Evidence-First Stock Research
Evidence-led stock research with valuation and safety
Evidence-First Stock Research / 证据优先股票研究
Build an auditable investment-research view from dated evidence to valuation and a bounded research stance. / 从带日期且可审计的证据出发,形成估值结论和边界明确的研究立场。
Non-negotiable boundaries / 不可妥协的边界
- Never place or transmit a trade, broker payload, or standing instruction. / 不得下单或传输交易指令、券商载荷或持续性指令。
- Never request credentials, account identifiers, portfolio values, cookies, API keys, or tokens. / 不得索取凭证、账户标识、组合市值、Cookie、API Key 或 Token。
- Treat every output as research, not personalized investment advice. / 所有输出均为研究内容,而非个性化投资建议。
- Do not provide exact quantities, amounts, order types, limit prices, or executable allocations. / 不提供精确数量、金额、订单类型、限价或可执行仓位指令。
- Fail closed when evidence, freshness, valuation, or source provenance is insufficient. / 当证据、时效性、估值或来源溯源不足时,必须保守失败。
- State that data may be delayed, incomplete, or wrong and that capital can be lost. / 明确说明数据可能延迟、不完整或错误,投资可能损失本金。
Read references/safety.md before producing a recommendation. / 输出建议前读取 references/safety.md。
Workflow / 工作流
1. Define the decision / 定义决策问题
Capture the symbols, market, research question, horizon, comparison set, and review date. / 记录标的、市场、研究问题、时间周期、比较集合和复盘日期。
Choose one mode: / 选择一种模式:
single_stock: one company or security. / 单一公司或证券。multi_compare: normalized peer comparison. / 标准化同业比较。sector_basket: sector, theme, or factor basket. / 行业、主题或因子篮子。thesis_review: test what changed against a prior thesis. / 对照既有论点检验变化。
For a reusable offline checklist, run: / 如需可复用的离线清单,请运行:
python3 scripts/plan_stock_research.py \
--symbols NVDA AMD AVGO \
--market US \
--mode multi_compare \
--question "Compare AI compute exposure"
2. Build the evidence ledger / 建立证据账本
Prefer primary sources: regulator filings, issuer investor-relations material, audited statements, official transcripts, and official macro or industry data. / 优先使用监管文件、公司投资者关系材料、审计财报、官方电话会文本,以及官方宏观或行业数据。
For each material fact, capture publisher, URL, publication time, retrieval time, period covered, timezone where relevant, and whether the value is raw or derived. / 对每个重要事实记录发布方、URL、发布时间、获取时间、覆盖期间、相关时区,以及该数值属于原始数据还是派生数据。
Read references/data-quality.md for source tiers, staleness gates, and conflict handling. / 来源分级、陈旧性门槛和冲突处理请读取 references/data-quality.md。
3. Normalize the operating picture / 标准化经营图景
Use a consistent period and currency. Separate reported facts, consensus estimates, management guidance, and analyst assumptions. / 使用一致的期间和币种,并区分已披露事实、市场一致预期、管理层指引和分析假设。
At minimum, examine: / 至少检查:
- revenue, margins, earnings, free cash flow, returns on capital, and balance-sheet risk; / 收入、利润率、盈利、自由现金流、资本回报和资产负债表风险;
- segment economics, dilution, buybacks, dividends, and capital intensity; / 分部经济性、稀释、回购、分红和资本密集度;
- direct positive evidence, direct negative evidence, catalysts, and invalidation conditions. / 直接正面证据、直接负面证据、催化剂和失效条件。
4. Value the security / 对证券估值
Valuation is mandatory for a completed view. Use methods appropriate to the business, such as peer and historical multiples, DCF or reverse DCF, SOTP, asset-backed valuation, or a bear/base/bull scenario table. / 完整观点必须包含估值;应根据业务选择同业与历史倍数、DCF 或反向 DCF、SOTP、资产价值法,或熊市、基准、牛市情景表。
State assumptions, fair-value or scenario range, sensitivity drivers, and why each method fits. / 说明假设、公允价值或情景区间、敏感性驱动因素,以及各方法适用的原因。
Read references/valuation.md before finalizing the valuation section. / 完成估值部分前读取 references/valuation.md。
5. Form a falsifiable thesis / 形成可证伪论点
State the strongest supported thesis, the non-consensus element if one exists, the KPI or event that can disprove it, and a timebox for review. / 说明证据最充分的论点、存在时的非共识部分、可推翻它的 KPI 或事件,以及复盘时间限制。
If no defensible variant view exists, say so. / 如果不存在可辩护的差异化观点,应明确说明。
6. Issue a bounded research stance / 给出有边界的研究立场
Use one of: / 使用以下之一:
positive candidate: evidence and valuation support deeper review. / 正面候选:证据和估值支持进一步研究。hold thesis: the existing thesis remains supported, without an execution instruction. / 持有论点:既有论点仍获支持,但不形成执行指令。watch: promising but blocked by missing evidence, freshness, or valuation. / 观察:具有潜力,但被证据、时效性或估值缺口阻断。pass: risk-reward or evidence quality is insufficient. / 放弃:风险收益或证据质量不足。risk review: direct negative evidence requires reassessment. / 风险复核:直接负面证据要求重新评估。
Include confidence, valuation range, trigger, invalidation condition, evidence grade, and unresolved gaps. / 包含置信度、估值区间、触发条件、失效条件、证据等级和未解决缺口。
7. Validate completion / 验证完成度
Represent the result as JSON and run the deterministic gate when useful: / 如适用,将结果表示为 JSON 并运行确定性门禁:
python3 scripts/validate_research_packet.py research.json --json
Do not call the research complete if the validator rejects missing valuation, undated sources, missing disclaimer, or execution fields. / 如果验证器发现缺失估值、来源未标日期、缺少免责声明或包含执行字段,不得声称研究已完成。
Output contract / 输出契约
Return concise Markdown containing: / 返回包含以下内容的简洁 Markdown:
- Research question and as-of time. / 研究问题与数据截至时间。
- Evidence ledger with source quality and conflicts. / 包含来源质量和冲突的证据账本。
- Operating and financial normalization. / 经营与财务标准化。
- Valuation assumptions and scenario range. / 估值假设与情景区间。
- Falsifiable thesis and disconfirming evidence. / 可证伪论点与反证。
- Research stance, confidence, trigger, and invalidation. / 研究立场、置信度、触发条件和失效条件。
- Risks, missing evidence, and next review date. / 风险、缺失证据和下次复盘日期。
- Disclaimer: research only, not personalized investment advice; data may be delayed or wrong; loss of capital is possible. / 免责声明:仅供研究,并非个性化投资建议;数据可能延迟或错误;投资可能损失本金。
Read references/workflows.md for mode-specific checklists. / 各模式专用清单请读取 references/workflows.md。
Version history / 版本历史
0.1.1(2026-08-30): Added an explicit Apache-2.0 OR MIT-0 dual-license contract for consistent GitHub and ClawHub distribution. / 新增明确的 Apache-2.0 OR MIT-0 双许可证契约,使 GitHub 与 ClawHub 分发保持一致。0.1.0(2026-08-30): Initial public community edition with evidence, valuation, safety, and no-trade gates. / 首个公开社区版本,包含证据、估值、安全和禁止交易门禁。
Related skills
- Downloads
- 466All time
- Security audit
- Pass
- Last updated
- 2mo ago
- Current version
- v0.1.1
- License
- MIT-0
