What it should and should not do它适合做什么，不适合做什么
Use an AI stock analysis tool to navigate supplied filings, compare passages, structure calculations, expose gaps, and maintain a reviewable record. Do not ask it to invent unseen documents, supply an undated price, infer intent, predict returns, or issue an automatic trade conclusion. Begin with one question and an official source packet, require a document and page or section locator for every material claim, independently check dates, periods, units, signs, and arithmetic, and leave unsupported points unresolved.
AI 股票分析工具适合减少文件导航工作、对照已提供申报中的段落、组织计算、暴露证据缺口，并维护可复核研究记录。不要要求它编造未提供的文件、给出无日期行情、推断管理层意图、预测收益或自动生成买卖结论。可靠流程从一个明确问题和正式来源包开始；每项重要主张都要附文件及页码或章节定位；日期、期间、单位、正负号和算术必须独立核对；没有来源支持的事项保留为未解决。
Give the AI work it can actually be checked on只把能够被核验的工作交给 AI
| Task | Useful role | Required control |
|---|---|---|
| Find disclosures in supplied files | List candidate pages and nearby text | Open the original location and read the full context |
| Align two filing periods | Create a date, scope, currency, and unit table | Check cover pages, headings, amendments, and column labels |
| Draft a calculation | Show inputs, formula, and result | Recalculate from unrounded source numbers |
| Compare management explanations | Place exact claims and dates side by side | Separate quotation, paraphrase, and inference |
| Generate follow-up questions | Point to gaps and possible alternatives | Decide relevance with human judgment |
| Predict price or returns | Not established by the filing packet | Do not convert document analysis into a forecast or trade signal |
| 任务 | 合适角色 | 必要控制 |
|---|---|---|
| 在已提供文件中寻找披露 | 列出候选页和相邻文本 | 打开原始位置并阅读完整上下文 |
| 对齐两期申报 | 建立日期、范围、币种和单位表 | 核对封面、标题、修订版本和列标签 |
| 起草计算 | 展示输入、公式和结果 | 用未舍入来源数字重新计算 |
| 比较管理层解释 | 把准确主张及日期并排放置 | 区分直接引文、转述和推断 |
| 生成追问 | 指出缺口和可能的替代解释 | 由人工判断其相关性 |
| 预测股价或收益 | 文件包不能建立这种能力 | 不要把文件分析变成预测或交易信号 |
Fluency is not evidence. A confident paragraph can contain the wrong issuer, a superseded amendment, a quarterly number described as annual, a value in thousands read as millions, or a citation that does not support the sentence. Judge the output by whether another reader can reproduce it from the source packet. Speed is useful only after the review path remains visible.
语言流畅不等于存在证据。一段自信的文字仍可能选错发行人、忽略修订、把季度数说成全年数、把千元读成百万元，或给出并不支持结论的引用。评价输出时，应看另一位读者能否依据同一来源包复现，而不是只看文字是否顺畅。只有保留复核路径，节省时间才真正有价值。
InfiniSynapse publishes this page and operates Stock Explained. Its public page describes uploading financial reports, financial data, or research material; translating complex figures into a plain-language company brief; separating facts, calculations, inferences, and information gaps; and continuing verification. This page does not claim that Stock Explained supplies real-time prices, predicts returns, guarantees accuracy, or makes autonomous trades.
产品关系与已核实边界InfiniSynapse 发布本页并运营“一眼看懂这只股票”。该工具公开页面说明：用户可上传财报、财务数据或研究资料，把复杂数字整理为通俗公司说明，并区分事实、计算、推断和信息缺口，继续核验。本文不声称 Stock Explained 提供实时价格、预测收益、保证准确或自动交易。
Seal the question inside an evidence envelope把问题放进边界明确的证据包
- Write one falsifiable question. “Did service renewals weaken in the first half?” is reviewable; “Is this a great stock?” bundles too many unstated decisions.
- Retrieve the filing from the relevant regulator or issuer site. Record issuer legal name, CIK or equivalent identifier, form, filing date, fiscal period, amendment status, currency, and unit.
- Include the earlier filing that defines the baseline. A 10-Q may update a 10-K, but it does not silently replace every annual policy or note.
- List what is deliberately excluded: live price, third-party consensus, private contracts, later events, or documents beyond the cutoff.
- Before uploading any non-public material, confirm authorization and the applicable privacy, retention, and sharing terms. If those terms are not established, use public filings or stop.
- 先写一个可证伪问题。“上半年服务续约是否变弱？”可以核验；“这是不是一只好股票？”则混入太多未说明的判断。
- 从适用监管机构或发行人网站取得申报文件，记录发行人法定名称、CIK 或对应标识、表格类型、提交日、财政期间、修订状态、币种与单位。
- 加入定义基线的早期文件。10-Q 可以更新 10-K，但不会在未说明时取代全部年度政策或附注。
- 列出明确排除项：实时价格、第三方一致预期、私人合同、截止日后事项，或来源包以外的文件。
- 上传任何非公开资料前，确认授权以及适用的隐私、保留与共享条款；若这些条件不明确，就只使用公开申报或停止。
Use original HTML or complete PDFs when possible, not screenshots of selected rows. Preserve filenames rather than calling everything “annual report.” Keep amendments beside the original. Scanned tables may need manual transcription and double entry because OCR can lose minus signs, footnote markers, or column order. A document manifest is part of the result, not clerical overhead.
尽量使用原始 HTML 或完整 PDF，不要只上传选定行的截图。保留文件名，不能把所有文件都笼统称为“年报”；修订文件应与原件并列。扫描表格可能需要人工抄录和双人复核，因为 OCR 会漏掉负号、脚注标记或列顺序。文件清单是研究结果的一部分，不是无意义的行政工作。
Make every answer carry its source contract要求每个答案携带来源契约
| Output field | Required content | Reject when |
|---|---|---|
| Reported fact | File, page or section, table/line, period, unit, and faithful wording | Locator is absent or text says something narrower |
| Calculation | Source inputs, formula, sign convention, and unrounded result | Inputs mix periods or cannot be found |
| Assumption | User-supplied premise, reason, and sensitivity | The output presents it as company disclosure |
| Interpretation | Narrow meaning plus a credible alternative | Causation or intent is asserted from correlation |
| Missing evidence | Exact document, metric, or event needed | A gap is filled with plausible-sounding text |
| Version record | Prompt, source manifest, cutoff, output date, and correction history | Later users cannot reproduce the answer |
| 输出字段 | 必要内容 | 拒绝条件 |
|---|---|---|
| 已披露事实 | 文件、页码或章节、表格或行、期间、单位与忠实措辞 | 没有定位，或原文支持范围更窄 |
| 计算 | 来源输入、公式、符号约定和未舍入结果 | 输入混用期间或无法找到 |
| 假设 | 用户提供的前提、理由和敏感度 | 输出把它写成公司披露 |
| 判断 | 范围有限的含义及可信替代解释 | 仅凭相关性断言因果或意图 |
| 缺失证据 | 仍需取得的准确文件、指标或事件 | 缺口被似是而非的文字填满 |
| 版本记录 | 提示、文件清单、截止日、输出日和更正历史 | 后来读者无法复现 |
Ask for locators without assuming they are reliable. PDF labels may differ from viewer page numbers; HTML filings use sections and anchors. If no locator appears, search the original for the phrase or number and add the verified location manually. A citation is a route to inspection, not a seal of correctness.
应要求返回来源定位，但不能假设界面每次都会给出可靠页码。PDF 印刷页码可能与阅读器页数不同，HTML 申报也常以章节与锚点而非固定页码定位。若输出无法提供定位，就在原文件中搜索短语或数字并手工补充已核实位置。引用只是通往复核的路线，不是正确性印章。
Worked file pass: did RivetVale renewals weaken?文件演练：RivetVale 的续约是否走弱？
RivetVale Instruments is fictional; every fact below is invented. RivetVale_20X5_10-K.pdf was filed February 20, 20X6 for the year ended December 31, 20X5. RivetVale_20X6_Q2_10-Q.pdf was filed August 4 for the 26 weeks ended June 30. Both use CU millions; the 10-K statements are audited and the 10-Q statements unaudited. The question is limited to renewal evidence available by August 4.
RivetVale Instruments 是一家虚构的工业传感器公司，以下名称、日期、数字与披露全部为编制示例。文件包包含 RivetVale_20X5_10-K.pdf：20X6 年 2 月 20 日提交，覆盖截至 20X5 年 12 月 31 日的年度；以及 RivetVale_20X6_Q2_10-Q.pdf：20X6 年 8 月 4 日提交，覆盖截至 20X6 年 6 月 30 日的 26 周。两份文件单位均为 CU 百万；10-K 报表经审计，10-Q 报表未经审计。问题仅限于截至 8 月 4 日可获得的服务续约证据。
| Evidence class | Invented source location | Extracted item | Status |
|---|---|---|---|
| Annual baseline | 20X5 10-K, p.42 revenue table | 20X5 service revenue CU48.0m | Reported fact |
| Comparable interim flow | Q2 10-Q, p.9 revenue table | H1 20X6 service revenue CU23.0m versus CU24.0m in H1 20X5 | Reported fact |
| Annual KPI definition | 20X5 10-K, p.31 | Gross revenue retention 92% for 20X5, excluding currency; management-defined | Reported fact, not updated |
| Point-in-time balance | 10-K note 5 p.67; 10-Q note 4 p.14 | Contract liabilities CU20.0m at year-end and CU16.0m at Q2 | Reported fact |
| Management explanation | Q2 10-Q MD&A p.23 | Billing shifted from annual prepayment toward quarterly invoicing | Reported explanation |
| Missing renewal evidence | No location in supplied Q2 filing | H1 renewal count, cohort retention, cancellations, and price/volume bridge | Unresolved |
| 证据类别 | 虚构来源定位 | 抽取项目 | 状态 |
|---|---|---|---|
| 年度基线 | 20X5 10-K 第 42 页收入表 | 20X5 服务收入 CU48.0 | 已披露事实 |
| 可比中期流量 | Q2 10-Q 第 9 页收入表 | 20X6 上半年服务收入 CU23.0，上年同期 CU24.0 | 已披露事实 |
| 年度 KPI 定义 | 20X5 10-K 第 31 页 | 20X5 总收入留存率 92%，排除汇率；管理层定义 | 已披露但未更新 |
| 时点余额 | 10-K 附注 5 第 67 页；10-Q 附注 4 第 14 页 | 年末合同负债 CU20.0，Q2 为 CU16.0 | 已披露事实 |
| 管理层解释 | Q2 10-Q 的 MD&A 第 23 页 | 开票方式由年度预付更多转向季度开票 | 已披露解释 |
| 缺失续约证据 | 提供的 Q2 文件中没有定位 | 上半年续约数、cohort 留存、取消及价格／销量桥 | 未解决 |
Comparable H1 service revenue changed by (23.0 − 24.0) ÷ 24.0 = −4.17%, rounded to −4.2%. Contract liabilities changed by (16.0 − 20.0) ÷ 20.0 = −20.0%, but those are point-in-time balances, not renewal rates. The packet supports weaker reported service revenue and advance billing, not weaker renewals. Management's billing-shift explanation is not independently proven.
完成计算并不等于回答完成计算层：可比上半年服务收入变化为（23.0−24.0）÷24.0＝−4.17%，四舍五入为 −4.2%。合同负债变化为（16.0−20.0）÷20.0＝−20.0%，但它比较的是两个时点余额，不是续约率。判断层：文件支持已报告服务收入走弱和预付开票减少，却不能证明续约变弱。开票方式变化是管理层提供的替代解释，尚未得到独立证明。
Pass three verification gates before keeping a claim重要主张保留前必须通过三道核验门
Match legal name, identifier, form, accession or filing date, reporting period, and amendment status. Confirm that the latest supplied filing did not restate the cited table or KPI definition.
身份与版本门核对法定名称、标识符、表格类型、accession number 或提交日、报告期间和修订状态；确认已提供的最新文件没有重述被引用表格或 KPI 定义。
Check column heading, year-to-date versus quarter, currency, unit, sign, footnote marker, subtotal scope, and rounding. Recompute the −4.17% from CU23.0m and CU24.0m rather than copying generated arithmetic.
表格与算术门核对列标题、年初至今或单季、币种、单位、正负号、脚注标记、小计范围和舍入；用 CU23.0 与 CU24.0 重算 −4.17%，不能直接复制生成结果。
Read the KPI definition and neighboring note. Ask what else could produce the movement. For RivetVale, lower prepayment, customer loss, price, mix, currency, and contract timing require different evidence.
含义与反证门阅读 KPI 定义和相邻附注，追问还有什么能造成这种变化。对 RivetVale 而言，预付减少、客户流失、价格、组合、汇率和合同时间分别需要不同证据。
Do not reward an answer for agreeing with the initial thesis. Test at least one disconfirming route. If Q3 later reports stable cohort retention but continued lower contract liabilities, the renewal-loss claim weakens and the billing-timing explanation strengthens. If cancellations rise while the KPI definition changes, confidence falls. A useful system keeps those conditions explicit instead of silently rewriting the earlier conclusion.
不能因为答案迎合初始观点就认为它更好，至少要检验一条反证路线。若 Q3 后续披露 cohort 留存稳定但合同负债继续下降，“续约流失”会被削弱，“开票时点变化”会得到更多支持；若取消上升且 KPI 定义又改变，结论置信度则下降。可靠系统应明确保留这些条件，而不是悄悄改写先前结论。
Finish with a record that survives the chat最终形成一份能离开对话继续使用的记录
| Record field | RivetVale entry |
|---|---|
| Question and cutoff | Did H1 20X6 service renewals weaken? Evidence through August 4, 20X6 |
| Supported facts | Comparable service revenue −4.2%; contract liabilities −20.0%; billing-shift explanation disclosed |
| Unsupported proposition | No updated renewal or cohort-retention measure in the supplied Q2 filing |
| Current judgment | Lower revenue and advance billing are visible; renewal deterioration is unconfirmed |
| Alternative explanation | Quarterly invoicing reduced advance balances without the same decline in customer retention |
| Next evidence | Q3 retention definition and value, cancellations, contract-liability roll-forward, and service price/volume bridge |
| Reproduction data | Two filenames, filing dates, page locators, formula cells, prompt version, reviewer, and correction log |
| 记录字段 | RivetVale 条目 |
|---|---|
| 问题与截止日 | 20X6 上半年服务续约是否变弱？证据截止 20X6 年 8 月 4 日 |
| 有支持的事实 | 可比服务收入 −4.2%；合同负债 −20.0%；已披露开票方式变化解释 |
| 无支持的命题 | 已提供 Q2 文件没有更新续约率或 cohort 留存指标 |
| 当前判断 | 收入和预付开票下降可见，但续约恶化尚未确认 |
| 替代解释 | 季度开票降低预收余额，但客户留存未同比下降 |
| 下一项证据 | Q3 留存定义与数值、取消、合同负债滚动及服务价格／销量桥 |
| 复现资料 | 两个文件名、提交日、来源定位、公式单元、提示版本、复核人和更正日志 |
Store the record separately from chat. When a filing arrives, append evidence and preserve the old cutoff. Move a conclusion from unresolved to supported or rejected only when identified evidence changes. AI can support continuity while source selection, accounting interpretation, privacy, and any investment decision remain with the human reviewer.
研究记录应与对话摘要分开保存。新申报到来时，追加证据行并保留旧截止日，不要覆盖历史。只有已识别证据发生变化，结论才能从“未解决”转为“支持”或“否定”。这样既能利用 AI 保持研究连续性，也把来源选择、会计解释、隐私和任何投资决定的责任留给人工复核者。
Use Stock Explained as a reviewable document layer把 Stock Explained 用作可复核的文件层
InfiniSynapse operates Stock Explained and publishes this guide. The verified public workflow accepts financial reports, financial data, or research material and organizes a plain-language company brief that distinguishes facts, calculations, inferences, and gaps. Start with public or authorized files, one narrow question, exact fiscal periods, currency and units, plus a written cutoff.
Request a locator and evidence status for material sentences, then verify both in the originals. Recompute decisive numbers and preserve gaps. Do not request a live quote, future result, unseen document, or automatic trade. Use confidential material only with authorization and after reviewing applicable privacy, storage, retention, and sharing terms; no privacy property is assumed here.
InfiniSynapse 运营 Stock Explained 并发布本指南。经公开页面核实，该工作流接受财报、财务数据或研究资料，整理通俗公司说明，并区分事实、计算、推断和缺口。先准备公开或已获授权的文件、一个窄问题、准确财政期间、币种与单位，并写明资料截止日。
要求每个重要句子返回来源定位和证据状态，但两者都必须在上传原件中复核；关键数字要重算，缺失项要保留。不要要求工具给出实时行情、未来结果、未提供文件或自动交易决定。涉及机密资料时，只有在获得授权并独立阅读适用的隐私、存储、保留与共享条款后才可使用；本文不假定任何未经核实的隐私属性。
Upload an authorized filing packet, require evidence locators, and export a record whose material claims can be checked against the originals.
上传已获授权的申报文件包，要求返回证据定位，并导出一份可把所有重要主张回查原件的记录。
Open Stock Explained打开“一眼看懂这只股票”Questions about a file-first AI workflow关于文件优先 AI 流程的具体问题
Yes, for document questions about a business, reported financials, risks, or filing changes. Valuation tied to market price requires a price, timestamp, currency, exchange, and share basis supplied from a reliable source. Never let the tool invent a missing quote.
AI 股票分析工具没有实时股价也能工作吗？可以，只要问题针对业务、已报告财务、风险或申报变化。与市场价格相连的估值必须提供可靠来源中的价格、时间戳、币种、交易所和股数口径，不能让工具编造缺失报价。
Usually not for an updated conclusion. A 10-K supplies the annual baseline, audit context, notes, risks, and policies; later 10-Qs, 8-Ks, amendments, and proxy material may update it. Select documents according to the question and cutoff rather than uploading volume without purpose.
一份 10-K 足以完成公司分析吗？通常不足以形成更新后的结论。10-K 提供年度基线、审计背景、附注、风险和政策；后续 10-Q、8-K、修订及代理材料可能更新它。应按问题和截止日选择文件，而不是无目的地堆积资料。
First align scope, period, unit, and definition. The 10-Q may present an interim update, a reclassification, or a changed estimate rather than a contradiction. Ask for both locators, read the surrounding notes, and record whether the difference is resolved or remains unexplained.
如果 10-Q 看起来与 10-K 冲突怎么办？先统一范围、期间、单位和定义。10-Q 可能是中期更新、重分类或估计变化，并非真正矛盾。要求返回两处定位，阅读相邻附注，并记录差异已经解决还是仍无解释。
No. The locator can point to the wrong column, omit qualifying text, or support only part of the claim. Open it, compare the wording, check the table headings and footnotes, and narrow the conclusion to what the source actually establishes.
带页码引用就能说明 AI 答案可靠吗？不能。定位可能指向错误列、遗漏限定文字，或只支持主张的一部分。必须打开原处，对照措辞、表头和脚注，并把结论缩小到来源真正能够证明的范围。
Do not upload material you are not authorized to share, personal data without a valid purpose and safeguards, credentials, or confidential company information merely for convenience. Check the actual tool terms and your legal or organizational obligations; public availability of an interface does not answer privacy questions.
哪些内容不应上传到股票分析工具？不要上传无权共享的材料、缺少合法目的和保障的个人数据、凭据，或仅为方便而上传的公司机密。应核对工具实际条款以及法律和组织义务；界面公开可用，并不能回答隐私问题。
A complete packet can improve evidence coverage, but it cannot remove uncertainty or decide personal objectives, risk tolerance, portfolio constraints, taxes, or price discipline. Keep the output as research evidence and judgment, not an autonomous instruction or promised return.
完整申报分析能否直接变成买卖指令？完整资料包可以提高证据覆盖，却不能消除不确定性，也不能决定个人目标、风险承受、组合限制、税务或价格纪律。输出应保持为研究证据与判断，而不是自动指令或收益承诺。
Official sources and operating boundary官方来源与操作边界
The Stock Explained page supports the product description used here. SEC and Investor.gov sources identify official filing access and the roles of 10-K, 10-Q, amendments, and exhibits. The NIST Generative AI Profile documents risks including confidently false content and citations, supporting the requirement for independent verification.
RivetVale, its documents, locators, and figures are fictional. This workflow is not an audit, assurance engagement, accounting opinion, privacy review, price feed, forecast, or individualized investment recommendation. Product interfaces and terms can change; verify the current public page and applicable terms before use.
Stock Explained 公开页面支持本文采用的产品描述。SEC 与 Investor.gov 来源说明正式申报获取方式，以及 10-K、10-Q、修订和附件的作用。NIST 生成式 AI 概览记录了自信输出错误内容或虚假引用等风险，因此重要结果必须独立核验。
RivetVale、其文件、定位和数字均为虚构。本流程不构成审计、鉴证、会计意见、隐私审查、价格数据源、预测或个性化投资建议。产品界面与条款可能变化，使用前应核对当前公开页面和适用条款。