check Checklist (the Core IP)

- ### 写之前就要警惕的"伪精确"陷阱

+ ### "Pseudo-precision" traps to watch for before writing

- 1. **概率加权期望值**:`30% × A + 50% × B + 20% × C = 期望 +X%` 几乎全是垃圾——概率分配纯主观,给读者错误精确感。**只列情景 + 触发条件 + 方向,不算加权期望**。

- 2. **第三方测算 MAU/份额**:QuestMobile/七麦/CBNData 等口径差异巨大(同一时点能差 2-3 倍)。**只用最可信的两个对比作 anchor,其他做定性描述**。

- 3. **历史增速线性外推**:`2025 年 +33% × 5 年复合 → 2030 年 X` 是金融文盲式预测。**情景假设 + 高/低区间 + 不是承诺**。

- 4. **未公开的持股比例**:未上市公司持股从未公开披露。**给区间,标"不可知"**。

- 5. **强归因**:竞争对手失败 = 因为 X。多重原因都列出来,**本文不做单一归因**。

+ 1. **Probability-weighted expected value**: `30% × A + 50% × B + 20% × C = expected +X%` is almost always garbage — the probabilities are pure subjective, giving readers false precision. **List scenarios + triggers + direction only; do not compute a weighted expectation.**

+ 2. **Third-party MAU/share estimates**: QuestMobile / 七麦 / CBNData differ hugely (2-3× at the same point). **Use only the two most-credible as anchors; describe the rest qualitatively.**

+ 3. **Linear extrapolation of historical growth**: `2025 +33% × 5y CAGR → 2030 X` is financial illiteracy. **Use scenario assumptions + high/low ranges; never a promise.**

+ 4. **Undisclosed shareholding**: unlisted-company stakes are never publicly disclosed. **Give a range, mark "unknowable".**

+ 5. **Strong attribution**: "competitor failed because of X." List multiple causes; **this article does no single attribution.**

- ### 修订时必跑的 7 项检查

+ ### The 7 mandatory revision checks

```

- □ 1. 跨篇数字一致性:总市值、Non-IFRS 净利润、关键持股 % 全系列对齐

- □ 2. 口径标注:Non-IFRS / GAAP / Non-IFRS-SBC / FCF 各用哪个,全文清楚

- □ 3. 重复加计扫描:已并表子公司不在"投资组合"里、SOTP 不双算

- □ 4. 横向比较公平性:不能"主业 PE(剔除现金+组合)" vs "对手 PE(不剔)"

- □ 5. 概率加权全删:见上一条

- □ 6. 绝对化表述全弱化:搜 "显然|必然|严重|教科书|完美" 逐条改

- □ 7. 第三方数据来源标注:每条非财报数据后跟"(来源:X)"

+ □ 1. Cross-article number consistency: market cap, Non-IFRS net income, key holding % aligned across the series

+ □ 2. Caliber labeling: Non-IFRS / GAAP / Non-IFRS-SBC / FCF — which is used, clear throughout

+ □ 3. Double-counting scan: consolidated subs are NOT in the "investment portfolio"; SOTP doesn't count them twice

+ □ 4. Peer-comparison fairness: don't compare "core-business PE (cash + portfolio stripped)" with "peer PE (not stripped)"

+ □ 5. Probability-weighted expectations deleted (see above)

+ □ 6. Absolute language softened: grep "obviously|inevitably|severely|textbook|perfect"

+ □ 7. Third-party data sourced: every non-filing data point followed by "(source: X)"

```

- ### 模型偏好(写之前先列已知硬错误风险)

- - 历史回报倍数:必须用累计投入口径(如 Riot 33 倍不是 58 倍)

- - 持股比例:必须看最新财报口径(如腾讯持有美团的比例随减持变化)

- - "派息分派"会计处理:视同处置收益按 IFRIC 17 在宣派日确认

- - 总股本会反弹:SBC 集中年初授予会让股本短期上升

+ ### Known hard-error risks (list before writing)

- ## 五、执行流程

+ - Historical return multiples: use cumulative-invested basis (e.g. Riot 33×, not 58×)

+ - Shareholding %: use the latest filing/financial-app basis (e.g. Tencent's Meituan stake changes with disposals)

+ - "Distribution accounting": treated as disposal gain under IFRIC 17, recognized on declaration date

+ - Share count rebounds: SBC granted in clusters at year-start can lift share count short-term

- ### 阶段 1:调研(写 01-02 篇前完成)

+ ## 5. Execution

- 1. 用 `get_financial_statements` 读公司近 5 年年报、最新季报

- 2. 用 `get_research_reports` / `web_search` 读至少 3 份独立卖方研报(找共识 + 反共识)

- 3. 可选:用 `run_swarm`(如 equity_research_team 或 value_investing_committee)先生成内部研究底稿

- 4. **与用户确认 8 篇的核心论点**(避免写完才发现方向不对)

+ ### Phase 1: Research (before writing 01-02)

- ### 阶段 2:写作(按 01→08 顺序写,不跳)

+ 1. `get_financial_statements` — last 5 years of annuals, latest quarterly

+ 2. `get_research_reports` / `web_search` — at least 3 independent sell-side reports (find consensus + dissent)

+ 3. Optional: `run_swarm` (e.g. equity_research_team or value_investing_committee) to generate an internal research draft

+ 4. **Confirm the 8-part core theses with the user** (avoid writing the wrong direction)

- - 每篇写完用 `write_file` 存到 `reports/{公司名}/《看懂{公司名}》/0X-XX.md`

- - 不立即发布——等用户审阅

- - 用户提修订意见后修改

+ ### Phase 2: Writing (01→08 in order, no skipping)

- ### 阶段 3:跨篇一致性扫描(08 篇全部写完后)

+ - After each part, `write_file` to `reports/{company}/《Understanding {company}》/0X-XX.md`

+ - Don't publish immediately — wait for user review

+ - Revise on feedback

- 这是本技能的关键差异化。用工具扫描 8 篇做以下检查:

+ ### Phase 3: Cross-Article Consistency Scan (after all 8)

- 1. 用 `read_file` 逐篇读 + `report_audit`(`command=extract`)抽取每篇的数字(市值、净利润、持股比例、PE 等)

- 2. **跨篇对比同一数字是否一致**——用 `financial_rigor`(`command=cross_validate`)把同一指标在不同篇的值做交叉验证,偏差 >1% 标口径差异

- 3. `read_file` 检查:同一术语(FBS、SBC、Non-IFRS)首次出现是否解释;引用关系("详见 06 篇")是否真的对应;要点回顾 vs 正文数字是否一致

- 4. 绝对化表述扫描:搜 "显然|必然|严重|完美" 逐条弱化

+ This is the key differentiator. Use tools to scan:

- ### 阶段 4:发布前最终核查

+ 1. `read_file` each part + `report_audit` (`command=extract`) to pull numbers (market cap, net income, holding %, PE) from each

+ 2. **Cross-check the same number across parts** — use `financial_rigor` (`command=cross_validate`) to cross-validate the same metric's values across articles; flag >1% deviation as a caliber mismatch

+ 3. `read_file` checks: is each term (FBS, SBC, Non-IFRS) defined at first use; do "see part 06" references actually resolve; do recaps match body numbers

+ 4. Absolute-language scan: grep "obviously|inevitably|severely|perfect" and soften each

- - 用 `report_audit`(`command=verdict`)对每篇做准出门:抽取数字 + 核验 + PASS/FAIL

- - 确认所有数字可追溯、无伪精确、无绝对化

+ ### Phase 4: Pre-publish Final Check

- ## 六、修订意见处理流程

+ - `report_audit` (`command=verdict`) as a gate on each part: extract numbers → verify → PASS/FAIL

+ - Confirm all numbers are traceable, no pseudo-precision, no absolutism

- 用户给修订意见时,按以下顺序:

+ ## 6. Revision-Feedback Handling

- ### 1. 先核查事实(不要直接改)

- 如果用户说"X 数据不对",先用 `get_financial_statements` / `web_search` 找原始数据交叉验证,给出"用户说的 vs 我查到的 vs 我之前用的"三方对比。

+ ### 1. Verify facts first (don't just change)

+ If the user says "X is wrong", use `get_financial_statements` / `web_search` to cross-check the original; present "user's number vs what I found vs what I used".

- ### 2. 判断修订级别

+ ### 2. Grade the revision

- | 级别 | 类型 | 处理 |

- |------|------|------|

- | 🔥 硬错误 | 数字错、归因错、口径错 | 必改 |

- | ⚠️ 主观化 | 强主观词、绝对化、流量党比喻 | 弱化或删除 |

- | 🔬 颗粒度 | 来源标注、口径细化 | 按可读性平衡 |

- | ❓ 不可靠 | 第三方测算差异大 | **删比改更稳** |

+ | Grade | Type | Handle |

+ |-------|------|--------|

+ | 🔥 Hard error | wrong number / attribution / caliber | Must fix |

+ | ⚠️ Subjective | strong subjective word / hype metaphor | Soften or cut |

+ | 🔬 Granularity | source label, caliber refinement | Balance against readability |

+ | ❓ Unreliable | large third-party discrepancies | **Deleting is safer than editing** |

- ### 3. 修订后联动检查

- 修一处先想"哪些地方还引用这个数字/概念":

- - 改了总市值 → 全系列联动改 PE / 主业 PE / 折让 / FCF Yield

- - 改了持股 % → 改 TOP 10 排序 + 历史持股表 + 减持清单

- - 改了术语口径 → 改首次定义 + 后续引用 + 要点回顾

+ ### 3. Cascade check after a fix

+ Before fixing one spot, think "where else is this number/concept referenced":

+ - Market cap changed → cascade to PE / core-business PE / discount / FCF yield

+ - Holding % changed → fix TOP-10 sort + historical holding table + disposal list

+ - Caliber changed → fix first definition + later references + recap

- ## 七、本技能不做什么

+ ## 7. What This Skill Does NOT Do

- - **不替读者做投资决策**——所有篇章末尾"不构成投资建议"

- - **不预测股价**——只给"情景 + 触发条件"

- - **不算"期望年化回报"加权值**——主观概率分配会误导读者

- - **不写"X 大佬也持有"**——用别人持仓为自己的判断背书是反价值投资的

- - **不强求 8 篇都写**——某篇没足够独立内容就合并或减篇数

+ - **Does not make investment decisions for the reader** — every part ends with "not investment advice"

+ - **Does not predict prices** — only "scenarios + triggers"

+ - **Does not compute a weighted "expected annualized return"** — subjective probability misleads

+ - **Does not write "famous investor X also holds"** — using someone else's holding to back your judgment is anti-value-investing

+ - **Does not force all 8 parts** — if a part lacks enough standalone content (e.g. management isn't distinctive), merge it or reduce the count

---

- **一句话总结**:写《看懂 X 系列》的核心能力 ≠ 写得好,而是**改得严**——大多数财经长文死于伪精确数字、主观加权期望值、绝对化表述。本技能的存在就是为了把这些坑全部标记出来,写之前避开,写之后扫干净(用 `report_audit` + `financial_rigor.cross_validate` 扫)。

+ **One-liner**: writing an "Understanding X" series is about **revising strictly, not writing well** — most finance long-form dies from pseudo-precise numbers, subjective weighted expectations, and absolute language. This skill exists to flag all those traps before writing and sweep them clean after (`report_audit` + `financial_rigor.cross_validate`).