A polished memo can still have a broken evidence chain写得流畅的备忘录，证据链仍可能断裂
Stock analysis red flags include relying on secondary sources, mismatching periods or units, treating correlation as causation, adjusting selectively, ignoring denominators or dilution, forcing incomparable peers together, publishing a single-point valuation, substituting management narrative for evidence, omitting refutation and update conditions, and using AI citations that cannot be traced. Audit each memo sentence as claim → primary source → calculation → assumption → judgment → refuter. Repair the chain before changing confidence; do not turn a process flaw into an allegation about the company.
股票分析过程的危险信号包括：只依赖二手来源，错配期间或单位，把相关关系当作因果，选择性调整，忽略分母或稀释，强行比较不可比同业，只给单点估值，以管理层叙述代替证据，不设置反证和更新条件，以及使用无法追溯的 AI 引用。应按“主张→第一手来源→计算→假设→判断→反证”审计每句话。先修复证据链，再改变置信度；不要把分析过程缺陷直接升级为对公司的指控。
Audit the memo before diagnosing the company诊断公司之前先审计备忘录
A company red flag is an observed disclosure or business condition that needs investigation. An analysis red flag is a failure in how evidence becomes a conclusion. The distinction matters. If a memo misreads a unit and reports excessive debt, the supported finding is “the memo’s unit conversion is wrong,” not “the company concealed leverage.” Repairing the method may remove the apparent company problem—or reveal a narrower question worth researching.
公司危险信号是需要调查的披露或经营状况；分析危险信号则是证据转化为结论时发生的方法故障。两者不能混淆。若备忘录误读单位并报告巨额债务，有证据支持的结论是“备忘录换算错误”，而不是“公司隐瞒杠杆”。修复方法后，表面上的公司问题可能消失，也可能留下一个更窄、更值得研究的问题。
| Link in the chain | Audit question | Red flag |
|---|---|---|
| Claim | Is it narrow enough to test? | “The business is excellent” with no measurable condition |
| Source | Can another reader open the original? | Aggregator snippet or AI quotation is the terminal citation |
| Transformation | Can the units, periods and formula be reproduced? | Hidden conversion, adjustment or denominator |
| Inference | Does evidence support the stated mechanism? | Coincidence is described as cause; narrative is treated as proof |
| Comparison | Are definitions and economics aligned? | Peer or historical values answer different questions |
| Refuter | What evidence would reduce confidence? | No contrary test, update event or expiry date |
| 链条环节 | 审计问题 | 危险信号 |
|---|---|---|
| 主张 | 是否足够具体，可以被检验？ | 只写“业务优秀”，没有可衡量条件 |
| 来源 | 另一位读者能否打开原文？ | 聚合器摘要或 AI 引语成为最终引用 |
| 转换 | 单位、期间和公式能否复现？ | 换算、调整或分母被隐藏 |
| 推断 | 证据是否支持所述机制？ | 把同时发生写成因果，把叙述当作证明 |
| 比较 | 定义与经济实质是否对齐？ | 同业或历史数字回答不同问题 |
| 反证 | 什么证据会降低置信度？ | 没有相反检验、更新事件或失效日期 |
Ten correct calculations cannot repair one material claim whose source does not exist. Likewise, a primary filing cannot rescue a conclusion that uses the wrong period or claims causation without a mechanism.
置信度受最弱的重大环节约束十项正确计算也无法修复一项来源不存在的重大主张；同样，即使引用第一手申报，错误期间或没有机制的因果断言仍然不能成立。
Open a deliberately broken memo打开一份故意写错的备忘录
NarrowPeak Metrology is a completely fictional manufacturer of laser measurement instruments with calibration-service contracts. The following CU amounts, periods, documents and market inputs are invented. Assume an analyst is reviewing the nine months ended September 30, 20X5. The supplied evidence pack contains a 10-Q, the prior comparable filing, an earnings release, a transcript and three fictional peer filings.
NarrowPeak Metrology 是一家完全虚构的激光测量仪器制造商，并提供校准服务合同。以下 CU 金额、期间、文件和市场输入均为虚构。假设分析者正在研究截至 20X5 年 9 月 30 日的九个月。证据包包含一份 10-Q、上年可比申报、业绩新闻稿、电话会文字稿和三家虚构同业申报。
“Revenue fell 12.5% from CU360m to CU315m, yet core operating profit is CU44m after removing restructuring. Debt is CU78,500m. Adjusted EPS is CU1.30. R&D growth caused service revenue growth. Three peers trade at 18×, proving fair value is CU23.40 per share. Management’s ‘best-ever pipeline’ guarantees recovery, and an AI summary confirms it.” Every sentence either mixes periods, loses units, hides a denominator, selects an adjustment, overstates causality, skips comparability, converts an assumption into certainty or lacks a traceable citation.
错误备忘录摘录“收入从 CU360m 降至 CU315m，下降 12.5%；剔除重组后核心营业利润为 CU44m；债务为 CU78,500m；调整后 EPS 为 CU1.30；研发增长导致服务收入增长；三家同业均为 18 倍，因此每股公允价值确定为 CU23.40；管理层所称‘史上最佳管线’保证复苏，AI 摘要也予以确认。”这些句子分别混淆期间、丢失单位、隐藏分母、选择性调整、夸大因果、跳过可比性、把假设写成确定事实，或缺少可追溯引用。
Do not repair this excerpt by making the prose more cautious. First preserve it as version M0. Give every sentence a claim ID, then create version M1 only from changes recorded in the memo repair log. This makes editorial improvement distinguishable from evidentiary repair.
不能仅靠把语气写得更保守来修复这段文字。先把原文保存为 M0 版本，为每句话分配 claim ID；随后只依据备忘录修复日志生成 M1。这样才能区分文字润色与证据修复。
Repair provenance, periods, units and denominators修复来源、期间、单位与分母
| Claim ID / process red flag | Source or calculation repair | Rewritten conclusion |
|---|---|---|
| C1 — secondary source only | Replace an aggregator revenue figure with statement page, accession, period and unit; retain the aggregator only as a discovery lead | The filed statement, not the summary, is the cited fact |
| C2 — period mismatch | M0 compares nine-month CU315m with prior full-year CU360m: (315 − 360) ÷ 360 = −12.5%. Match prior nine-month CU288m: (315 − 288) ÷ 288 = +9.4% | Nine-month reported revenue increased 9.4%; drivers remain untested |
| C3 — unit loss | Debt note is presented in CU thousands. CU78,500 thousand ÷ 1,000 = CU78.5m | Debt is CU78.5m at the stated date, before any net-debt interpretation |
| C4 — selective adjustment | GAAP operating profit CU36m includes CU8m restructuring expense and CU5m disposal gain. M0 adds only the expense. A symmetric analyst scenario is 36 + 8 − 5 = CU39m | GAAP is CU36m; CU39m is a labeled sensitivity, not reported “core” profit |
| C5 — denominator and dilution | Adjusted earnings assumption CU31.2m ÷ basic weighted-average shares CU24m = CU1.30. Use diluted weighted-average CU26m: 31.2 ÷ 26 = CU1.20 | M0 overstates this adjusted per-share scenario by 8.3%; verify the reconciliation and dilution instruments |
| 主张 ID/过程危险信号 | 来源或计算修复 | 改写后的结论 |
|---|---|---|
| C1——只有二手来源 | 用报表页码、accession、期间与单位替换聚合器收入数字；聚合器只保留为发现线索 | 被引用事实来自正式申报，而不是摘要 |
| C2——期间错配 | M0 用九个月 CU315m 对比上年全年 CU360m：（315 − 360）÷ 360 = −12.5%。改用上年九个月 CU288m：（315 − 288）÷ 288 = +9.4% | 九个月已报告收入增长 9.4%，但驱动因素尚未检验 |
| C3——单位丢失 | 债务附注以 CU 千元列示；CU78,500 千元 ÷ 1,000 = CU78.5m | 指定日期债务为 CU78.5m；尚未进行净债务解释 |
| C4——选择性调整 | GAAP 营业利润 CU36m 包含 CU8m 重组费用与 CU5m 处置收益。M0 只加回费用；对称的分析者情景为 36 + 8 − 5 = CU39m | GAAP 为 CU36m；CU39m 是已标注敏感性，不是已报告“核心”利润 |
| C5——分母与稀释 | 调整后盈利假设 CU31.2m ÷ 基本加权平均股数 CU24m = CU1.30；改用稀释加权平均 CU26m：31.2 ÷ 26 = CU1.20 | M0 把该调整后每股情景高估 8.3%；仍需核验调节与稀释工具 |
Facts, calculations and adjustments occupy different columns. CU36m, CU8m and CU5m are fictional reported facts. The arithmetic producing CU39m is reproducible. Calling the restructuring item non-recurring is an analyst assumption until its history is checked. Treating CU39m as the better indicator of future profit is judgment. This separation stops arithmetic accuracy from disguising a debatable premise.
事实、计算与调整必须位于不同栏。CU36m、CU8m 与 CU5m 是虚构的已报告事实；得到 CU39m 的算术可以复现；把重组项目称为非经常性，是在检查历史前的分析者假设；认为 CU39m 更能代表未来利润，则属于判断。这样分层可以防止正确算术掩盖有争议的前提。
Repair causality, comparables and valuation claims修复因果、可比集与估值主张
| Claim ID / process red flag | Repair test | Remaining confidence boundary |
|---|---|---|
| C6 — correlation as causation | R&D and service revenue rose together. Test product launches, installed base, pricing, acquisition scope and lag; seek management disclosure rather than infer direction | The memo may report correlation, not that R&D caused the revenue |
| C7 — incomparable peers | Align product/service mix, customer market, capital intensity, accounting basis, fiscal period, growth and margin before retaining each peer | A larger peer list is not better if its economics are different |
| C8 — single-point valuation | The memo assumes CU1.20 normalized EPS and 12×–16× comparable P/E: 1.20 × 12 = CU14.40 and 1.20 × 16 = CU19.20 | The range is a scenario, not fair value fact; both EPS and multiple need their own evidence |
| C9 — management narrative as evidence | Preserve the exact “best-ever pipeline” quote, speaker and date; find definition, amount, conversion, cancellation and margin disclosures | The fact is that management said it; recovery is not thereby proved |
| C10 — no refuter or update condition | Write disconfirmation tests: falling orders, cancellation, weaker conversion margin, share issuance, or revised definitions; attach the next filing date | Confidence expires if the memo is not updated when the trigger arrives |
| C11 — untraceable AI citation | Open the cited file and locate the exact passage, page/table, entity, period and unit. If absent, delete the quotation and record a source gap | Fluent wording is not evidence; no primary location means no supported claim |
| 主张 ID/过程危险信号 | 修复测试 | 剩余置信边界 |
|---|---|---|
| C6——把相关当因果 | 研发和服务收入同时增长；检验新品、安装基础、定价、收购范围与滞后，并寻找管理层披露，而不是推断方向 | 备忘录最多报告相关关系，不能断言研发导致收入 |
| C7——同业不可比 | 保留同业前，对齐产品/服务组合、客户市场、资本强度、会计基础、财政期间、增长与利润率 | 经济实质不同的情况下，同业数量更多并不会更好 |
| C8——单点估值 | 备忘录假设正常化 EPS 为 CU1.20，可比 P/E 为 12–16 倍：1.20 × 12 = CU14.40，1.20 × 16 = CU19.20 | 区间是情景，不是公允价值事实；EPS 与倍数都需要独立证据 |
| C9——管理层叙述替代证据 | 保留“史上最佳管线”的原话、发言人与日期，再找定义、金额、转化、取消和利润率披露 | 事实只是管理层说过这句话，并不能据此证明复苏 |
| C10——没有反证或更新条件 | 写出证伪检验：订单下降、取消、转化利润率减弱、增发或定义改变，并附下一申报日期 | 触发点到来后若不更新，原置信度即失效 |
| C11——AI 引用不可追溯 | 打开被引用文件，定位准确段落、页码或表格、主体、期间与单位；若不存在，删除引语并记录来源缺口 | 流畅文字不是证据；没有第一手位置，就没有获支持的主张 |
CU14.40–CU19.20 is precise arithmetic applied to assumptions. It does not express the probability of either endpoint, establish peer comparability or create a margin of safety. Show sensitivity and uncertainty instead of calling CU16.80 the answer.
增加小数位不会提高区间可信度CU14.40–CU19.20 是对假设进行的精确算术，不代表端点概率，不能证明同业可比，也不会自动产生安全边际。应展示敏感性与不确定性，而不是把 CU16.80 称为唯一答案。
Publish the memo repair log beside version M1在 M1 版本旁发布备忘录修复日志
| Log field | Required entry | NarrowPeak example |
|---|---|---|
| Claim ID and M0 text | Exact original sentence; never rewrite history | C2: “Revenue fell 12.5%” |
| Broken link | Source, scope, transform, inference, comparison or refuter | Period: nine-month versus full-year |
| Primary anchor | Document, accession/version, statement/note, page, period and unit | Current and prior nine-month income statements |
| Repair operation | Replacement source, formula, denominator, reclassification or narrower wording | (315 − 288) ÷ 288 |
| M1 text | New conclusion with fact, assumption and judgment labels | Reported nine-month revenue increased 9.4% |
| Counterevidence | Strongest fact that could weaken M1 | Acquisition contribution may explain the increase |
| Confidence and expiry | Reasoned level plus dated update trigger | Conditional until organic bridge; revisit next filing |
| 日志字段 | 必填内容 | NarrowPeak 示例 |
|---|---|---|
| 主张 ID 与 M0 原文 | 保留准确原句，不能重写历史 | C2：“收入下降 12.5%” |
| 断裂环节 | 来源、范围、转换、推断、比较或反证 | 期间：九个月对全年 |
| 第一手锚点 | 文件、accession/版本、报表或附注、页码、期间与单位 | 本期与上年九个月利润表 |
| 修复操作 | 替换来源、公式、分母、重分类或缩窄措辞 | （315 − 288）÷ 288 |
| M1 文本 | 带事实、假设与判断标签的新结论 | 已报告九个月收入增长 9.4% |
| 反面证据 | 可能削弱 M1 的最强事实 | 收购贡献可能解释增长 |
| 置信度与失效点 | 有理由的等级及带日期更新触发点 | 取得内生增长桥前为条件性；下期复核 |
Confidence is high only for the narrow statement that the source and calculation establish. C2 can support a 9.4% reported change; it cannot support organic growth without acquisition, currency and scope evidence. “Conditional” is not a vague hedge when the missing document and expiry event are named. “Certain” is not a valid substitute for a closed evidence chain.
高置信度只适用于来源与计算真正建立的窄主张。C2 可以支持已报告变化 9.4%，但在没有收购、汇率与范围证据时不能支持内生增长。只要明确缺失文件与失效事件，“条件性”就不是含糊回避；而“确定”也不能替代闭合证据链。
Use a release gate that tests reasoning, not optimism用发布门槛检验推理，而不是乐观程度
- Open every material citation and confirm entity, filing version, page, period and unit.
- Reproduce each calculation from visible inputs; name numerator, denominator, sign and conversion.
- Apply each adjustment policy consistently to gains and charges across comparable periods.
- Remove peers that fail the written comparability criteria; explain what the smaller set can and cannot benchmark.
- Replace causal language with a tested mechanism or a correlation statement.
- Attach one credible refuter and one dated update trigger to every thesis-driving judgment.
- Reject any AI-produced quotation, number or citation that cannot be found in the source file.
- 打开每项重大引用，确认主体、申报版本、页码、期间与单位。
- 用可见输入复现每项计算，注明分子、分母、符号与换算。
- 对可比期间的收益与费用一致应用调整政策。
- 删除不符合书面可比标准的同业，并说明缩小后的集合能与不能衡量什么。
- 用经过检验的机制或相关关系表述替换因果语言。
- 为每个驱动论点的判断附上一项可信反证和一个带日期更新触发点。
- 拒绝任何无法在原文件中找到的 AI 引语、数字或引用。
Use uploaded files to test every memo claim用上传文件检验备忘录中的每项主张
InfiniSynapse is this website’s analysis tool. Upload the memo as version M0, complete primary filings, prior comparable periods, earnings materials, peer filings and the dated valuation-input sheet. Ask for a claim inventory that maps each sentence to document, page or table, period, unit, formula and confidence boundary. A missing citation should remain a gap; the tool should not fill it from an unprovided source or invent a quotation.
Request a memo repair log with M0 text, broken link, primary anchor, recalculation, M1 text, counterevidence and expiry trigger. Then open every material source link and reproduce every number from the original. Check that quoted words actually appear, tables did not shift during extraction, diluted shares were not replaced by basic or ending shares, and an AI summary did not become the terminal source. Tool output organizes verification; it does not certify the memo.
InfiniSynapse 是本网站提供的分析工具。上传 M0 版备忘录、完整第一手申报、上年可比期间、业绩材料、同业申报与带日期的估值输入表。要求生成主张清单，把每句话映射到文件、页码或表格、期间、单位、公式与置信边界。缺少引用就保留为缺口；工具不应从未提供的来源补全，也不能编造引语。
要求输出备忘录修复日志，包含 M0 原文、断裂环节、第一手锚点、重算、M1 文本、反面证据与失效触发点。随后打开每个重大回链，用原文重算所有数字；检查引语是否确实存在、抽取时表格是否错位、稀释股数是否被基本或期末股数替换，以及 AI 摘要是否变成最终来源。工具用于组织核验，不会为备忘录提供认证。
Provide the memo, primary filings, comparison periods, peer documents and dated inputs before asking for a revised conclusion.
在要求改写结论前，先提供备忘录、第一手申报、可比期间、同业文件与带日期输入。
Open Stock Explained打开“一眼看懂这只股票”Questions readers ask next读者接下来常问的问题
No. A secondary source can help discover a question or locate a filing. The red flag appears when it becomes the terminal evidence for a material number or quotation that could be checked against an available primary source.
使用二手来源是否一定是股票分析危险信号？不一定。二手来源可以帮助发现问题或定位申报。当它替代可取得的第一手来源，成为重大数字或引语的最终证据时，才构成过程危险信号。
It proves that management made the quoted statement if the recording or company transcript is authentic and traceable. It does not prove the underlying forecast, causal mechanism or outcome. Those require independent disclosed evidence and later testing.
管理层引语可以算证据吗？若录音或公司文字稿真实且可追溯，它能证明管理层说过这句话，却不能证明底层预测、因果机制或结果；后者需要独立披露证据和后续检验。
Record exact start and end dates, weeks, acquisition scope, currency and seasonal exposure. A nominal quarter-to-quarter comparison may be unsuitable. Narrow the claim, use a comparable-week or year-to-date view if supported, and disclose the remaining mismatch.
财政日历不同时，应该如何比较期间？记录准确起止日、周数、收购范围、币种与季节性敞口。名义季度之间未必可比；在有依据时缩窄主张、使用可比周数或年初至今口径，并披露剩余错配。
No. It means the memo needs a consistent, disclosed policy and must consider economically similar gains and charges. Keep GAAP results visible. Any alternative measure is an analyst scenario unless it is clearly sourced as the company’s reported measure.
对称调整是否意味着所有收益与费用都必须剔除？不是。它要求备忘录采用一致且公开的政策，并考虑经济性质相似的收益与费用。GAAP 结果必须保持可见；除非明确来自公司披露，替代指标都属于分析者情景。
Yes, if the inclusion criteria are written before seeing the multiples and the surviving companies share relevant economics. Report where comparability fails. Three defensible peers may be more informative than fifteen selected only by an industry label.
规模很小的同业集合仍然有用吗？可以，前提是在看到倍数前写好纳入标准，并且保留下来的公司具有相关经济特征。还要报告不可比之处。三家可辩护同业可能比十五家仅按行业标签选择的公司更有信息量。
Remove the claim from the supported conclusion, preserve the failed citation in the repair log, and search the supplied originals by distinctive terms. If no passage exists, mark a source gap. Do not repair it with another uncited AI answer.
找不到 AI 引用时应该怎么办？把该主张从有支持的结论中删除，在修复日志保留失败引用，并用特征词搜索已提供原文。若仍无对应段落，就标记来源缺口；不能再用一个无引用 AI 答案修补。
Sources and limits资料来源与边界
SEC materials support statement roles, filing retrieval and non-GAAP review. FASB’s Concepts Statement supports comparability and verifiability; the NIST profile supports disciplined generative-AI risk review. None validates fictional NarrowPeak, its peers, amounts or memo.
An analysis red flag identifies work that needs repair. It does not establish fraud, reporting error, business weakness, fair value or future return. The calculations here are educational examples, not individualized investment advice.
SEC 材料支持报表职责、申报检索与非 GAAP 复核；FASB Concepts Statement 支持可比性与可验证性；NIST 文件支持生成式 AI 风险复核。任何来源都不会验证虚构 NarrowPeak、同业、金额或备忘录。
分析危险信号只表示工作需要修复，不会据此认定舞弊、报告错误、业务弱点、公允价值或未来回报。文中计算是教育性示例，不是个性化投资建议。
- U.S. Securities and Exchange Commission — Beginners’ Guide to Financial Statements
- U.S. Securities and Exchange Commission — Search Filings and EDGAR Access
- SEC Division of Corporation Finance — Non-GAAP Financial Measures Interpretations
- Financial Accounting Standards Board — Concepts Statement No. 8
- National Institute of Standards and Technology — Generative AI Profile (NIST AI 600-1)