# paper-qa - Doramagic AI Context Pack

> 定位：安装前体验与判断资产。它帮助宿主 AI 有一个好的开始，但不代表已经安装、执行或验证目标项目。

## 充分原则

- **充分原则，不是压缩原则**：AI Context Pack 应该充分到让宿主 AI 在开工前理解项目价值、能力边界、使用入口、风险和证据来源；它可以分层组织，但不以最短摘要为目标。
- **压缩策略**：只压缩噪声和重复内容，不压缩会影响判断和开工质量的上下文。

## 给宿主 AI 的使用方式

你正在读取 Doramagic 为 paper-qa 编译的 AI Context Pack。请把它当作开工前上下文：帮助用户理解适合谁、能做什么、如何开始、哪些必须安装后验证、风险在哪里。不要声称你已经安装、运行或执行了目标项目。

## Claim 消费规则

- **事实来源**：Repo Evidence + Claim/Evidence Graph；Human Wiki 只提供显著性、术语和叙事结构。
- **事实最低状态**：`supported`
- `supported`：可以作为项目事实使用，但回答中必须引用 claim_id 和证据路径。
- `weak`：只能作为低置信度线索，必须要求用户继续核实。
- `inferred`：只能用于风险提示或待确认问题，不能包装成项目事实。
- `unverified`：不得作为事实使用，应明确说证据不足。
- `contradicted`：必须展示冲突来源，不得替用户强行选择一个版本。

## 它最适合谁

- **AI 研究者或研究型 Agent 构建者**：README 明确围绕研究、实验或论文工作流展开。 证据：`README.md` Claim：`clm_0002` supported 0.86
- **正在使用 Claude/Codex/Cursor/Gemini 等宿主 AI 的开发者**：README 或插件配置提到多个宿主 AI。 证据：`README.md` Claim：`clm_0003` supported 0.86

## 它能做什么

- **命令行启动或安装流程**（需要安装后验证）：项目文档中存在可执行命令，真实使用需要在本地或宿主环境中运行这些命令。 证据：`README.md`, `packages/paper-qa-nemotron/README.md` Claim：`clm_0001` supported 0.86

## 怎么开始

- `pip install paper-qa` 证据：`README.md` Claim：`clm_0004` supported 0.86, `clm_0006` supported 0.86, `clm_0007` supported 0.86, `clm_0008` supported 0.86 等
- `curl -o my_papers/PaperQA2.pdf https://arxiv.org/pdf/2409.13740` 证据：`README.md` Claim：`clm_0005` supported 0.86
- `pip install paper-qa>=5` 证据：`README.md` Claim：`clm_0006` supported 0.86
- `pip install paper-qa[local]` 证据：`README.md` Claim：`clm_0007` supported 0.86
- `pip install paper-qa[nemotron]` 证据：`packages/paper-qa-nemotron/README.md` Claim：`clm_0008` supported 0.86, `clm_0009` supported 0.86, `clm_0010` supported 0.86
- `pip install paper-qa-nemotron` 证据：`packages/paper-qa-nemotron/README.md` Claim：`clm_0008` supported 0.86, `clm_0009` supported 0.86, `clm_0010` supported 0.86
- `pip install paper-qa-nemotron[sagemaker]` 证据：`packages/paper-qa-nemotron/README.md` Claim：`clm_0010` supported 0.86

## 继续前判断卡

- **当前建议**：需要管理员/安全审批
- **为什么**：继续前可能涉及密钥、账号、外部服务或敏感上下文，建议先经过管理员或安全审批。

### 30 秒判断

- **现在怎么做**：需要管理员/安全审批
- **最小安全下一步**：先跑 Prompt Preview；若涉及凭证或企业环境，先审批再试装
- **先别相信**：真实输出质量不能在安装前相信。
- **继续会触碰**：命令执行、本地环境或项目文件、环境变量 / API Key

### 现在可以相信

- **适合人群线索：AI 研究者或研究型 Agent 构建者**（supported）：有 supported claim 或项目证据支撑，但仍不等于真实安装效果。 证据：`README.md` Claim：`clm_0002` supported 0.86
- **适合人群线索：正在使用 Claude/Codex/Cursor/Gemini 等宿主 AI 的开发者**（supported）：有 supported claim 或项目证据支撑，但仍不等于真实安装效果。 证据：`README.md` Claim：`clm_0003` supported 0.86
- **能力存在：命令行启动或安装流程**（supported）：可以相信项目包含这类能力线索；是否适合你的具体任务仍要试用或安装后验证。 证据：`README.md`, `packages/paper-qa-nemotron/README.md` Claim：`clm_0001` supported 0.86
- **存在 Quick Start / 安装命令线索**（supported）：可以相信项目文档出现过启动或安装入口；不要因此直接在主力环境运行。 证据：`README.md` Claim：`clm_0004` supported 0.86, `clm_0006` supported 0.86, `clm_0007` supported 0.86, `clm_0008` supported 0.86

### 现在还不能相信

- **真实输出质量不能在安装前相信。**（unverified）：Prompt Preview 只能展示引导方式，不能证明真实项目中的结果质量。
- **宿主 AI 版本兼容性不能在安装前相信。**（unverified）：Claude、Cursor、Codex、Gemini 等宿主加载规则和版本差异必须在真实环境验证。
- **不会污染现有宿主 AI 行为，不能直接相信。**（inferred）：Skill、plugin、AGENTS/CLAUDE/GEMINI 指令可能改变宿主 AI 的默认行为。
- **可安全回滚不能默认相信。**（unverified）：除非项目明确提供卸载和恢复说明，否则必须先在隔离环境验证。
- **真实安装后是否与用户当前宿主 AI 版本兼容？**（unverified）：兼容性只能通过实际宿主环境验证。
- **项目输出质量是否满足用户具体任务？**（unverified）：安装前预览只能展示流程和边界，不能替代真实评测。
- **安装命令是否需要网络、权限或全局写入？**（unverified）：这影响企业环境和个人环境的安装风险。 证据：`README.md`

### 继续会触碰什么

- **命令执行**：包管理器、网络下载、本地插件目录、项目配置或用户主目录。 原因：运行第一条命令就可能产生环境改动；必须先判断是否值得跑。 证据：`README.md`, `packages/paper-qa-nemotron/README.md`
- **本地环境或项目文件**：安装结果、插件缓存、项目配置或本地依赖目录。 原因：安装前无法证明写入范围和回滚方式，需要隔离验证。 证据：`README.md`, `packages/paper-qa-nemotron/README.md`
- **环境变量 / API Key**：项目入口文档明确出现 API key、token、secret 或账号凭证配置。 原因：如果真实安装需要凭证，应先使用测试凭证并经过权限/合规判断。 证据：`README.md`, `docs/tutorials/settings_tutorial.md`, `docs/tutorials/where_do_I_get_papers.md`, `packages/paper-qa-nemotron/README.md` 等
- **宿主 AI 上下文**：AI Context Pack、Prompt Preview、Skill 路由、风险规则和项目事实。 原因：导入上下文会影响宿主 AI 后续判断，必须避免把未验证项包装成事实。

### 最小安全下一步

- **先跑 Prompt Preview**：用安装前交互式试用判断工作方式是否匹配，不需要授权或改环境。（适用：任何项目都适用，尤其是输出质量未知时。）
- **只在隔离目录或测试账号试装**：避免安装命令污染主力宿主 AI、真实项目或用户主目录。（适用：存在命令执行、插件配置或本地写入线索时。）
- **不要使用真实生产凭证**：环境变量/API key 一旦进入宿主或工具链，可能产生账号和合规风险。（适用：出现 API、TOKEN、KEY、SECRET 等环境线索时。）
- **安装后只验证一个最小任务**：先验证加载、兼容、输出质量和回滚，再决定是否深用。（适用：准备从试用进入真实工作流时。）

### 退出方式

- **保留安装前状态**：记录原始宿主配置和项目状态，后续才能判断是否可恢复。
- **记录安装命令和写入路径**：没有明确卸载说明时，至少要知道哪些目录或配置需要手动清理。
- **准备撤销测试 API key 或 token**：测试凭证泄露或误用时，可以快速止损。
- **如果没有回滚路径，不进入主力环境**：不可回滚是继续前阻断项，不应靠信任或运气继续。

## 哪些只能预览

- 解释项目适合谁和能做什么
- 基于项目文档演示典型对话流程
- 帮助用户判断是否值得安装或继续研究

## 哪些必须安装后验证

- 真实安装 Skill、插件或 CLI
- 执行脚本、修改本地文件或访问外部服务
- 验证真实输出质量、性能和兼容性

## 边界与风险判断卡

- **把安装前预览误认为真实运行**：用户可能高估项目已经完成的配置、权限和兼容性验证。 处理方式：明确区分 prompt_preview_can_do 与 runtime_required。 Claim：`clm_0011` inferred 0.45
- **命令执行会修改本地环境**：安装命令可能写入用户主目录、宿主插件目录或项目配置。 处理方式：先在隔离环境或测试账号中运行。 证据：`README.md`, `packages/paper-qa-nemotron/README.md` Claim：`clm_0012` supported 0.86
- **待确认**：真实安装后是否与用户当前宿主 AI 版本兼容？。原因：兼容性只能通过实际宿主环境验证。
- **待确认**：项目输出质量是否满足用户具体任务？。原因：安装前预览只能展示流程和边界，不能替代真实评测。
- **待确认**：安装命令是否需要网络、权限或全局写入？。原因：这影响企业环境和个人环境的安装风险。

## 开工前工作上下文

### 加载顺序

- 先读取 how_to_use.host_ai_instruction，建立安装前判断资产的边界。
- 读取 claim_graph_summary，确认事实来自 Claim/Evidence Graph，而不是 Human Wiki 叙事。
- 再读取 intended_users、capabilities 和 quick_start_candidates，判断用户是否匹配。
- 需要执行具体任务时，优先查 role_skill_index，再查 evidence_index。
- 遇到真实安装、文件修改、网络访问、性能或兼容性问题时，转入 risk_card 和 boundaries.runtime_required。

### 任务路由

- **命令行启动或安装流程**：先说明这是安装后验证能力，再给出安装前检查清单。 边界：必须真实安装或运行后验证。 证据：`README.md`, `packages/paper-qa-nemotron/README.md` Claim：`clm_0001` supported 0.86

### 上下文规模

- 文件总数：103
- 重要文件覆盖：40/103
- 证据索引条目：57
- 角色 / Skill 条目：10

### 证据不足时的处理

- **missing_evidence**：说明证据不足，要求用户提供目标文件、README 段落或安装后验证记录；不要补全事实。
- **out_of_scope_request**：说明该任务超出当前 AI Context Pack 证据范围，并建议用户先查看 Human Manual 或真实安装后验证。
- **runtime_request**：给出安装前检查清单和命令来源，但不要替用户执行命令或声称已执行。
- **source_conflict**：同时展示冲突来源，标记为待核实，不要强行选择一个版本。

## Prompt Recipes

### 适配判断

- 目标：判断这个项目是否适合用户当前任务。
- 预期输出：适配结论、关键理由、证据引用、安装前可预览内容、必须安装后验证内容、下一步建议。

```text
请基于 paper-qa 的 AI Context Pack，先问我 3 个必要问题，然后判断它是否适合我的任务。回答必须包含：适合谁、能做什么、不能做什么、是否值得安装、证据来自哪里。所有项目事实必须引用 evidence_refs、source_paths 或 claim_id。
```

### 安装前体验

- 目标：让用户在安装前感受核心工作流，同时避免把预览包装成真实能力或营销承诺。
- 预期输出：一段带边界标签的体验剧本、安装后验证清单和谨慎建议；不含真实运行承诺或强营销表述。

```text
请把 paper-qa 当作安装前体验资产，而不是已安装工具或真实运行环境。

请严格输出四段：
1. 先问我 3 个必要问题。
2. 给出一段“体验剧本”：用 [安装前可预览]、[必须安装后验证]、[证据不足] 三种标签展示它可能如何引导工作流。
3. 给出安装后验证清单：列出哪些能力只有真实安装、真实宿主加载、真实项目运行后才能确认。
4. 给出谨慎建议：只能说“值得继续研究/试装”“先补充信息后再判断”或“不建议继续”，不得替项目背书。

硬性边界：
- 不要声称已经安装、运行、执行测试、修改文件或产生真实结果。
- 不要写“自动适配”“确保通过”“完美适配”“强烈建议安装”等承诺性表达。
- 如果描述安装后的工作方式，必须使用“如果安装成功且宿主正确加载 Skill，它可能会……”这种条件句。
- 体验剧本只能写成“示例台词/假设流程”：使用“可能会询问/可能会建议/可能会展示”，不要写“已写入、已生成、已通过、正在运行、正在生成”。
- Prompt Preview 不负责给安装命令；如用户准备试装，只能提示先阅读 Quick Start 和 Risk Card，并在隔离环境验证。
- 所有项目事实必须来自 supported claim、evidence_refs 或 source_paths；inferred/unverified 只能作风险或待确认项。

```

### 角色 / Skill 选择

- 目标：从项目里的角色或 Skill 中挑选最匹配的资产。
- 预期输出：候选角色或 Skill 列表，每项包含适用场景、证据路径、风险边界和是否需要安装后验证。

```text
请读取 role_skill_index，根据我的目标任务推荐 3-5 个最相关的角色或 Skill。每个推荐都要说明适用场景、可能输出、风险边界和 evidence_refs。
```

### 风险预检

- 目标：安装或引入前识别环境、权限、规则冲突和质量风险。
- 预期输出：环境、权限、依赖、许可、宿主冲突、质量风险和未知项的检查清单。

```text
请基于 risk_card、boundaries 和 quick_start_candidates，给我一份安装前风险预检清单。不要替我执行命令，只说明我应该检查什么、为什么检查、失败会有什么影响。
```

### 宿主 AI 开工指令

- 目标：把项目上下文转成一次对话开始前的宿主 AI 指令。
- 预期输出：一段边界明确、证据引用明确、适合复制给宿主 AI 的开工前指令。

```text
请基于 paper-qa 的 AI Context Pack，生成一段我可以粘贴给宿主 AI 的开工前指令。这段指令必须遵守 not_runtime=true，不能声称项目已经安装、运行或产生真实结果。
```

## 角色 / Skill 索引

- 共索引 10 个角色 / Skill / 项目文档条目。

- **PaperQA2**（project_doc）：! GitHub https://img.shields.io/badge/GitHub-black?logo=github&logoColor=white https://github.com/Future-House/paper-qa ! PyPI version https://badge.fury.io/py/paper-qa.svg https://badge.fury.io/py/paper-qa ! tests https://github.com/Future-House/paper-qa/actions/workflows/tests.yml/badge.svg https://github.com/Future-House/paper-qa ! License https://img.shields.io/badge/License-Apache 2.0-blue.svg ! PyPI Python Ver… 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`README.md`
- **paper-qa-pypdf**（project_doc）：! GitHub https://img.shields.io/badge/GitHub-black?logo=github&logoColor=white https://github.com/Future-House/paper-qa/tree/main/packages/paper-qa-docling ! PyPI version https://badge.fury.io/py/paper-qa-docling.svg https://badge.fury.io/py/paper-qa-docling ! tests https://github.com/Future-House/paper-qa/actions/workflows/tests.yml/badge.svg https://github.com/Future-House/paper-qa ! License https://img.shields.io… 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`packages/paper-qa-docling/README.md`
- **paper-qa-nemotron**（project_doc）：! GitHub https://img.shields.io/badge/GitHub-black?logo=github&logoColor=white https://github.com/Future-House/paper-qa/tree/main/packages/paper-qa-nemotron ! tests https://github.com/Future-House/paper-qa/actions/workflows/tests.yml/badge.svg https://github.com/Future-House/paper-qa ! License https://img.shields.io/badge/License-Apache 2.0-blue.svg ! PyPI Python Versions https://img.shields.io/pypi/pyversions/paper… 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`packages/paper-qa-nemotron/README.md`
- **paper-qa-pymupdf**（project_doc）：! GitHub https://img.shields.io/badge/GitHub-black?logo=github&logoColor=white https://github.com/Future-House/paper-qa/tree/main/packages/paper-qa-pymupdf ! PyPI version https://badge.fury.io/py/paper-qa-pymupdf.svg https://badge.fury.io/py/paper-qa-pymupdf ! tests https://github.com/Future-House/paper-qa/actions/workflows/tests.yml/badge.svg https://github.com/Future-House/paper-qa ! License https://img.shields.io… 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`packages/paper-qa-pymupdf/README.md`
- **paper-qa-pypdf**（project_doc）：! GitHub https://img.shields.io/badge/GitHub-black?logo=github&logoColor=white https://github.com/Future-House/paper-qa/tree/main/packages/paper-qa-pypdf ! PyPI version https://badge.fury.io/py/paper-qa-pypdf.svg https://badge.fury.io/py/paper-qa-pypdf ! tests https://github.com/Future-House/paper-qa/actions/workflows/tests.yml/badge.svg https://github.com/Future-House/paper-qa ! License https://img.shields.io/badge… 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`packages/paper-qa-pypdf/README.md`
- **Contributing to PaperQA**（project_doc）：Thank you for your interest in contributing to PaperQA! Here are some guidelines to help you get started. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`CONTRIBUTING.md`
- **PaperQA2 for Clinical Trials**（project_doc）：PaperQA2 now natively supports querying clinical trials in addition to any documents supplied by the user. It uses a new tool, the aptly named clinical trials search tool. Users don't have to provide any clinical trials to the tool itself, it uses the clinicaltrials.gov API to retrieve them on the fly. As of January 2025, the tool is not enabled by default, but it's easy to configure. Here's an example where we quer… 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/tutorials/querying_with_clinical_trials.md`
- **python3**（project_doc）：Measuring PaperQA2 with LFRQA This tutorial is available as a Jupyter notebook here https://github.com/Future-House/paper-qa/blob/main/docs/tutorials/running on lfrqa.md 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/tutorials/running_on_lfrqa.md`
- **python3**（project_doc）：This tutorial is available as a Jupyter notebook here https://github.com/Future-House/paper-qa/blob/main/docs/tutorials/settings tutorial.ipynb . 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/tutorials/settings_tutorial.md`
- **Where to get papers**（project_doc）：You can use papers from https://openreview.net/ https://openreview.net/ as your database! Here's a helper that fetches a list of all papers from a selected conference like ICLR, ICML, NeurIPS , queries this list to find relevant papers using LLM, and downloads those relevant papers to a local directory which can be used with paper-qa on the next step. Install openreview-py with 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/tutorials/where_do_I_get_papers.md`

## 证据索引

- 共索引 57 条证据。

- **PaperQA2**（documentation）：! GitHub https://img.shields.io/badge/GitHub-black?logo=github&logoColor=white https://github.com/Future-House/paper-qa ! PyPI version https://badge.fury.io/py/paper-qa.svg https://badge.fury.io/py/paper-qa ! tests https://github.com/Future-House/paper-qa/actions/workflows/tests.yml/badge.svg https://github.com/Future-House/paper-qa ! License https://img.shields.io/badge/License-Apache 2.0-blue.svg ! PyPI Python Versions https://img.shields.io/pypi/pyversions/paper-qa 证据：`README.md`
- **paper-qa-pypdf**（documentation）：! GitHub https://img.shields.io/badge/GitHub-black?logo=github&logoColor=white https://github.com/Future-House/paper-qa/tree/main/packages/paper-qa-docling ! PyPI version https://badge.fury.io/py/paper-qa-docling.svg https://badge.fury.io/py/paper-qa-docling ! tests https://github.com/Future-House/paper-qa/actions/workflows/tests.yml/badge.svg https://github.com/Future-House/paper-qa ! License https://img.shields.io/badge/License-Apache 2.0-blue.svg ! PyPI Python Versions https://img.shields.io/pypi/pyversions/paper-qa-docling 证据：`packages/paper-qa-docling/README.md`
- **paper-qa-nemotron**（documentation）：! GitHub https://img.shields.io/badge/GitHub-black?logo=github&logoColor=white https://github.com/Future-House/paper-qa/tree/main/packages/paper-qa-nemotron ! tests https://github.com/Future-House/paper-qa/actions/workflows/tests.yml/badge.svg https://github.com/Future-House/paper-qa ! License https://img.shields.io/badge/License-Apache 2.0-blue.svg ! PyPI Python Versions https://img.shields.io/pypi/pyversions/paper-qa-nemotron 证据：`packages/paper-qa-nemotron/README.md`
- **paper-qa-pymupdf**（documentation）：! GitHub https://img.shields.io/badge/GitHub-black?logo=github&logoColor=white https://github.com/Future-House/paper-qa/tree/main/packages/paper-qa-pymupdf ! PyPI version https://badge.fury.io/py/paper-qa-pymupdf.svg https://badge.fury.io/py/paper-qa-pymupdf ! tests https://github.com/Future-House/paper-qa/actions/workflows/tests.yml/badge.svg https://github.com/Future-House/paper-qa ! License https://img.shields.io/badge/license-AGPLv3-blue.svg ! PyPI Python Versions https://img.shields.io/pypi/pyversions/paper-qa-pymupdf 证据：`packages/paper-qa-pymupdf/README.md`
- **paper-qa-pypdf**（documentation）：! GitHub https://img.shields.io/badge/GitHub-black?logo=github&logoColor=white https://github.com/Future-House/paper-qa/tree/main/packages/paper-qa-pypdf ! PyPI version https://badge.fury.io/py/paper-qa-pypdf.svg https://badge.fury.io/py/paper-qa-pypdf ! tests https://github.com/Future-House/paper-qa/actions/workflows/tests.yml/badge.svg https://github.com/Future-House/paper-qa ! License https://img.shields.io/badge/License-Apache 2.0-blue.svg ! PyPI Python Versions https://img.shields.io/pypi/pyversions/paper-qa-pypdf 证据：`packages/paper-qa-pypdf/README.md`
- **Contributing to PaperQA**（documentation）：Thank you for your interest in contributing to PaperQA! Here are some guidelines to help you get started. 证据：`CONTRIBUTING.md`
- **License**（source_file）：Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ 证据：`LICENSE`
- **License**（source_file）：Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ 证据：`packages/paper-qa-docling/LICENSE`
- **License**（source_file）：Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ 证据：`packages/paper-qa-nemotron/LICENSE`
- **License**（source_file）：GNU AFFERO GENERAL PUBLIC LICENSE Version 3, 19 November 2007 证据：`packages/paper-qa-pymupdf/LICENSE`
- **License**（source_file）：Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ 证据：`packages/paper-qa-pypdf/LICENSE`
- **PaperQA2 for Clinical Trials**（documentation）：PaperQA2 now natively supports querying clinical trials in addition to any documents supplied by the user. It uses a new tool, the aptly named clinical trials search tool. Users don't have to provide any clinical trials to the tool itself, it uses the clinicaltrials.gov API to retrieve them on the fly. As of January 2025, the tool is not enabled by default, but it's easy to configure. Here's an example where we query only clinical trials, without using any documents: 证据：`docs/tutorials/querying_with_clinical_trials.md`
- **Measuring PaperQA2 with LFRQA**（documentation）：Measuring PaperQA2 with LFRQA This tutorial is available as a Jupyter notebook here https://github.com/Future-House/paper-qa/blob/main/docs/tutorials/running on lfrqa.md 证据：`docs/tutorials/running_on_lfrqa.md`
- **Setup**（documentation）：This tutorial is available as a Jupyter notebook here https://github.com/Future-House/paper-qa/blob/main/docs/tutorials/settings tutorial.ipynb . 证据：`docs/tutorials/settings_tutorial.md`
- **Where to get papers**（documentation）：You can use papers from https://openreview.net/ https://openreview.net/ as your database! Here's a helper that fetches a list of all papers from a selected conference like ICLR, ICML, NeurIPS , queries this list to find relevant papers using LLM, and downloads those relevant papers to a local directory which can be used with paper-qa on the next step. Install openreview-py with 证据：`docs/tutorials/where_do_I_get_papers.md`
- **Contracrow**（structured_config）：{ "llm": "claude-3-5-sonnet-20240620", "llm config": null, "summary llm": "claude-3-5-sonnet-20240620", "summary llm config": null, "embedding": "hybrid-text-embedding-3-large", "embedding config": null, "temperature": 0.0, "batch size": 1, "texts index mmr lambda": 1.0, "verbosity": 0, "answer": { "evidence k": 30, "evidence retrieval": true, "evidence summary length": "about 300 words", "evidence skip summary": false, "answer max sources": 15, "answer length": "about 200 words, but can be longer", "max concurrent requests": 4, "answer filter extra background": false }, "parsing": { "use doc details": true, "reader config": { "chunk chars": 7000, "overlap": 250 }, "citation prompt": "Provi… 证据：`src/paperqa/configs/contracrow.json`
- **Debug**（structured_config）：{ "llm": "claude-3-haiku-20240307", "summary llm": "claude-3-haiku-20240307", "answer": { "evidence k": 2, "evidence summary length": "25 to 50 words", "answer max sources": 2, "answer length": "50 to 100 words", "max concurrent requests": 5 }, "parsing": { "use doc details": false, "defer embedding": true }, "prompts": { "use json": false, "context inner": "{name}: {text}" } } 证据：`src/paperqa/configs/debug.json`
- **Tier2 Limits**（structured_config）：{ "answer": { "evidence k": 8, "answer max sources": 3, "max concurrent requests": 8 }, "parsing": { "use doc details": true, "reader config": { "chunk chars": 7000, "overlap": 250 } }, "prompts": { "use json": true }, "llm config": { "rate limit": { "gpt-4o": "450000 per 1 minute", "gpt-4o-2024-08-06": "450000 per 1 minute", "gpt-4o-2024-05-13": "450000 per 1 minute", "gpt-4o-mini": "2000000 per 1 minute", "gpt-4o-mini-2024-07-18": "2000000 per 1 minute", "gpt-4-turbo": "450000 per 1 minute", "gpt-4-turbo-2024-04-09": "450000 per 1 minute", "gpt-4-0613": "40000 per 1 minute", "gpt-4-0314": "40000 per 1 minute", "gpt-4": "40000 per 1 minute", "gpt-3.5-turbo-0125": "2000000 per 1 minute", "g… 证据：`src/paperqa/configs/tier2_limits.json`
- **Tier4 Limits**（structured_config）：{ "answer": { "evidence k": 10, "answer max sources": 5, "max concurrent requests": 8 }, "parsing": { "use doc details": true, "reader config": { "chunk chars": 7000, "overlap": 250 } }, "prompts": { "use json": true }, "llm config": { "rate limit": { "gpt-4o": "2000000 per 1 minute", "gpt-4o-2024-08-06": "2000000 per 1 minute", "gpt-4o-2024-05-13": "2000000 per 1 minute", "gpt-4o-mini": "10000000 per 1 minute", "gpt-4o-mini-2024-07-18": "10000000 per 1 minute", "gpt-4-turbo": "800000 per 1 minute", "gpt-4-turbo-2024-04-09": "800000 per 1 minute", "gpt-4-0613": "300000 per 1 minute", "gpt-4-0314": "300000 per 1 minute", "gpt-4": "300000 per 1 minute", "gpt-3.5-turbo-0125": "10000000 per 1 m… 证据：`src/paperqa/configs/tier4_limits.json`
- **Tier5 Limits**（structured_config）：{ "answer": { "evidence k": 15, "answer max sources": 5, "max concurrent requests": 8 }, "parsing": { "use doc details": true, "reader config": { "chunk chars": 7000, "overlap": 250 } }, "prompts": { "use json": true }, "llm config": { "rate limit": { "gpt-4o": "30000000 per 1 minute", "gpt-4o-2024-08-06": "30000000 per 1 minute", "gpt-4o-2024-05-13": "30000000 per 1 minute", "gpt-4o-mini": "150000000 per 1 minute", "gpt-4o-mini-2024-07-18": "150000000 per 1 minute", "gpt-4-turbo": "2000000 per 1 minute", "gpt-4-turbo-2024-04-09": "2000000 per 1 minute", "gpt-4-0613": "1000000 per 1 minute", "gpt-4-0314": "1000000 per 1 minute", "gpt-4": "1000000 per 1 minute", "gpt-3.5-turbo-0125": "500000… 证据：`src/paperqa/configs/tier5_limits.json`
- **Wikicrow**（structured_config）：{ "llm": "gpt-4-turbo-2024-04-09", "llm config": null, "summary llm": "gpt-4-turbo-2024-04-09", "summary llm config": null, "embedding": "hybrid-text-embedding-3-small", "embedding config": null, "temperature": 0.0, "batch size": 1, "texts index mmr lambda": 1.0, "verbosity": 0, "answer": { "evidence k": 25, "evidence retrieval": true, "evidence summary length": "about 300 words", "evidence skip summary": false, "answer max sources": 12, "answer length": "about 200 words, but can be longer", "max concurrent requests": 4, "answer filter extra background": false }, "parsing": { "use doc details": true, "reader config": { "chunk chars": 7000, "overlap": 1750 }, "citation prompt": "Provide the… 证据：`src/paperqa/configs/wikicrow.json`
- **Init**（source_file）：all = "parse pdf to pages" 证据：`packages/paper-qa-docling/src/paperqa_docling/__init__.py`
- **NOTE: the list value here is a two-item list of page text, page media.**（source_file）：DOCLING VERSION = version docling. name DOCLING IMAGES SCALE PER DPI = ⋮---- path = Path path ⋮---- pipeline options = PdfPipelineOptions ⋮---- pipeline options = PdfPipelineOptions custom pipeline options or {} converter = DocumentConverter ⋮---- result = converter.convert ⋮---- doc = result.document NOTE: the list value here is a two-item list of page text, page media. It's mutable so we can append text and media as found content: dict str, list = collections.defaultdict lambda: "", total length = count media = 0 ⋮---- NOTE: docling pages are 1-indexed page nums = prov.page no for prov in item.prov if isinstance item, TextItem FormulaItem : Handle items with text item text = item.text ⋮--… 证据：`packages/paper-qa-docling/src/paperqa_docling/reader.py`
- **Init**（source_file）：all = "parse pdf to pages" 证据：`packages/paper-qa-nemotron/src/paperqa_nemotron/__init__.py`
- **Explicitly specify OpenAI-compatible provider over Nvidia NIM provider,**（source_file）：get session = None ⋮---- logger = logging.getLogger name NVIDIA API NEMOTRON PARSE RATE LIMIT = class NemotronLengthError ValueError ⋮---- r""" Error for nemotron-parse running out of context, indicated by the 'length' finish reason. This 'length' finish reason comes from the Nvidia NIM wrapping nemotron-parse version 1.1 when the model starts babbling e.g. repeating '\\n' . It's been seen with the markdown bbox tool on large figures. Retrying is a possible method to skirt this error, but it's a bad idea as a 'length' finish reason means nemotron-parse ran out of context, and retrying until success just provides a flawed output. """ class NemotronBBoxError ValueError NemotronParseToolName:… 证据：`packages/paper-qa-nemotron/src/paperqa_nemotron/api.py`
- **Apply white border padding to increase bounding box reliability**（source_file）：logger = logging.getLogger name WHITE RGB = 255, 255, 255 DEFAULT BORDER SIZE = 60 ⋮---- canvas = Image.new ⋮---- pdf doc = pdfium.PdfDocument path ⋮---- page = pdf doc page num ⋮---- render kwargs: dict str, Any = {} ⋮---- rendered page = page.render render kwargs rendered page pil = rendered page.to pil ⋮---- Apply white border padding to increase bounding box reliability ⋮---- image data uri = encode image to base64 padded pil, format="PNG" ⋮---- image data uri = encode image to base64 rendered page pil, format="PNG" offset x = offset y = padded height = padded width = 0 ⋮---- """Parse a PDF using Nvidia's nemotron-parse VLM. Args: path: Path to the PDF file to parse. page size limit: Se… 证据：`packages/paper-qa-nemotron/src/paperqa_nemotron/reader.py`
- **Init**（source_file）：all = 证据：`packages/paper-qa-pymupdf/src/paperqa_pymupdf/__init__.py`
- **Reader**（source_file）：def setup pymupdf python logging - None BLOCK TEXT INDEX = 4 PYMUPDF PIXMAP ATTRS = { ⋮---- page = file.load page page num ⋮---- blocks = page.get text "blocks", sort=False text = "\n".join ⋮---- text = page.get text "text", sort=True ⋮---- """Worker function for parallel full-page screenshot parsing. NOTE: must be top-level for pickling. """ ⋮---- pix = page.get pixmap dpi=dpi media metadata: dict str, JsonValue = {"type": "screenshot"} { ⋮---- media = ParsedMedia index=0, data=pix.tobytes , info=media metadata ⋮---- content: dict str, str tuple str, list ParsedMedia = {} total length = count media = 0 ⋮---- page iter = resolve page range page range, file.page count path str = str path arg… 证据：`packages/paper-qa-pymupdf/src/paperqa_pymupdf/reader.py`
- **Init**（source_file）：all = 证据：`packages/paper-qa-pypdf/src/paperqa_pypdf/__init__.py`
- **On 12/30/2025 with pypdf==6.4.2, a PageObject.extract text call on**（source_file）：pdfium = None ⋮---- pdfplumber = None ⋮---- @unique class MediaMode StrEnum ⋮---- NONE = "" No media extraction FULL PAGE = "full-page" INDIVIDUAL CLUSTERING = INDIVIDUAL = "individual" def str self - str ⋮---- @property def metadata value self - str PDFIUM BITMAP ATTRS = {"width", "height", "stride", "n channels", "mode"} SCALE TO DPI = 72 ⋮---- render kwargs = {} ⋮---- pdf reader = pypdf.PdfReader file ⋮---- pages: dict str, str tuple str, list ParsedMedia = {} total length = count media = 0 ⋮---- media mode = MediaMode.NONE ⋮---- media mode = MediaMode.FULL PAGE ⋮---- media mode = MediaMode.INDIVIDUAL CLUSTERING ⋮---- media mode = MediaMode.INDIVIDUAL ⋮---- pdf doc = pdfium.PdfDocument s… 证据：`packages/paper-qa-pypdf/src/paperqa_pypdf/reader.py`
- **Init**（source_file）：all = 证据：`src/paperqa/__init__.py`
- **Init**（source_file）：logger = logging.getLogger name LOG VERBOSITY MAP: dict int, dict str, int = { ⋮---- MAX PRESET VERBOSITY: int = max k for k in LOG VERBOSITY MAP PAPERQA PKG ROOT LOGGER = logging.getLogger name .split ".", maxsplit=1 0 INITIATED FROM CLI = False def is running under cli - bool def set up rich handler install: bool = True - RichHandler ⋮---- rich handler = RichHandler ⋮---- def configure log verbosity verbosity: int = 0 - None ⋮---- key = min verbosity, MAX PRESET VERBOSITY ⋮---- def configure cli logging verbosity: int Settings = 0 - None ⋮---- verbosity = verbosity.verbosity ⋮---- """Query PaperQA via an agent.""" ⋮---- """Search using a pre-built PaperQA index.""" ⋮---- index name = sett… 证据：`src/paperqa/agents/__init__.py`
- **SEE: https://regex101.com/r/L0L5MH/1**（source_file）：logger = logging.getLogger name POPULATE FROM SETTINGS = None ⋮---- llm model = llm model or settings.get llm summary llm model = summary llm model or settings.get summary llm embedding model = embedding model or settings.get embedding model tools: list Tool = ⋮---- def make tool fn: Callable, tool type: type NamedTool = tool type - Tool ⋮---- tool = make tool ⋮---- gather evidence tool = GatherEvidence ⋮---- tool = make tool gather evidence tool.gather evidence ⋮---- generate answer tool = GenerateAnswer ⋮---- tool = make tool generate answer tool.gen answer ⋮---- tool = make tool Reset .reset ⋮---- tool = make tool Complete .complete ⋮---- tools.append tool Place at the end ⋮---- SEE: htt… 证据：`src/paperqa/agents/env.py`
- **partial formatting**（source_file）：logger = logging.getLogger name def get year ts: datetime None = None - str ⋮---- ts = datetime.now ⋮---- search prompt = "" ⋮---- partial formatting search prompt = template.replace "{date}", get year ⋮---- TODO: move to use tools instead of DIY schema in prompt search prompt = ⋮---- model: LLMModel = LiteLLMModel name=llm ⋮---- model = llm messages = result = await model.call single search query = cast "str", result.text queries = s for s in search query.split "\n" if len s 3 queries = re.sub r"^\d+\.\s ", "", q for q in queries remove quotes ⋮---- table = Table title="Prior Answers" ⋮---- table = Table title="PDF Search" ⋮---- docs = cast "Docs", obj doc = docs.texts 0 .doc ⋮---- display… 证据：`src/paperqa/agents/helpers.py`
- **Models**（source_file）：logger = logging.getLogger name class SupportsPickle Protocol ⋮---- def reduce self - str tuple Any, ... : ... def getstate self - object: ... def setstate self, state: object - None: ... class AgentStatus StrEnum ⋮---- FAIL = "fail" SUCCESS = "success" TRUNCATED = "truncated" UNSURE = "unsure" class MismatchedModelsError Exception ⋮---- LOG METHOD NAME: ClassVar str = "warning" class AnswerResponse BaseModel ⋮---- model config = ConfigDict populate by name=True session: PQASession = Field alias="answer" bibtex: dict str, str None = None status: AgentStatus timing info: dict str, dict str, float None = None duration: float = 0.0 stats: dict str, str None = None ⋮---- async def get summary s… 证据：`src/paperqa/agents/models.py`
- **Search**（source_file）：logger = logging.getLogger name class AsyncRetryError Exception class SearchDocumentStorage StrEnum ⋮---- JSON MODEL DUMP = auto PICKLE COMPRESSED = auto PICKLE UNCOMPRESSED = auto def extension self - str def write to string self, data: BaseModel SupportsPickle - bytes ⋮---- OPENED INDEX CACHE: dict tuple str, str , tuple Index, int = {} DONT USE OPENED INDEX CACHE = def reap opened index cache - None ⋮---- """Delete any unreferenced Index instances from the Index cache.""" ⋮---- class SearchIndex ⋮---- """Wrapper around a tantivy.Index exposing higher-level behaviors for documents.""" REQUIRED FIELDS: ClassVar list str = "file location", "body" ⋮---- fields = self.REQUIRED FIELDS ⋮---- if… 证据：`src/paperqa/agents/search.py`
- **first see if we are nested; i.e. we want order**（source_file）：logger = logging.getLogger name DEFAULT CLIENTS: Collection type MetadataPostProcessor MetadataProvider = ALL CLIENTS: Collection type MetadataPostProcessor MetadataProvider = MetadataClientQuerier: TypeAlias = class DocMetadataTask BaseModel ⋮---- model config = ConfigDict arbitrary types allowed=True providers: Collection MetadataProvider = Field processors: Collection MetadataPostProcessor = Field ⋮---- def repr self - str class DocMetadataClient ⋮---- """Metadata client for querying multiple metadata providers and processors. Args: http client: Async HTTP client to allow for connection pooling. metadata clients: list of MetadataProvider and MetadataPostProcessor instances or classes to… 证据：`src/paperqa/clients/__init__.py`
- **we're suppressing this error to not fail on 403 or 500 errors from providers**（source_file）：logger = logging.getLogger name class ClientQuery BaseModel ⋮---- model config = ConfigDict arbitrary types allowed=True client: httpx.AsyncClient class TitleAuthorQuery ClientQuery ⋮---- title: str authors: list str = Field default factory=list title similarity threshold: float = 0.75 fields: Collection str None = None ⋮---- @model validator mode="before" @classmethod def ensure fields are present cls, data: dict str, Any - dict str, Any ⋮---- @field validator "title similarity threshold" @classmethod def zero and one cls, v: float, info: ValidationInfo - float class DOIQuery ClientQuery ⋮---- doi: str ⋮---- @model validator mode="before" @classmethod def add doi to fields and validate cls… 证据：`src/paperqa/clients/client_models.py`
- **Init**（source_file）：all = "ZoteroDB" 证据：`src/paperqa/contrib/__init__.py`
- **Openreview Paper Helper**（source_file）：openreview = None logger = logging.getLogger name class PaperSuggestion BaseModel ⋮---- submission id: str = Field description="The ID of the submission" explanation: str = Field description="Reasoning for why this paper is relevant" class RelevantPapersResponse BaseModel ⋮---- suggested papers: list PaperSuggestion = Field reasoning step by step: str = Field RELEVANT PAPERS SCHEMA = RelevantPapersResponse.model json schema class OpenReviewPaperHelper ⋮---- def get venues self - list str def get submissions self - list Any def create submission string self, submissions: list Any - str ⋮---- submission info string = "" ⋮---- paper = { ⋮---- async def fetch relevant papers self, question: str… 证据：`src/paperqa/contrib/openreview_paper_helper.py`
- **Filter:**（source_file）：class ZoteroPaper BaseModel ⋮---- key: str title: str pdf: Path num pages: int zotero key: str details: dict def str self - str class ZoteroDB zotero.Zotero ⋮---- """An extension of pyzotero.zotero.Zotero to interface with paperqa. This class automatically reads in your ZOTERO USER ID and ZOTERO API KEY from your environment variables. If you do not have these, see step 2 of https://github.com/urschrei/pyzotero quickstart. This class will download PDFs from your Zotero library and store them in ~/.paperqa/zotero by default. To use this class, call the iterate method, which returns a paperqa.Docs object. """ ⋮---- library id = os.environ "ZOTERO USER ID" ⋮---- api key = os.environ "ZOTERO AP… 证据：`src/paperqa/contrib/zotero.py`
- **NVIDIA's reference orders bbox as xmin, ymin, xmax, ymax**（source_file）：REPO ROOT = Path file .parents 3 STUB DATA DIR = REPO ROOT / "tests" / "stub data" class TestNemotronParseBBox ⋮---- def test bbox validation self - None ⋮---- bbox = NemotronParseBBox xmin=0.1, xmax=0.9, ymin=0.2, ymax=0.8 ⋮---- bbox full = NemotronParseBBox xmin=0.0, xmax=1.0, ymin=0.0, ymax=1.0 ⋮---- def test bbox to page coordinates self - None def test bbox to original coordinates self, subtests: pytest.Subtests - None ⋮---- aspect ratio = original width / original height new height = original height new width = original width ⋮---- new height = target h new width = int new height aspect ratio ⋮---- new width = target w new height = int new width / aspect ratio resized width = new widt… 证据：`packages/paper-qa-nemotron/tests/test_api.py`
- **Clinical Trials**（structured_config）：{ "answer": { "evidence k": 15, "answer max sources": 5, "max concurrent requests": 10 }, "agent": { "tool names": "gather evidence", "paper search", "gen answer", "clinical trials search", "complete" }, "parsing": { "use doc details": true, "reader config": { "chunk chars": 9000, "overlap": 750 } } } 证据：`src/paperqa/configs/clinical_trials.json`
- **Fast**（structured_config）：{ "answer": { "evidence k": 5, "evidence summary length": "25 to 50 words", "answer max sources": 3, "answer length": "50 to 100 words", "max concurrent requests": 5 }, "parsing": { "use doc details": false }, "prompts": { "use json": false, "context inner": "{name}: {text}" }, "agent": { "agent type": "fake" } } 证据：`src/paperqa/configs/fast.json`
- **High Quality**（structured_config）：{ "answer": { "evidence k": 20, "answer max sources": 5, "max concurrent requests": 10 }, "parsing": { "use doc details": true, "reader config": { "chunk chars": 7000, "overlap": 250 } } } 证据：`src/paperqa/configs/high_quality.json`
- **Openreview**（structured_config）：{ "llm": "gemini/gemini-2.0-flash-exp", "llm config": { "model name": "gemini/gemini-2.0-flash-exp", "litellm params": { "model": "gemini/gemini-2.0-flash-exp", "api key": null } }, "summary llm": "gemini/gemini-2.0-flash-exp", "summary llm config": { "model name": "gemini/gemini-2.0-flash-exp", "litellm params": { "model": "gemini/gemini-2.0-flash-exp", "api key": null } }, "embedding": "ollama/granite3-dense", "paper directory": "my papers", "verbosity": 3, "agent": { "agent llm": "gemini/gemini-2.0-flash-exp", "agent llm config": { "model name": "gemini/gemini-2.0-flash-exp", "litellm params": { "model": "gemini/gemini-2.0-flash-exp", "api key": null } }, "return paper metadata": false }… 证据：`src/paperqa/configs/openreview.json`
- **Search Only Clinical Trials**（structured_config）：{ "answer": { "evidence k": 15, "answer max sources": 5, "max concurrent requests": 10 }, "agent": { "tool names": "gather evidence", "gen answer", "clinical trials search", "complete" }, "parsing": { "use doc details": true, "reader config": { "chunk chars": 9000, "overlap": 750 } } } 证据：`src/paperqa/configs/search_only_clinical_trials.json`
- **Tier1 Limits**（structured_config）：{ "answer": { "evidence k": 5, "evidence summary length": "25 to 50 words", "answer max sources": 3, "answer length": "50 to 100 words", "max concurrent requests": 5 }, "parsing": { "use doc details": false }, "prompts": { "use json": true, "context inner": "{name}: {text}" }, "llm config": { "rate limit": { "gpt-4o": "30000 per 1 minute", "gpt-4o-2024-08-06": "30000 per 1 minute", "gpt-4o-2024-05-13": "30000 per 1 minute", "gpt-4o-mini": "200000 per 1 minute", "gpt-4o-mini-2024-07-18": "200000 per 1 minute", "gpt-4-turbo": "30000 per 1 minute", "gpt-4-turbo-2024-04-09": "30000 per 1 minute", "gpt-4-0613": "10000 per 1 minute", "gpt-4-0314": "10000 per 1 minute", "gpt-4": "10000 per 1 minut… 证据：`src/paperqa/configs/tier1_limits.json`
- **Tier3 Limits**（structured_config）：{ "answer": { "evidence k": 8, "answer max sources": 3, "max concurrent requests": 8 }, "parsing": { "use doc details": true, "reader config": { "chunk chars": 7000, "overlap": 250 } }, "prompts": { "use json": true }, "llm config": { "rate limit": { "gpt-4o": "800000 per 1 minute", "gpt-4o-2024-08-06": "800000 per 1 minute", "gpt-4o-2024-05-13": "800000 per 1 minute", "gpt-4o-mini": "4000000 per 1 minute", "gpt-4o-mini-2024-07-18": "4000000 per 1 minute", "gpt-4-turbo": "600000 per 1 minute", "gpt-4-turbo-2024-04-09": "600000 per 1 minute", "gpt-4-0613": "80000 per 1 minute", "gpt-4-0314": "80000 per 1 minute", "gpt-4": "80000 per 1 minute", "gpt-3.5-turbo-0125": "4000000 per 1 minute", "g… 证据：`src/paperqa/configs/tier3_limits.json`
- **.gitattributes**（source_file）：tests/cassettes/ linguist-generated=true 证据：`.gitattributes`
- **Swap**（source_file）：Swap . .s a-v a-z ! .svg comment out if you don't need vector files . .sw a-p . s a-rt-v a-z . ss a-gi-z . sw a-p 证据：`.gitignore`
- **.mailmap**（source_file）：Andrew White Ahmet Celebi At4 Anush008 Anush Dmitrii Magas eamag Geemi Wellawatte Geemi Wellawatte Harry Vu Harry Vu harryvu-futurehouse James Braza Mayk Caldas maykcaldas Mayk Caldas Michael Skarlinski mskarlin Odhran O'Donoghue odhran-o-d Odhran O'Donoghue Samantha Cox takeru fukushima 证据：`.mailmap`
- **.Pre Commit Config**（source_file）：default language version: python: python3 repos: - repo: https://github.com/pre-commit/pre-commit-hooks rev: v6.0.0 hooks: - id: check-added-large-files exclude: ?x ^ packages/paper-qa-nemotron/tests/cassettes. src/paperqa/clients/client data. tests/stub data. tests/cassettes. uv\.lock $ - id: check-case-conflict - id: check-merge-conflict - id: check-shebang-scripts-are-executable - id: check-symlinks - id: check-toml - id: check-yaml - id: debug-statements - id: detect-private-key - id: end-of-file-fixer - id: fix-byte-order-marker - id: mixed-line-ending - id: trailing-whitespace - repo: https://github.com/astral-sh/ruff-pre-commit rev: v0.14.4 hooks: - id: ruff-check args: --fix, --exit… 证据：`.pre-commit-config.yaml`
- **.python-version**（source_file）：3.13 证据：`.python-version`
- **Citation**（source_file）：--- cff-version: 1.2.0 message: - If you use this software, please cite it using the metadata from this file. authors: - family-names: Skarlinski given-names: Michael D. - family-names: Cox given-names: Sam - family-names: Laurent given-names: Jon M. - family-names: Braza given-names: James D. - family-names: Hinks given-names: Michaela - family-names: Hammerling given-names: Michael J. - family-names: Ponnapati given-names: Manvitha - family-names: Rodriques given-names: Samuel G. - family-names: White given-names: Andrew D. title: "Language agents achieve superhuman synthesis of scientific knowledge" identifiers: - type: doi value: 10.48550/arXiv.2409.13740 description: ArXiv DOI - type:… 证据：`CITATION.cff`
- **2024 10 16 Litqa2 Splits**（source_file）：// Train, evaluation, and test DOIs here were generated from historical data // aggregated across many runs made while writing DOI 10.48550/arXiv.2409.13740 { train: { question ids: "04dbe07d-8b2c-4daf-b5b2-ef0e93f1fd2a", "0708b62f-9652-49eb-8ba6-28878afa7445", "0a9d6516-95ef-4d7b-a28d-d7cde27b7b55", "0bac8974-554c-439a-a9a2-22fa509c8d5d", "0d5cf8a7-a240-4a8f-be4e-c16712f90d79", "0eeb7ea9-fc80-4dee-9418-1c328c3ab653", "0eede7a8-fe1f-42d3-a2c6-478083648644", "10cece36-a507-4a93-9600-13f3e0e677f8", "12a20d8d-cd49-47eb-9a19-6a38519ee3dc", "14fd2b75-76fb-4c29-a21d-c557b2bcf2ff", "178a5e56-340f-4ba8-a3e5-f024ca016f40", "1e5f5199-84f4-4133-ab87-2372fa6ca722", "1f1b07d7-39ce-4665-9b70-4ab77e3c87aa… 证据：`docs/2024-10-16_litqa2-splits.json5`
- **Full list: https://pypi.python.org/pypi?%3Aaction=list classifiers**（source_file）：build-system build-backend = "setuptools.build meta" requires = "setuptools =64", "setuptools scm =8" 证据：`pyproject.toml`

## 宿主 AI 必须遵守的规则

- **把本资产当作开工前上下文，而不是运行环境。**：AI Context Pack 只包含证据化项目理解，不包含目标项目的可执行状态。 证据：`README.md`, `packages/paper-qa-docling/README.md`, `packages/paper-qa-nemotron/README.md`
- **回答用户时区分可预览内容与必须安装后才能验证的内容。**：安装前体验的消费者价值来自降低误装和误判，而不是伪装成真实运行。 证据：`README.md`, `packages/paper-qa-docling/README.md`, `packages/paper-qa-nemotron/README.md`

## 用户开工前应该回答的问题

- 你准备在哪个宿主 AI 或本地环境中使用它？
- 你只是想先体验工作流，还是准备真实安装？
- 你最在意的是安装成本、输出质量、还是和现有规则的冲突？

## 验收标准

- 所有能力声明都能回指到 evidence_refs 中的文件路径。
- AI_CONTEXT_PACK.md 没有把预览包装成真实运行。
- 用户能在 3 分钟内看懂适合谁、能做什么、如何开始和风险边界。

---

## Doramagic Context Augmentation

下面内容用于强化 Repomix/AI Context Pack 主体。Human Manual 只提供阅读骨架；踩坑日志会被转成宿主 AI 必须遵守的工作约束。

## Human Manual 骨架

使用规则：这里只是项目阅读路线和显著性信号，不是事实权威。具体事实仍必须回到 repo evidence / Claim Graph。

宿主 AI 硬性规则：
- 不得把页标题、章节顺序、摘要或 importance 当作项目事实证据。
- 解释 Human Manual 骨架时，必须明确说它只是阅读路线/显著性信号。
- 能力、安装、兼容性、运行状态和风险判断必须引用 repo evidence、source path 或 Claim Graph。

- **PaperQA2 概览与快速开始**：importance `high`
  - source_paths: README.md, src/paperqa/__init__.py, src/paperqa/configs/high_quality.json, src/paperqa/configs/fast.json, src/paperqa/paths.py
- **核心架构、设置系统与数据流**：importance `high`
  - source_paths: src/paperqa/docs.py, src/paperqa/settings.py, src/paperqa/types.py, src/paperqa/core.py, src/paperqa/utils.py
- **LLM 与嵌入模型配置 (含 Azure / 本地 LLM)**：importance `high`
  - source_paths: src/paperqa/llms.py, src/paperqa/settings.py, src/paperqa/agents/main.py, src/paperqa/agents/models.py, src/paperqa/configs/tier1_limits.json
- **文档读取、PDF 解析与外部元数据源**：importance `medium`
  - source_paths: src/paperqa/readers.py, src/paperqa/clients/crossref.py, src/paperqa/clients/semantic_scholar.py, src/paperqa/clients/openalex.py, src/paperqa/clients/unpaywall.py

## Repo Inspection Evidence / 源码检查证据

- repo_clone_verified: true
- repo_inspection_verified: true
- repo_commit: `d7675d7b7eddeb3535e8c260399c5bbeeb818c50`
- inspected_files: `README.md`, `pyproject.toml`, `uv.lock`, `docs/tutorials/querying_with_clinical_trials.md`, `docs/tutorials/running_on_lfrqa.md`, `docs/tutorials/settings_tutorial.md`, `docs/tutorials/where_do_I_get_papers.md`, `packages/paper-qa-docling/README.md`, `packages/paper-qa-docling/pyproject.toml`, `packages/paper-qa-docling/src/paperqa_docling/__init__.py`, `packages/paper-qa-docling/src/paperqa_docling/reader.py`, `packages/paper-qa-docling/tests/test_paperqa_docling.py`, `packages/paper-qa-nemotron/README.md`, `packages/paper-qa-nemotron/pyproject.toml`, `packages/paper-qa-nemotron/src/paperqa_nemotron/__init__.py`, `packages/paper-qa-nemotron/src/paperqa_nemotron/api.py`, `packages/paper-qa-nemotron/src/paperqa_nemotron/reader.py`, `packages/paper-qa-nemotron/tests/cassettes/TestNvidiaAPI.test_detection_only[0].yaml`, `packages/paper-qa-nemotron/tests/cassettes/TestNvidiaAPI.test_detection_only[1].yaml`, `packages/paper-qa-nemotron/tests/cassettes/TestNvidiaAPI.test_markdown_bbox[0].yaml`

宿主 AI 硬性规则：
- 没有 repo_clone_verified=true 时，不得声称已经读过源码。
- 没有 repo_inspection_verified=true 时，不得把 README/docs/package 文件判断写成事实。
- 没有 quick_start_verified=true 时，不得声称 Quick Start 已跑通。

## Doramagic Pitfall Constraints / 踩坑约束

这些规则来自 Doramagic 发现、验证或编译过程中的项目专属坑点。宿主 AI 必须把它们当作工作约束，而不是普通说明文字。

### Constraint 1: 来源证据：Is it possible to choose which articles to index and query in the paper directory?

- Trigger: GitHub 社区证据显示该项目存在一个安装相关的待验证问题：Is it possible to choose which articles to index and query in the paper directory?
- Host AI rule: 来源显示可能已有修复、规避或版本变化，说明书中必须标注适用版本。
- Why it matters: 可能影响升级、迁移或版本选择。
- Evidence: community_evidence:github | https://github.com/Future-House/paper-qa/issues/1330 | 来源讨论提到 python 相关条件，需在安装/试用前复核。
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 2: 来源证据：CMYK images in PDFs crash indexing with OSError

- Trigger: GitHub 社区证据显示该项目存在一个配置相关的待验证问题：CMYK images in PDFs crash indexing with OSError
- Host AI rule: 来源显示可能已有修复、规避或版本变化，说明书中必须标注适用版本。
- Why it matters: 可能阻塞安装或首次运行。
- Evidence: community_evidence:github | https://github.com/Future-House/paper-qa/issues/1310 | 来源讨论提到 python 相关条件，需在安装/试用前复核。
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 3: 能力判断依赖假设

- Trigger: README/documentation is current enough for a first validation pass.
- Host AI rule: 将假设转成下游验证清单。
- Why it matters: 假设不成立时，用户拿不到承诺的能力。
- Evidence: capability.assumptions | https://github.com/Future-House/paper-qa | README/documentation is current enough for a first validation pass.
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 4: 维护活跃度未知

- Trigger: 未记录 last_activity_observed。
- Host AI rule: 补 GitHub 最近 commit、release、issue/PR 响应信号。
- Why it matters: 新项目、停更项目和活跃项目会被混在一起，推荐信任度下降。
- Evidence: evidence.maintainer_signals | https://github.com/Future-House/paper-qa | last_activity_observed missing
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

- Trigger: no_demo
- Evidence: downstream_validation.risk_items | https://github.com/Future-House/paper-qa | no_demo; severity=medium
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 6: 存在评分风险

- Trigger: no_demo
- Why it matters: 风险会影响是否适合普通用户安装。
- Evidence: risks.scoring_risks | https://github.com/Future-House/paper-qa | no_demo; severity=medium
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 7: 来源证据：Dependency Dashboard

- Trigger: GitHub 社区证据显示该项目存在一个安全/权限相关的待验证问题：Dependency Dashboard
- Why it matters: 可能影响授权、密钥配置或安全边界。
- Evidence: community_evidence:github | https://github.com/Future-House/paper-qa/issues/399 | 来源讨论提到 python 相关条件，需在安装/试用前复核。
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 8: 来源证据：Insecure pickle deserialization in PaperQA2 persisted search indexes can lead to code execution when querying a poisone…

- Trigger: GitHub 社区证据显示该项目存在一个安全/权限相关的待验证问题：Insecure pickle deserialization in PaperQA2 persisted search indexes can lead to code execution when querying a poisoned index
- Why it matters: 可能影响授权、密钥配置或安全边界。
- Evidence: community_evidence:github | https://github.com/Future-House/paper-qa/issues/1325 | 来源讨论提到 python 相关条件，需在安装/试用前复核。
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 9: issue/PR 响应质量未知

- Trigger: issue_or_pr_quality=unknown。
- Host AI rule: 抽样最近 issue/PR，判断是否长期无人处理。
- Why it matters: 用户无法判断遇到问题后是否有人维护。
- Evidence: evidence.maintainer_signals | https://github.com/Future-House/paper-qa | issue_or_pr_quality=unknown
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 10: 发布节奏不明确

- Trigger: release_recency=unknown。
- Host AI rule: 确认最近 release/tag 和 README 安装命令是否一致。
- Why it matters: 安装命令和文档可能落后于代码，用户踩坑概率升高。
- Evidence: evidence.maintainer_signals | https://github.com/Future-House/paper-qa | release_recency=unknown
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。