Generate a controlled draft, not an investment verdict生成受控初稿，而不是自动投资结论
A useful stock analysis report generator starts with a production brief: question, reader, cutoff date, company scope, required sections, permitted calculations, source hierarchy, and prohibited claims. It inventories supplied files and builds from claim blocks. Each block is labeled reported fact, calculation, assumption, judgment, counterpoint, or gap; material facts receive locators, while calculations show inputs and formulas. Release controls compare repeated numbers, test periods and units, verify original citations, require a serious contrary case, and preserve approved versions. Generation accelerates assembly but should not decide whether to buy or sell, invent evidence, or hide uncertainty. The deliverable is the reviewed version, not the first answer.
有用的股票分析报告生成器先接收一份明确生产工单：要回答的问题、读者、资料截止日、公司范围、必选章节、允许的计算、来源优先级与禁止出现的主张。随后清点每份用户文件，并以小型“主张组件”组装报告。每个组件都注明它是报告事实、计算、假设、判断、反方证据还是未决缺口；重大事实带来源定位，计算展示输入与公式。发布前还要检查全文重复数字是否一致、期间和单位是否匹配、引用能否在原件打开，要求存在有力度的反方观点，并保留版本历史与复核签字。生成器可以加快组装与改写，却不应自动给出买卖决定、虚构缺失证据，或用流畅语言隐藏不确定性。真正可交付的是通过审阅的版本，不是第一份生成答案。
Write the production order before choosing sections先写生产工单，再决定报告章节
| Brief field | Instruction | Failure prevented |
|---|---|---|
| Decision question | One answerable question, not “analyze everything” | Generic company biography with no conclusion boundary |
| Reader and length | Name the reviewer and maximum body length | Detail that obscures the decision |
| Cutoff and entity | State date, legal issuer, consolidation scope, currency, and unit | Later information or subsidiary data leaking into the draft |
| Required modules | Select only business mechanism, current evidence, cash/financing, countercase, gaps, and triggers needed | A fixed template dictating irrelevant sections |
| Allowed calculations | Declare formulas and whether adjustments are company-defined or analyst-defined | Unlabeled normalized figures |
| Evidence rules | Define source priority, locator format, and what counts as unresolved | A summary article outranking the filing |
| Prohibited outputs | No automatic recommendation, live price claim, or invented forecast | Polished speculation presented as evidence |
| 工单字段 | 具体指令 | 可防止的失败 |
|---|---|---|
| 判断问题 | 只写一个可回答问题，而不是“分析一切” | 生成没有结论边界的公司简介 |
| 读者与长度 | 指定复核读者和正文上限 | 细节淹没真正判断 |
| 截止日与主体 | 写明日期、法律发行人、合并范围、币种与单位 | 后续信息或子公司单体数据混入 |
| 必选模块 | 只选择本问题需要的业务机制、当前证据、现金/融资、反方、缺口和触发器 | 固定模板强塞无关章节 |
| 允许计算 | 声明公式，并区分公司定义调整与分析者调整 | 规范化数字缺少标签 |
| 证据规则 | 定义来源优先级、定位格式与何种情况算未决 | 摘要文章凌驾正式申报 |
| 禁止输出 | 不得自动建议买卖、声称实时价格或虚构预测 | 把顺滑猜测包装成证据 |
Sections follow the brief. Refinancing needs a debt-maturity module; retention needs supplied cohort evidence. Neither needs every valuation method. Freeze the cutoff so reviewers know what could enter the draft. Valuation requires a dated price or authorized market source. Never fill missing market value from memory or imply an undated number is current.
章节必须服从工单。若问题是近期再融资，就需要债务到期模块；若问题是客户留存，且来源提供 cohort 数据，就应加入 cohort 证据。两者都不需要默认塞进所有估值方法。生成前冻结资料截止日，复核者才能知道哪些信息有资格进入初稿。若确需估值，用户必须提供注明日期的价格或获授权行情来源；生成器不能凭记忆补市场价值，也不能暗示一个无日期数字仍是当前值。
Give six files six distinct jobs六份文件各有职责，不得相互冒名
RailLumen Diagnostics is an invented rail-inspection hardware and monitoring-services company. All FY20X4 and Q2 20X5 documents, figures, names, and conclusions below are illustrative US-dollar material, not a real issuer. The production question is: “Does recent growth show repeatable service economics without increasing cash and financing risk?”
虚构来源包RailLumen Diagnostics 是虚构的铁路检测硬件与监测服务公司。下列 FY20X4 与 20X5 Q2 文件、数字、名称和结论均为美元教学材料，不对应真实发行人。生产问题是：“近期增长是否显示可重复的服务经济性，同时没有增加现金与融资风险？”
| ID and file | Authority in this report | Do not use it for |
|---|---|---|
| S1 — FY20X4 annual filing | Business model, audited-year baseline, risks, accounting policies | A Q2 20X5 balance or later event |
| S2 — Q2 20X5 quarterly filing | Current statements, notes, YTD cash flow, shares, debt, customer concentration | Management’s non-GAAP definition unless included |
| S3 — Q2 earnings release | Headline results, adjusted EBITDA bridge, management framing | Replacing filed statement values |
| S4 — Q2 investor presentation | Backlog chart, service mix, operating KPIs, definitions in slide footnotes | Treating every chart label as an accounting line |
| S5 — earnings-call transcript | What management and analysts said, with speaker and section | Proof that an explanation is economically correct |
| S6 — credit amendment excerpt | Revolver size, maturity, covenant definitions and dates in the excerpt | Terms outside the provided pages or legal advice |
| 编号与文件 | 在本报告中的职责 | 不能用来证明 |
|---|---|---|
| S1 — FY20X4 年度申报 | 商业模式、经审计年度基线、风险与会计政策 | 20X5 Q2 余额或后续事件 |
| S2 — 20X5 Q2 季度申报 | 本期报表、附注、YTD 现金流、股数、债务、客户集中 | 文件未包含的管理层 non-GAAP 定义 |
| S3 — Q2 业绩稿 | 头条业绩、调整后 EBITDA 桥与管理层表述 | 替换正式申报数值 |
| S4 — Q2 投资者演示 | 订单积压图、服务占比、经营 KPI 与幻灯片脚注定义 | 把每个图表标签当会计行项目 |
| S5 — 财报电话会文字稿 | 管理层和分析师说了什么，保留发言人与章节 | 证明解释在经济上必然正确 |
| S6 — 信贷修订节选 | 节选中循环额度、到期日、契约定义与日期 | 推断未提供页面中的条款或给法律意见 |
The manifest records filename, issuer, type, period, publication or filing date, currency, unit, version, page count, searchable-text status, and source ID. Missing pages are gaps, not permission to use general knowledge. InfiniSynapse documentation says scanned PDFs are unsupported in its knowledge-base context; prefer searchable originals and never promise reliable image-only extraction. Remove credentials, account numbers, personal data, and unrelated confidential material.
来源清单要记录文件名、发行主体、文件类型、报告期、发布或申报日期、币种、显示单位、版本、页数、文字是否可搜索与来源 ID。缺页就是缺口，不能用常识补齐。InfiniSynapse 文档明确说明知识库场景不支持扫描 PDF，因此应尽量取得可搜索原件，绝不能承诺可靠抽取纯图像材料。上传前删除凭证、账户号码、个人数据与本问题无关的机密内容。
Assemble claims before composing paragraphs先组装主张组件，再生成自然段
Store the exact value or disclosure, period, unit, basis, source ID, and locator. “Revenue was $104m in Q2 20X5” is a fact after S2 is checked.
报告事实保存精确数值或披露、期间、单位、口径、来源 ID 与定位。核对 S2 后，“20X5 Q2 收入为 1.04 亿美元”才是事实。
Store approved inputs, formula, rounding, and result. Revenue growth = ($104m ÷ $92m) − 1 = 13.0%. The output is derived, not directly reported.
计算保存已批准输入、公式、舍入和结果。收入增长=1.04 亿÷9,200 万−1=13.0%。这是推导值，不是直接报告数。
Declare an input that the source does not establish. Assuming all backlog converts within twelve months would be an analyst assumption and is not used in this report.
假设声明来源没有证实的输入。“假设全部积压订单在十二个月内转化”属于分析者假设，本报告不采用。
Connect evidence proportionately. “Cash conversion weakens the growth case” is a review judgment, not a line from management.
判断按证据强度连接事实。“现金转化削弱增长论点”是复核判断，不是管理层原话。
Attach the strongest evidence that could reverse the draft conclusion: higher service mix and backlog may support future repeatability despite current cash use.
反方主张附上最可能推翻初稿结论的证据：即使当前使用现金，服务占比与积压订单上升仍可能支持未来重复性。
Name the missing evidence and its consequence. No backlog conversion schedule means timing cannot be concluded.
缺口写明缺失证据及影响。没有积压订单转化时间表，就不能判断实现时点。
A claim enters prose only with an owner and status: proposed, source-checked, calculated, challenged, approved, or rejected. Citation coverage = material factual or calculated claims with usable locators ÷ all such material claims. Seven of eight is 87.5%; repairing the eighth makes 100%. Coverage measures locator completeness, not truth. A wrong source can be cited perfectly, or support only part of a sentence. Review support, not mere presence.
主张组件只有在具备负责人和状态后才能进入正文：待处理、来源已核、计算已核、受到挑战、批准或拒绝。引用覆盖率=带可用定位的重大事实或计算主张÷全部重大事实或计算主张。若八项中七项有定位，覆盖率为 87.5%；补齐第八项后为 100%。这个比率只衡量定位完整性，不代表内容必然真实。错误来源也可以被完美引用，而且一条引用可能只支持句子的一半；复核时既要看“有没有”，也要看“是否真正支持”。
Build the RailLumen report from approved components用已批准组件生成 RailLumen 短报告
| Draft section | Approved content | Label and source |
|---|---|---|
| Conclusion boundary | Growth evidence improved, but cash conversion, concentration, and financing keep repeatability unconfirmed | Judgment; supported and contrary components below |
| Operating result | Q2 revenue $104m versus $92m; growth 13.0%. Gross margin $45m ÷ $104m = 43.3% versus $39m ÷ $92m = 42.4%, up 0.9 points | Facts + calculations; S2 pages 6–7 |
| Cost of growth | Operating margin $11m ÷ $104m = 10.6% versus $10m ÷ $92m = 10.9%, down 0.3 points | Calculation; S2 page 7 |
| Cash conversion | 20X5 YTD OCF $14m − PP&E purchases $18m = −$4m FCF under the stated analyst convention | Facts + calculation; S2 page 10 |
| Financing | Debt $72m less cash $26m implies $46m net debt; quarter-end shares 42m versus 40m diluted weighted-average shares | Calculation + facts; S2 pages 8, 20, 24 |
| Forward evidence | Backlog rose from $140m at year-end to $168m, or 20.0%, but is not revenue and lacks a conversion schedule | Facts + gap; S4 slide 12 and definition footnote |
| Countercase and risk | Service mix rose, supporting repeatability; one municipal consortium represented 28% of Q2 revenue, increasing concentration risk | Reported facts + judgment; S2 note 11, S4 slide 9 |
| 初稿章节 | 批准内容 | 标签与来源 |
|---|---|---|
| 结论边界 | 增长证据改善，但现金转化、集中度与融资使重复性仍未确认 | 判断；由下列支持与反方组件支撑 |
| 经营结果 | Q2 收入 1.04 亿，对比 9,200 万，增长 13.0%；毛利率 4,500 万÷1.04 亿=43.3%，上年为 3,900 万÷9,200 万=42.4%，上升 0.9 个百分点 | 事实+计算；S2 第 6—7 页 |
| 增长成本 | 营业利润率 1,100 万÷1.04 亿=10.6%，上年为 1,000 万÷9,200 万=10.9%，下降 0.3 个百分点 | 计算；S2 第 7 页 |
| 现金转化 | 20X5 YTD OCF 1,400 万−购买 PP&E 1,800 万=−400 万 FCF，采用已声明的分析者口径 | 事实+计算；S2 第 10 页 |
| 融资 | 现金 2,600 万、债务 7,200 万，按债务减现金口径得净债务 4,600 万；期末股数 4,200 万，稀释加权平均股数 4,000 万 | 计算+事实；S2 第 8、20、24 页 |
| 前瞻证据 | 积压订单从年末 1.40 亿增至 1.68 亿，即 +20.0%，但它不是收入且没有转化时间表 | 事实+缺口；S4 第 12 页及定义脚注 |
| 反方与风险 | 服务占比上升，支持重复性；一个市政联合体占 Q2 收入 28%，提高集中风险 | 报告事实+判断；S2 附注 11、S4 第 9 页 |
The report must answer the brief and expose where important sentences came from. FY20X4 revenue of $360m is background, not the Q2 growth denominator. S3 reports $18m Q2 adjusted EBITDA; that management-defined measure stays beside its reconciliation and is not renamed operating cash flow. S6 confirms only the excerpted facility terms. Refinancing risk remains a judgment requiring maturity, covenant, liquidity, and scenario evidence.
短报告无需复述六份文件的全部内容，只需回答工单，并暴露每个重大句子的来源。FY20X4 收入 3.60 亿只是背景，不能充当 Q2 增长分母。S3 报告 Q2 调整后 EBITDA 1,800 万，但这一管理层定义指标必须与调节表并存，不能改名为经营现金流。S6 信贷节选只能确认其中列出的额度条款；再融资风险是否可接受仍是判断，需要到期日、契约、流动性与情景证据共同支持。
Three generated errors stopped at three different gates三项生成错误，被三道不同审阅门拦下
| Rejected generated sentence | Gate that fails | Reviewed correction |
|---|---|---|
| “Revenue fell 71.1%, from $360m to $104m.” | Period gate: FY20X4 twelve months compared with Q2 20X5 three months | Use Q2 $92m to $104m: ($104m ÷ $92m) − 1 = +13.0% |
| “The company generated $168m of backlog revenue.” | Definition gate: backlog is an operating measure, not recognized revenue; conversion timing is missing | Report backlog of $168m, +20.0% from $140m, separately from Q2 revenue |
| “Operating cash flow improved to $18m.” | Metric gate: $18m is Q2 adjusted EBITDA in S3; S2 reports $14m YTD OCF | Keep adjusted EBITDA with its bridge; report YTD OCF $14m and analyst-convention FCF −$4m |
| 被拒生成句 | 失败的审阅门 | 复核后修正 |
|---|---|---|
| “收入从 3.60 亿降至 1.04 亿，下降 71.1%。” | 期间门：FY20X4 十二个月与 20X5 Q2 三个月错配 | 使用 Q2 的 9,200 万到 1.04 亿：（1.04 亿÷9,200 万）−1=+13.0% |
| “公司产生了 1.68 亿积压订单收入。” | 定义门：backlog 是经营指标，不是已确认收入；转化时点缺失 | 把 1.68 亿 backlog 及较 1.40 亿增长 20.0% 与 Q2 收入分开报告 |
| “经营现金流改善至 1,800 万。” | 指标门：1,800 万是 S3 的 Q2 调整后 EBITDA；S2 报告 YTD OCF 1,400 万 | 调整后 EBITDA 保留调节桥；报告 YTD OCF 1,400 万及分析者口径 FCF −400 万 |
None of these errors is repaired by adding softer language. The values must return to their source, definition, and period. Keep the rejected sentence, reason code, corrected component, reviewer, and timestamp so the same defect does not reappear during later rewriting.
语言流畅也不能让错误过关这三项错误都不能靠增加模糊措辞修复，数字必须回到正确来源、定义与期间。保留被拒句、原因代码、修正组件、复核人和时间戳，避免后续改写时同一错误再次出现。
Promote versions only when their gate is complete每道门完成后，版本才能晋级
- V0 assembly: ingest only manifest-approved files and produce claim components, not a polished conclusion.
- V1 source review: open every material locator, repair period, unit, sign, entity, definition, and quotation errors, and mark missing evidence.
- V2 numerical review: recompute subtotals, growth, percentage-point changes, cash measures, share denominators, and every repeated number across sections.
- V3 challenge review: add the strongest contrary evidence, test whether the conclusion overstates causality, and distinguish company adjustments from analyst assumptions.
- V4 editorial review: remove duplication, keep qualifications beside claims, and confirm that tables and narrative use the same approved components.
- Release candidate: record source-package hash or version names, cutoff date, citation coverage, unresolved gaps, approver, approval date, and intended audience before any user-chosen export.
- V0 组装：只接收清单批准的文件，先生成主张组件，不急于写顺滑结论。
- V1 来源复核：逐个打开重大定位，修复期间、单位、符号、主体、定义与引语错误，并标记缺失证据。
- V2 数字复核：重算小计、增长率、百分点变化、现金指标、股数分母，以及各章节重复出现的数字。
- V3 挑战复核：加入最强反方证据，检查结论是否夸大因果，并区分公司调整与分析者假设。
- V4 编辑复核：删除重复内容，让限制紧邻主张，并确认表格与正文引用同一批批准组件。
- 候选发布版：在任何由用户选择的导出前，记录来源包哈希或版本名、截止日、引用覆盖率、未决缺口、批准人、批准日期与目标读者。
Do not advertise an export format absent from verified product materials. The record is format-neutral. Final QA asks whether material figures reproduce, citations open to supporting evidence, identical metrics agree, formulas disclose denominators, counterarguments are substantive, gaps stay visible, and approval is dated. Interpretation may fail despite 100% citation coverage, so numerical, source, challenge, and editorial approvals remain separate.
若已核验的产品资料没有明确某种导出格式，就不能擅自宣传。生产记录应保持格式中立，用户可以在其合法选择的位置复核。最终 QA 要确认：重大数字均可复算、每条引用都能打开并真正支持主张、同名指标全文一致、公式披露分母、反方观点有实质内容、缺口仍可见、批准带日期。即使引用覆盖率达到 100%，解释仍可能错误，因此数字、来源、挑战和编辑审批必须分开。
Use Stock Explained as the drafting surface, not the approver把 Stock Explained 当作起草界面，而不是批准人
InfiniSynapse publishes this page and operates Stock Explained, so product statements here are first-party. Public descriptions support organizing a plain-language stock brief from files and text supplied by the user. Broader InfiniSynapse documentation supports file and data-source uploads plus knowledge bases for definitions and prior analyses. This page does not claim a particular export format, automatic market prices or filing collection, confidentiality beyond documented controls, perfect citations, guaranteed accuracy, or automatic investment conclusions.
Upload the production brief and manifest before the documents. Request claim components with label, source ID, locator, period, unit, formula, contrary evidence, gap, and status; generate only approved sections. Open every material locator in the original, recalculate derived numbers, find inconsistent repeats, and challenge uncited causality. Record repairs in new versions. Missing or conflicting evidence stays visible until human review resolves it.
InfiniSynapse 发布本页并运营 Stock Explained，因此这里的产品陈述属于第一方信息。公开描述支持依据用户提供的文件与文本整理通俗股票说明；更广泛的 InfiniSynapse 文档支持文件和数据源上传，也支持用知识库保存定义与历史分析。本页不宣称特定导出格式、自动行情或申报收集、超出文档控制的保密保证、完美引用、准确性保证或自动投资结论。
先上传生产工单与六份文件清单，再上传正文文件。要求主张组件包含标签、来源 ID、定位、期间、单位、公式、反方证据、缺口与状态，然后只生成工单批准的章节。生成后，在上传原件中打开每个重大定位，复算所有推导数，搜索全文重复指标是否冲突，并挑战无引用的因果语言。每次修复形成新版本，不能覆盖审阅轨迹；缺失或冲突证据要保持可见，直到人工复核者解决。
Provide a dated brief and complete source package, generate labeled claim blocks, and verify every material output against the originals.
提供注明日期的工单和完整来源包，生成带标签的主张组件，并用原件核验每项重大输出。
Open Stock Explained打开“一眼看懂这只股票”Questions at the report approval desk报告审批台常见问题
No. A template supplies static headings and fields. A generator assembles selected sections and prose from a particular source package. That added flexibility requires manifest controls, claim labels, citation review, numerical checks, version history, and human approval.
报告生成器等于报告模板吗？不等于。模板提供静态标题与字段；生成器依据特定来源包组装所选章节与正文。灵活性更高，也因此需要来源清单、主张标签、引用复核、数字检查、版本历史与人工批准。
A usable citation identifies the source and opens to a page, note, table, slide, or transcript section that actually supports the claim. A homepage link, inaccessible file, or locator supporting only a different period does not count.
生成股票报告中，什么才算可用引用？可用引用应标明来源，并能打开到真正支持主张的页、附注、表格、幻灯片或文字稿章节。首页链接、无法访问的文件，或只支持另一期间的定位都不算。
No. Coverage measures whether material factual and calculated claims have locators. It does not prove that extraction, arithmetic, classification, source authority, or interpretation is correct. Those require separate source, numerical, and challenge reviews.
引用覆盖率 100% 就代表报告准确吗？不代表。覆盖率只衡量重大事实与计算主张是否带定位，不能证明抽取、算术、分类、来源权威或解释正确；这些仍要经过独立的来源、数字与挑战复核。
It should not. Supply an authorized price with a date when valuation is in scope. Forecasts must come from a cited company range or a clearly labeled analyst assumption. Missing data remains a gap rather than a value invented from model memory.
生成器能补齐缺失股价或预测吗？不应这样做。估值在范围内时，应提供有日期的授权价格；预测必须来自有引用的公司区间，或清楚标记的分析者假设。缺失数据应保持为缺口，不能从模型记忆中虚构。
A named human reviewer with access to the originals. Complex accounting, legal, tax, or investment questions may need a qualified specialist. The tool can draft and organize checks, but it does not assume professional responsibility or provide an automatic recommendation.
最终生成报告由谁批准？应由能访问原件的具名人工复核者批准。复杂会计、法律、税务或投资问题可能需要合格专业人士。工具可以起草并组织检查，但不承担专业责任，也不自动给出建议。
Official sources for product scope and review discipline产品范围与审阅纪律的官方来源
InfiniSynapse publishes this page and operates Stock Explained. Product descriptions are first-party statements, not independent testing. Capabilities and interfaces may change, so verify current official materials before relying on a workflow.
This page is educational and does not provide accounting, audit, legal, tax, market-data, forecasting, or individualized investment advice. RailLumen is fictional. Generated text can contain omissions or errors; primary documents and qualified human review control material conclusions.
InfiniSynapse 发布本页并运营 Stock Explained。产品描述属于第一方陈述，不是独立测试。能力与界面可能变化，依赖某项流程前应核对当前官方资料。
本页仅供学习，不提供会计、审计、法律、税务、市场数据、预测或个性化投资建议。RailLumen 是虚构公司。生成文本可能遗漏或出错；重大结论应由第一手文件与合格人工复核控制。