hk-ipo-analysis

name

hk-ipo-analysis

description

Analyze Hong Kong IPO (new share) listings end-to-end and produce a retail-investor-friendly subscription decision memo. Use when the user mentions a HK stock code in IPO/招股 phase, asks "要不要认购/打新", references 招股书/招股章程/招股价/暗盘/破发/孖展, shares an IPO prospectus PDF or hkexnews link, or asks to evaluate any "(待)上市" Hong Kong company. Covers locating the prospectus on HKEX disclosure (申请版本 + 正式招股章程), pulling deal terms (price range, lot, entry fee, timetable, cornerstones, clawback) from etnet/AASTOCKS/sina, parsing prospectus financials (revenue, margin, cashflow, customer concentration), building a peer-PS/PE comparison, structuring risk factors, ending with an explicit "if I were the user, would I subscribe" stance, and giving post-listing position management rules. Not for already-listed seasoned equities — for those use ordinary equity research instead.

HK IPO Analysis

A repeatable workflow for evaluating a Hong Kong new-share offering and giving the user a clear subscription stance — not just a summary.

When to use

Trigger on any of:

• A 4–5 digit HK stock code in IPO window (e.g. 01236, 01609, 00100).
• User shares a prospectus PDF or hkexnews link.
• Mentions: 招股 / 招股书 / 招股章程 / 招股价 / 入场费 / 每手 / 暗盘 / 破发 / 孖展 / cornerstone / 基石 / clawback / 回拨 / 国际配售.
• Question shape: "要不要认购 X"、"分析下这只新股"、"打新 X"、"X 上市能不能买".

If the company is already listed and seasoned (post-IPO with normal trading history), this skill is the wrong tool — fall back to standard equity analysis.

Workflow checklist

Copy this checklist and tick as you go:

- [ ] Step 0: Time-window gate — is the public offer still open?
- [ ] Step 1: Lock down deal terms (price, lot, fee, timetable, structure, greenshoe)
- [ ] Step 2: Locate AND DOWNLOAD the prospectus PDF locally
- [ ] Step 3: Business + industry positioning + narrative-claim verification + 品类重定义检查
- [ ] Step 3.5: Pre-IPO history check (prior failed IPOs, founder cash-out, valuation anchor)
- [ ] Step 4: Financial track record (revenue, margin, cashflow, R&D, 同店模型 if 线下零售)
- [ ] Step 5: Concentration & dependency risks (clients/suppliers + 客户对称性)
- [ ] Step 6: Implied valuation + peer comparison + 主题进度条
- [ ] Step 7: Risk structure — single-issue dealbreaker scan FIRST, then 4×3 matrix
- [ ] Step 8: Subscription stance using the decision rubric
- [ ] Step 9: Post-listing position management rules (核心仓/交易仓 + 阶梯止盈 + 复评 trigger)

Do not skip steps. If a step's data isn't available, write "未披露/未取得" rather than fabricating.

Helper scripts (optional, faster than manual web fetches)

# Fetch deal terms from etnet + AASTOCKS, output JSON-ish table
scripts/fetch_ipo_terms.sh <code>            # e.g. scripts/fetch_ipo_terms.sh 01609

# Find latest prospectus URL on hkexnews and download to ./prospectus/
scripts/find_prospectus.sh <code> [zh|en]    # default: zh

If scripts fail, fall back to manual etnet/AASTOCKS/hkexnews URLs.

Step 0 — Time-window gate

Compare today against the IPO timetable from etnet/AASTOCKS:

Bucket	What changes
⏳ Pre-S-1 / Pre-招股书（公司还没递表）	远期前瞻分析；只能给"watching only"，禁止给立场或仓位建议；记录关键追踪节点（路演、analyst day、SEC 关键词）
🟦 Before public offer opens	标准认购分析；强调入场费/孖展/基石数学
🟢 Within public offer window	标准认购分析；同上，加上中签率估算
🟡 After public offer closed but before listing day	Reframe：认购不可能；问题变成"暗盘是否追入" / "上市首日是否买"。用暗盘报价（不是 IPO 价）评估追价风险
🟠 Already listed (within 6 months)	半新股阶段；分析变成"现在追入 vs 等基石解锁"；考虑 lockup expiry 时间表
🔴 Already listed (>6 months, seasoned)	Wrong skill；换标准 equity research

State the bucket explicitly at the top of the report. Do not pretend the user can still subscribe when they can't.

Step 1 — Deal terms

Pull these in parallel from at least two independent sources (etnet, AASTOCKS, sina/east money):

Field	Why it matters
Deal type	普通 IPO / 18A（未盈利生物科技）/ 18C（未商业化特专科技；含港股 18C 第 N 家序号） / SPAC / Roll-up — 不同类型估值锚和监管门槛差别巨大
Price range (HK$ low–high)	Defines max subscription cost and implied market cap
价格区间收窄方向	定价上限定 = 需求强；定价中位 = 中性；定价下限或收窄至下半区 = 需求弱（不一定基本面差，配合 Step 6 主题进度条解读）
Lot size + entry fee	Decides minimum capital + retail accessibility
入场费档位	< HK$3,000 = 散户友好（高暗盘活跃度、低中签率）；3K–10K = 中等；>10K = 偏机构
Offer structure (HK public + international)	Float thinness, retail allocation pool
Cornerstones (names, amount, lock-up, type score)	Quality of demand + post-lockup overhang — 用 prospectus-checklist.md 的评分表
Clawback table	How much retail share scales up if oversubscribed
Greenshoe / 绿鞋	有/无 + 上限。无绿鞋 = 上市头 30 天裸奔，列为 🔴
Use of proceeds (% split)	Reveals stage: R&D-heavy / capacity / refinance / M&A
Timetable	招股期 / 定价日 / 暗盘 / 上市日
孖展超购倍数（招股期内每日跟踪）	中签率 + 散户情绪温度
公开发售超购倍数	用于 Step 7 dealbreaker 扫描第 9 项；< 5× 是教科书警告
A+H 标的标识	A 股代码 + AH 折价% — 折价是独立估值维度，不混在估值评分里
Director Share Program / 用户社区配售机制（美股 IPO 特有）	例如 Reddit 把 1.76M 股分配给 mods 和老用户 — 是真实的 community moat 信号，不是噱头；记录配售比例和锁定期

Cite the source for each number. If two sources disagree, flag it.

⚠️ 需求过热反指标

很多人以为"超购倍数越高 = 越值得参与"。实际上：

信号	隐含含义	Stance 影响
超购 > 3000 倍	中签率极低（< 1 手 / 10 万）+ 上市当天获利盘巨大	降低预期持有时长
基石数量 > 15 家	6 个月后供给冲击大	避免长期持仓
公开发售 100% 触发回拨上限（50%）	公开发售部分会被超额申购摊薄	中签率进一步下降

需求过热不等于"基本面好"——它经常意味着"故事被一次性兑现"，参考 01879 曦智案例。

Step 2 — Prospectus retrieval

Try in this order:

1. Run helper: scripts/find_prospectus.sh <code> zh — searches hkexnews and downloads to prospectus/<code>_c.pdf.
2. If helper fails, HKEX disclosure direct PDF at https://www1.hkexnews.hk/app/sehk/<year>/<listing-id>/documents/sehk<YYMMDD><nnnnn>_c.pdf — verify with curl -sI returning HTTP 200 and content-type: application/pdf.
3. If unknown, web-search site:hkexnews.hk <company name in 繁體> 招股章程 and "sehk<YY>" pdf.

⚠️ Always download to local immediately

HKEX replaces application proofs with newer versions without notice — the URL you cite today may 404 tomorrow (this happened in the 01609 case). The helper script writes to prospectus/<code>_<lang>.pdf; if going manual:

curl -sL "<url>" -o prospectus/<code>_<lang>.pdf

Versions

• Application Proof (草擬本): business/financials/risk are real; price/share fields show [編纂]. Acceptable for everything except citing final 招股价.
• Post-Hearing Information Pack (PHIP): closer to final.
• Formal Prospectus (招股章程): registered with Companies Registry; required for citing the final 招股价.

Save the URL and the local copy. State explicitly which version you read and the file's last-modified date.

Step 3 — Business & positioning

Extract from prospectus chapters 「概要」「业务」「行业概览」:

• What does the company actually sell? (avoid pitch deck language)
• Where in the value chain? (component / module / system integrator / brand)
• Market size and the company's stated share — note the methodology (revenue-based? unit-based? niche-defined?).
• Who are the competitors named in 招股书 itself, not the friendly comp set the bankers picked.
• Geographic mix (CN domestic vs overseas) — affects FX, regulatory exposure, premium.

🔍 Narrative-claim verification (mandatory sub-check)

For every "global/China #N" or "largest/leading" claim:

1. Find the slice definition in the industry consultant report (CIC / Frost & Sullivan / iResearch). The narrower the slice, the weaker the moat.
2. Cross-check with total market share. Common pattern: "China #1 by visual-perception-tech-core revenue" = 1.6% total share. Discount accordingly.
3. Check gap to #2. If <1pp, the ranking is fragile.

For every self-described positioning (e.g. "tech company", "全球化"), pull the supporting metric:

• "Tech company" → check R&D % of revenue vs sector average. <50% of sector avg = self-claim contradicted.
• "全球化" → check overseas revenue %.
• "Innovation-driven" → check patent count growth + R&D headcount ratio.

Flag any self-claim vs reality contradiction and carry it to Step 7 dealbreaker scan.

🚀 Category-redefinition check (P0 — added after 老铺 backtest)

某些公司不是在已有品类里做老二老三，而是在重新定义品类——这种情况会出现"经营模型与同行有数量级差异"，必须在 Step 6 切换估值锚，不能机械按老品类 PE/PS 判断。

判别问题（任一为「是」则触发再定价提醒）：

问题	阈值	含义
单店 / 单产品 / 单 SKU 收入是同行的几倍？	≥ 3×	经营模型已超出同品类
客单价是同行的几倍？	≥ 3×	已经在另一个价格带
同店增长 / 单 SKU 增长是同行的几倍？	≥ 5×	不是周期共振，是真实再定价
净利率是同行的几倍？	≥ 2×	商业模型质变
公司自我定位是否使用「文化奢侈品」「品类开创者」「new luxury」「重新定义 X」等表述？	关键词命中	市场可能未识别（这种公司 招股期常被错估）

触发后的动作：

• 在 Step 6 估值章节明确写「这不应按 <老品类> 估值，参考锚为 <新品类> 同行 PE/PS X×」。
• 在 Step 8 stance 评分卡的「估值合理性」一栏给 ✅✅（双勾），并在 Step 9 持仓管理的核心仓比例上限上调（普通 30% → 50%）。
• 在最终立场里显式承认这是潜在 5×+ 标的，避免给"中庸立场"。

Step 3.5 — Pre-IPO history

Often buried across prospectus chapters 「歷史、發展及公司架構」+「主要股東」+「股本」, but decision-critical. Check:

Item	Where to look	Why it matters
Prior IPO attempts (A-share / HK previously)	「歷史」section + web search " 撤单 OR 失效"	Sponsor withdrawal usually signals substantive issues; policy-driven 撤回 ≠ sponsor 主动 = 不触发 dealbreaker
Founder / co-founder share transfers in last 24 months	「歷史」+ 「主要股東」share-transfer table	Large pre-IPO cash-out (>50% of original holding) is a single-issue dealbreaker
Controlling shareholder changes	「主要股東」	Recent change of control = risk
Pre-IPO last-round valuation	「歷史」最新一輪融資 valuation per share	见下方「估值倒挂分情景」
Sponsor / auditor changes	Cover + 「申請文件」	Multiple changes are a red flag
Application lapses (失效)	hkexnews submission history	Multiple lapses (>2) → drafting / disclosure quality issues

State each finding explicitly. If clean, write "无重大 Pre-IPO 红旗".

💡 Pre-IPO 估值倒挂分情景判断 (P1 — added after Reddit backtest)

如果 IPO 价格 < 最近一轮 Pre-IPO 估值（即倒挂 / down-round），不要一刀切判定为负面信号。要分情景：

情景	信号特征	判断
A. 真打折（基本面恶化）	倒挂 + 收入增速放缓 + 毛利压缩 + 用户流失	🔴 谨慎；倒挂幅度反映真实价值损耗
B. 板块系统性折价（市场冷淡）	倒挂 + 同板块多家公司估值同步下移 + 自身基本面稳定	🟡 中性；可视为 IPO 折扣
C. 弹簧效应（基本面改善 + 题材重燃）	倒挂 + 收入增速回升 + 新催化（AI 题材 / 政策红利）+ 倒挂幅度被 IPO 后题材回潮快速抹平	🟢 潜在反弹超调，可能是 +3× 以上的机会（Reddit 案例）

判别问题：「倒挂的根因是公司变差了，还是 Pre-IPO 投资人当时买贵了？」前者是真打折，后者是好机会。

Step 4 — Financial track record

Build a small table from the prospectus 「财务资料」 / 「管理层讨论与分析」:

Metric	FY-2	FY-1	FY (latest)	Interim	Trend?
Revenue					growth rate
Gross margin					up/down
Operating margin
Net profit (IFRS)
Adjusted net profit					esp. for biotech/AI
Operating cash flow					sign matters
R&D % of revenue					for tech
Cash & equivalents					runway

Flag inflection signals: GM bottoming and turning up, OCF flipping positive, segment-mix shift driving margins.

🏪 线下零售 / 消费品的额外字段 (P1 — added after 老铺 backtest)

如果公司是 brick-and-mortar 零售 / 餐饮 / 消费品 标的，必须额外抓这几个字段——它们对未来增速的预测力远高于纯财务指标：

Metric	为什么关键
同店增长 (Same-store sales growth, SSSG)	区分"开店扩张"和"真实需求"。>50% 同店增长 = 极强信号，可能是品类再定义（触发 Step 3 检查）
单店年收入 (Avg revenue per store)	与同行对比；高出 3-5× = 商业模型质变
客单价 / ATV (Average transaction value)	客单价远高于同行 = 不在同一个价格带，估值锚要换
门店数与门店密度	网络饱和度 + 区域天花板
会员复购率 / 客户留存率	长尾收入质量

如果这几项里有任意 2+ 触发「品类重定义」阈值，回到 Step 3 跑 category-redefinition check。

Step 5 — Concentration & dependency risks

Mandatory pull from prospectus 「业务」 + 「风险因素」:

• Top 5 customers % of revenue (each year).
• Top 5 suppliers % of purchase.
• Single-product or single-region dependency.
• Related-party transactions of material size.
• Any pending litigation, regulatory probes, IP disputes.
• Subsidies / tax benefits as % of profit (if profit exists).

🔄 客户依赖对称性检查 (P1 — added after CRWV backtest)

Top-1 客户占比高（>50%）未必是单向依赖——要看双方切换成本是否对称：

维度	单向依赖（🔴）	共生关系（🟡 / 🟢）
客户切走的难度	容易（你是商品化供应）	难（你是稀缺产能）
客户的替代选择	多家可替换	少数几家或仅你一家
客户对你的认股 / 战略入股	无	有（说明客户也在锁定你）
业务上下游捆绑	仅采购关系	联合研发 / 长期合同 / 共建数据中心
行业自身供需平衡	客户产能过剩	客户自身资源不足，反过来需要你（CRWV vs Microsoft）

判断口径：「如果你和客户翻脸，是你死得更快还是客户死得更快？」

• 你死得快 → 真单向依赖（🔴）
• 双方都很疼 → 共生关系（🟡）
• 客户死得更快 → 反向依赖优势（🟢，罕见但很强）

共生关系不应触发 dealbreaker 第 4 项（即使 Top-1 > 70%）。

Step 6 — Implied valuation + peer comparison

1. Compute implied market cap at low / mid / high of the price range using post-IPO total shares (not just H-share float). Cross-check against etnet's 「上市市值」 figure.
2. Choose primary multiple:
   • Profitable + stable: P/E (use latest-year & forward).
   • Loss-making growth: P/S (forward 1–2 years).
   • Asset-heavy: P/B.
3. Build a 3–5 row peer table — name peers explicitly, list their HK code, recent market cap, latest revenue/profit, multiple. Note when picks deviate from prospectus's chosen peer set and why.
4. State: "the IPO is priced at X× vs peer median Y× → premium/discount of Z%".

📈 主题进度条检查 (P0 — added after CRWV / Reddit backtest)

如果公司处于一个明显的宏观主题（AI / 机器人 / 新能源 / 太空 / 量子等），不能只用同行 PS/PE 机械判断——必须先评估主题阶段：

阶段	特征	估值行为	判断口径
🌱 早期（< 12 个月，第一波叙事）	标杆公司刚上市；行业 capex 刚启动；同行少且估值波动剧烈	PS / PE 会反复重估 1-3× ；机械按当前同业中位估值会严重低估	给"等回调买"风险高，可能错过 3-5× 反弹
🔥 中期（12-36 个月，主题铺开）	多家公司上市；估值开始稳定；具体业绩开始检验叙事	PS / PE 在一个区间震荡；机械同业对比可用	标准 Step 6 估值方法适用
📊 晚期（> 36 个月，主题成熟）	行业格局稳定；估值回归基本面；增量资金减少	纯基本面定价；同业平均可信	强调下行保护
❄️ 已退潮（叙事破灭后）	多家破发或下跌 50%+ ；新发行少	估值压到行业历史低位	反向机会

判别参考：

• 早期信号：主题词在过去 6 个月新闻数 +500%+ / 同行公司不到 5 家上市 / 行业 capex 同比 +100%+。
• 早期阶段对 IPO 立场的影响：估值贵不要直接 Skip，至少给"first-week observe"或小额申购，等下一波情绪。

参考 CRWV 案例：定价从 $47-55 弱化到 $40 看似负面，但 AI infra 主题处于早期，6 月行情翻 5× 印证了"早期主题 + 定价弱化 ≠ 长期看跌"。

Step 7 — Structured risk factors

🚨 Single-issue dealbreaker scan (do this FIRST)

Each item below, if HIT, caps position size at "1 lot lottery" regardless of other positives. Run the scan explicitly and report ✅ HIT / ❌ pass for each:

#	Dealbreaker	Threshold
1	Founder / co-founder cash-out > 50% of original holding within 24 months	e.g. 55% → 4.9% (天星案例)
2	Failed prior A-share/HK IPO with sponsor unilateral withdrawal (policy-driven 撤回 不算)	e.g. 科创板撤单
3	Negative operating cash flow for 3+ consecutive years AND no clear inflection	OCF / revenue ratio worsening
4	Top-5 customer concentration > 70% AND rising AND 单向依赖（不对称）	共生关系不触发，参考 Step 5 客户对称性检查
5	Self-claim vs reality contradiction	e.g. "tech co" but R&D % < ½ sector avg (乐动案例)
6	Controlling shareholder changed within 24 months	governance instability
7	Multiple sponsor or auditor changes (>1)	drafting / due-diligence quality
8	Material undecided litigation > 50% of FY net profit (or LTM revenue if loss-making)	tail liability
9	公开发售超购倍数 < 5× AND 基石占比 < 30% AND 区间下限定价	三者同时 = 需求严重不足（黑芝麻案例：1.52× 超购）

Report total dealbreakers HIT. 2+ HIT → no position larger than 1 lot, no margin, regardless of other axes.

4×3 Risk matrix

Then group remaining risks (do not just list):

• Deal-level: pricing aggressiveness, float thinness, cornerstone lockup expiry, clawback uncertainty, broker margin financing risk, greenshoe absent = 🔴.
• Company-level: profitability not yet proven, GM compression, concentration, key-man, share-incentive dilution.
• Sector / regulatory: domestic policy (集采 / VBP / data), overseas data/IP/tariff, technology displacement.
• Macro / market: HSI sentiment, sector IPO pipeline crowding, recent comparable break-issue.

For each, mark severity 🔴 High / 🟡 Medium / 🟢 Low with a one-sentence reason.

Step 8 — Subscription stance (the decision rubric)

End every analysis with an explicit personal stance using this rubric. Do not hedge into uselessness.

Score each axis ✅ / ⚠️ / ❌:

1. Valuation reasonableness — is the implied multiple defensible vs peers + growth?（先做 Step 6 主题进度条 + Step 3 品类重定义后再评分）
2. Quality of growth — is revenue growth backed by margin expansion or just volume at falling price?
3. Path to profit / cash — is breakeven visible within 12–24 months on disclosed trajectory?
4. Demand signals — cornerstone quality, expected oversubscription, sector heat.
5. Downside protection — what's a realistic break-issue scenario, and can the user stomach it?
6. A+H 折价（仅 A+H 标的）— 折价 >30% 是独立安全垫，单独评分不混入"估值"。

Then commit to one of (HK 标准选项):

• Skip — risk/reward unattractive at this price.
• One lot cash, treat as lottery — story interesting but not high-conviction; cap at money you can lose entirely.
• Multi-lot cash, no margin — quality + price both acceptable.
• Margin / 孖展 — only when conviction is high AND clawback math + interest cost work; usually rare.
• Wait for post-listing entry — fundamentals look good but IPO pricing leaves no margin of safety.

⚠️ 如果 Step 3 「品类重定义」检查 HIT，必须额外输出 「潜在 5× 以上标的，建议核心仓上限上调到 50% 而非默认 30%」 的提示——避免给「中庸立场」错杀。

🇺🇸 US IPO 分支

美股 IPO 散户参与机制不同（无"打新一手"概念），所以最终立场要写：

• IPO Allocation 申购 — 仅当用户有 IPO Access（嘉信、富途美股、Robinhood IPO Access、Fidelity IPO 等）；按 1-2 lots 配额申购。
• First-week observe — 没有配额；上市第一周观察价格行为，开盘 +30% 内分批入；>+50% 等回调。
• Wait for first earnings — 等首份季报后再决定。
• Skip — 估值或基本面不符合标准。

如果命中 Step 1 的「Director Share Program / 用户社区配售机制」，在立场里加备注：「community moat 是真信号，提高对 retail oversubscription 的信任度」（Reddit 案例验证）。

⏳ 远期 IPO（Pre-S-1）分支

如果是远期 IPO（尚未递表 SEC / 港交所），禁止给立场。只能给：

• Watching only — 等 S-1 / 招股章程出来后按完整 8 步重跑。
• 列出关键追踪节点（路演、analyst day、SEC 关键词订阅、上游母体股价）。

Always state: "If it were my own money, I would [X] because [one-sentence reason]." This is a personal preference statement, not investment advice — make that disclaimer once at the end.

Step 9 — Post-listing position management (added after backtest, P0)

立场不是分析的终点——上市后必须有持仓管理规则，否则避险了反而会错过反弹（CRWV 案例），或捕到大牛但没拿住（老铺案例）。

🎯 Tier the position

仓位类型	上限（占总仓）	持有逻辑
核心仓（Core）	默认 30%；命中 Step 3 品类重定义则上限 50%	长期持有，跟踪基本面季度变化；除非 dealbreaker 出现否则不动
交易仓（Trade）	70%（默认）/ 50%（核心仓上调时）	按下方阶梯止盈规则操作

如果是 Skip 立场则不存在持仓管理；如果是 1 手 lottery 立场则全部归为核心仓（全亏接受）。

📐 Staged take-profit ladder（替代单一阈值）

不要写"+50% 落袋一半"这种单点规则，会在 +14× 标的上过早卖飞（老铺案例）。改用阶梯：

涨幅触发（vs 入场价）	交易仓兑现比例	累计兑现
+30%	25%	25%
+50%	25%	50%
+100%	25%	75%
+200%	剩余全部	100%
+500%	—	（只剩核心仓继续滚动）

核心仓不在阶梯内，只在以下事件出现时减仓：

• 首份季报增速 < 招股期承诺的 -30% 以上
• Step 3 品类重定义假设被推翻（同店增速回归同行水平）
• 单年涨幅超过 +500% 且估值已到行业极值

🔁 Mandatory re-evaluation triggers

任何一个发生时，必须回到 Step 6 重做估值 + Step 8 重出立场：

Trigger	频率 / 时点
跌破 IPO 价 -20%	立即
基石解锁前 30 天	一次
首份季报发布	立即
上市满 6 个月	一次（特别针对定价弱化标的，如 CRWV）
主题进度条阶段切换（早期 → 中期）	触发即跑
重大新合约 / 新客户公告	立即

Output format

Step 9 在最终报告里以 「持仓管理建议」 section 出现，包括：

## 持仓管理建议
- 核心仓上限：X%
- 交易仓阶梯止盈表
- 复评 trigger 列表
- （首条触发后的预期动作模板）

Output template

Use this structure for the final memo:

# <Company> (<code>.HK) — IPO 评估

## ⏱️ 时间窗口
[认购窗口内 / 已截止 / 已上市 — Step 0 bucket]

## 🎯 最终立场（如果是我）
[1-line action: 跳过 / 一手现金 / 多手现金 / 孖展 / 等上市后买]
[1-sentence reason]

## 一句话定位
[business in plain Chinese, no pitch words]

## 发行条款速查
[deal terms table from Step 1, with greenshoe explicit]

## 招股书来源
[URL + local file + which version + last-modified date]

## 业务与行业位置 + 行研话术核查 + 品类重定义检查
[Step 3 findings]

## Pre-IPO 历史
[Step 3.5 table or "无重大红旗"; 倒挂分情景]

## 财务轨迹
[Step 4 table + inflection commentary; 同店模型 if 线下零售]

## 集中度与依赖风险
[Step 5 bullets + 客户对称性判断 if 单一大客户]

## 估值与可比公司 + 主题进度条
[Step 6 implied market cap + peer table + 主题阶段判断]

## 风险结构化
[Step 7: 🚨 Dealbreaker scan (9 项) FIRST + 4×3 matrix]

## 认购立场详细评分
[Step 8 rubric table + chosen action + one-sentence reason]

## 持仓管理建议
[Step 9: 核心仓 / 交易仓上限 + 阶梯止盈表 + 复评 trigger]

[disclaimer line]

Put time window and final stance at the very top so the user can decide whether to read further.

Anti-patterns

• ❌ Quoting招股价 from a single source without cross-check.
• ❌ Citing "global #1" without explaining the slice.
• ❌ Confusing Application Proof's [編纂] placeholders with "no price disclosed".
• ❌ Ending with "建议关注" or other non-committal banker speak — the user explicitly wants a stance.
• ❌ Using pre-IPO investor entry valuations as "fair value" anchor — they are not.
• ❌ Assuming retail clawback ratios — confirm from the actual prospectus's clawback table.
• ❌ 机械按同业 PS/PE 估值，忽略主题进度条阶段（CRWV 教训）。
• ❌ 看到「同行高出几倍的单店模型」仍按老品类估值（老铺教训）。
• ❌ 客户集中度高 = 单向依赖（共生关系不算，CRWV 教训）。
• ❌ Pre-IPO 倒挂 = 一定看跌（弹簧效应可能反弹超调，Reddit 教训）。
• ❌ 给完立场就完结，不输出持仓管理 / 复评 trigger（错过反弹的根因）。

Additional resources

• Prospectus deep-read checklist + cornerstone scoring: see prospectus-checklist.md.
• HKEX disclosure URL patterns and probe commands: see hkex-urls.md.
• Worked case studies — see cases/README.md for the index. Pick one HK + one US example before first real run to calibrate depth, tone, and stance discipline.