# Daily Workflow 4 — `paper_reader`: PDF URL / arXiv / Local PDF -> Translation + Core Thesis & Claims + Detailed Experimental Data Extraction (XLSX/CSV) + Method Diagram (D2) + Mindmap + PDF Dossier.
#
#   yanshi workflow run examples/generators/paper_reader.yml \
#     --input source "https://arxiv.org/abs/1706.03762" --output-dir out
name: paper_reader
description: "输入论文 PDF 文件、arXiv 链接或 PDF URL，完成核心章节双语精译、核心思想与论点提炼、详细实验数据表格提取（XLSX/CSV）、算法架构图、思维导图与精读报告 PDF"

inputs:
  source:
    description: "论文 arXiv 链接、PDF URL 或本地上传 PDF（支持 Studio 一键选择本地 PDF）"
    default: "https://arxiv.org/abs/1706.03762"
  target_lang:
    description: "翻译与精读报告目标语言"
    default: "简体中文"
  max_pages:
    description: "最大解析页数上限（含正文、实验表与关键附录）"
    default: 24
  focus:
    description: "重点研读方向"
    default: "问题动机、核心方法创新、数学/架构机制、消融与主实验数据对比、局限性"
  export_pdf:
    description: "同时导出完整论文精读报告 PDF"
    default: true

nodes:
  - id: extract
    type: pdf_extract
    parameters:
      max_chars: 120000

  - id: context_query
    type: template
    dependencies: ["extract"]
    input_mapping:
      paper_text: "{{ nodes.extract.outputs.output }}"
    parameters:
      template: "{{ paper_text | truncate(length=200) | trim }} paper key contributions benchmarks"

  - id: scholarly_context
    type: web_search
    dependencies: ["context_query"]
    input_mapping:
      query: "{{ nodes.context_query.outputs.output }}"
    parameters:
      max_results: 5

  - id: translation_prompt
    type: template
    dependencies: ["extract"]
    input_mapping:
      paper_text: "{{ nodes.extract.outputs.output }}"
    parameters:
      template: |
        你是一位资深计算机与科学文献双语译者。请针对以下提取自论文原件的文本，面向「{{ target_lang }}」读者完成高质量忠实精译（信、达、专有名词保留英文括注，公式符号保持原样）：
        1. `paper_title_original` 与 `paper_title_translated`；
        2. `abstract_bilingual`：摘要原文与{{ target_lang }}逐段对照精译；
        3. `section_translations`：按论文 4-6 个最核心章节顺序（引言、核心方法/模型设计、关键理论/架构细节、结论），输出每一节的高密度精准{{ target_lang }}精译（每节控制在 200-350 字核心译文）与术语批注。

        论文全文提取内容：
        {{ paper_text | truncate(length=52000) }}

  - id: translate_paper
    type: llm
    dependencies: ["translation_prompt"]
    input_mapping:
      prompt: "{{ nodes.translation_prompt.outputs.output }}"
    parameters:
      max_tokens: 8192
      system: "你是严谨的学术论文翻译专家，术语准确、逻辑忠实于原文，仅输出合法 JSON。"
      json_schema:
        type: object
        required: [paper_title_original, paper_title_translated, abstract_bilingual, section_translations]
        properties:
          paper_title_original: { type: string }
          paper_title_translated: { type: string }
          authors_and_venue: { type: string }
          abstract_bilingual: { type: string }
          section_translations:
            type: array
            minItems: 2
            maxItems: 7
            items:
              type: object
              required: [section_heading, translated_markdown, key_terms]
              properties:
                section_heading: { type: string }
                translated_markdown: { type: string }
                key_terms:
                  type: array
                  items: { type: string }

  - id: thesis_prompt
    type: template
    dependencies: ["extract", "scholarly_context"]
    input_mapping:
      paper_text: "{{ nodes.extract.outputs.output }}"
      web_ctx: "{{ nodes.scholarly_context.outputs.markdown }}"
    parameters:
      template: |
        请以顶会审稿人与领域研究者的视角，围绕「{{ focus }}」深度精读以下论文，并用「{{ target_lang }}」提炼：
        1. `one_sentence_insight`：一句话直击论文最本质的核心思想突破；
        2. `problem_and_bottleneck`：既有方法面临的根本瓶颈与本文切入点；
        3. `core_claims`：论文提出的 3-6 条核心论点，每条论点给出对应的理论推导或实验证据支撑（注明章节/图表出处）；
        4. `method_mechanics`：算法/架构关键机制拆解（含关键公式含义与复杂度权衡）；
        5. `limitations_and_future`：边界条件、局限性与后续演进脉络；
        6. `method_d2_brief`：用于 D2 绘制「论文核心算法/模型数据流架构图」的精确结构说明；
        7. `mindmap_tree`：论文全景知识思维导图树（根节点为论文简称，含研究动机、核心思想、关键方法、实验证据、启发与局限等分支，深度不超过 3 层）。

        论文正文：
        {{ paper_text | truncate(length=52000) }}

        学术背景补充：
        {{ web_ctx | truncate(length=6000) }}

  - id: analyze_thesis
    type: llm
    dependencies: ["thesis_prompt"]
    input_mapping:
      prompt: "{{ nodes.thesis_prompt.outputs.output }}"
    parameters:
      max_tokens: 8192
      system: "你是严谨的顶会审稿人，一切论点与数据必须严格忠于论文原文，严禁编造数值，仅输出合法 JSON。"
      json_schema:
        type: object
        required:
          - one_sentence_insight
          - problem_and_bottleneck
          - core_claims
          - method_mechanics
          - limitations_and_future
          - method_d2_brief
          - mindmap_tree
        properties:
          one_sentence_insight: { type: string }
          problem_and_bottleneck: { type: string }
          core_claims:
            type: array
            minItems: 2
            maxItems: 8
            items:
              type: object
              required: [claim, evidence_in_paper, source_section]
              properties:
                claim: { type: string }
                evidence_in_paper: { type: string }
                source_section: { type: string }
          method_mechanics: { type: string }
          limitations_and_future: { type: string }
          method_d2_brief: { type: string }
          mindmap_tree:
            type: object
            required: [label, children]
            properties:
              label: { type: string }
              children:
                type: array
                items:
                  type: object
                  required: [label]
                  properties:
                    label: { type: string }
                    children:
                      type: array
                      items:
                        type: object
                        required: [label]
                        properties:
                          label: { type: string }
                          children:
                            type: array
                            items:
                              type: object
                              required: [label]
                              properties:
                                label: { type: string }

  - id: experiments_prompt
    type: template
    dependencies: ["extract"]
    input_mapping:
      paper_text: "{{ nodes.extract.outputs.output }}"
    parameters:
      template: |
        请从以下论文全文中提取最具代表性的 8-18 行关键实验设置与定量对比数据（涵盖主实验、基线对比、消融实验 Ablation Study、吞吐/参数量/开销对比）：
        1. `experimental_setup`：数据集、评价指标、训练/推理硬件与关键超参数总结；
        2. `ablation_findings`：关键消融实验发现与变量贡献度解读；
        3. `metrics_rows`：结构化实验数据行数组（8-18 行核心数据），每行包含统一字段：
           - `table_or_figure`（原论文图表编号，如 "Table 1" / "Table 3"）
           - `task_or_dataset`（任务或数据集名称）
           - `model_or_variant`（对比基线或本文模型变体）
           - `metric_name`（指标名称，如 BLEU / Accuracy / Latency / Params）
           - `metric_value`（论文原始数值字符串，原样保留）
           - `delta_vs_baseline`（相对最强基线或基准变体的提升/变化说明）
           - `supports_claim`（该组数据支撑的核心论点）

        论文全文：
        {{ paper_text | truncate(length=56000) }}

  - id: extract_experiments
    type: llm
    dependencies: ["experiments_prompt"]
    input_mapping:
      prompt: "{{ nodes.experiments_prompt.outputs.output }}"
    parameters:
      max_tokens: 8192
      system: "你是定量科学数据分析专家，严格提取论文真实实验表格与正文报告的原始数据，绝不虚构指标，仅输出合法 JSON。"
      json_schema:
        type: object
        required: [experimental_setup, ablation_findings, metrics_rows]
        properties:
          experimental_setup: { type: string }
          ablation_findings: { type: string }
          metrics_rows:
            type: array
            minItems: 1
            maxItems: 18
            items:
              type: object
              required:
                - table_or_figure
                - task_or_dataset
                - model_or_variant
                - metric_name
                - metric_value
                - delta_vs_baseline
                - supports_claim
              properties:
                table_or_figure: { type: string }
                task_or_dataset: { type: string }
                model_or_variant: { type: string }
                metric_name: { type: string }
                metric_value: { type: string }
                delta_vs_baseline: { type: string }
                supports_claim: { type: string }

  - id: exp_rows_json
    type: template
    dependencies: ["extract_experiments"]
    input_mapping:
      exp: "{{ nodes.extract_experiments.outputs.output }}"
    parameters:
      output_format: json
      template: "{{ exp.metrics_rows | json_encode() }}"

  - id: experiments_xlsx
    type: render_table
    dependencies: ["exp_rows_json"]
    input_mapping:
      rows: "{{ nodes.exp_rows_json.outputs.output }}"
    parameters:
      format: xlsx
      columns:
        - table_or_figure
        - task_or_dataset
        - model_or_variant
        - metric_name
        - metric_value
        - delta_vs_baseline
        - supports_claim

  - id: experiments_csv
    type: render_table
    dependencies: ["exp_rows_json"]
    input_mapping:
      rows: "{{ nodes.exp_rows_json.outputs.output }}"
    parameters:
      format: csv
      columns:
        - table_or_figure
        - task_or_dataset
        - model_or_variant
        - metric_name
        - metric_value
        - delta_vs_baseline
        - supports_claim

  - id: method_fig_desc
    type: template
    dependencies: ["analyze_thesis", "translate_paper"]
    input_mapping:
      thesis: "{{ nodes.analyze_thesis.outputs.output }}"
      trans: "{{ nodes.translate_paper.outputs.output }}"
    parameters:
      template: |
        为论文《{{ trans.paper_title_original }}》绘制一张清晰展示其核心模型/算法数据流机制的 D2 架构图（direction: right，8-14 个关键模块节点，标注张量/数据流转换关系）：
        {{ thesis.method_d2_brief }}

  - id: method_diagram
    type: figure
    dependencies: ["method_fig_desc"]
    input_mapping:
      description: "{{ nodes.method_fig_desc.outputs.output }}"
    parameters:
      language: d2
      png: true
      png_width: 1800
      max_attempts: 4

  - id: mindmap_tree_json
    type: template
    dependencies: ["analyze_thesis"]
    input_mapping:
      thesis: "{{ nodes.analyze_thesis.outputs.output }}"
    parameters:
      output_format: json
      template: '{"tree": {{ thesis.mindmap_tree | json_encode() }}}'

  - id: mindmap_svg
    type: render_mindmap
    dependencies: ["mindmap_tree_json"]
    input_mapping:
      tree: "{{ nodes.mindmap_tree_json.outputs.tree }}"

  - id: mindmap_png
    type: render_svg
    dependencies: ["mindmap_svg"]
    input_mapping:
      svg: "{{ nodes.mindmap_svg.outputs.mindmap_svg_output }}"
    parameters:
      width: 1600

  - id: mindmap_html
    type: render_mindmap
    dependencies: ["mindmap_tree_json"]
    input_mapping:
      tree: "{{ nodes.mindmap_tree_json.outputs.tree }}"
    parameters:
      format: html

  - id: assemble_dossier
    type: template
    dependencies:
      - extract
      - translate_paper
      - analyze_thesis
      - extract_experiments
      - method_diagram
      - mindmap_png
    input_mapping:
      pages: "{{ nodes.extract.outputs.page_count }}"
      trans: "{{ nodes.translate_paper.outputs.output }}"
      thesis: "{{ nodes.analyze_thesis.outputs.output }}"
      exp: "{{ nodes.extract_experiments.outputs.output }}"
      method_png: "{{ nodes.method_diagram.outputs.method_diagram_png }}"
      mm_png: "{{ nodes.mindmap_png.outputs.mindmap_png_output }}"
    parameters:
      template: |
        # {{ trans.paper_title_translated }}

        > **原文标题**：*{{ trans.paper_title_original }}*
        {% if trans.authors_and_venue %}> **作者与出处**：{{ trans.authors_and_venue }}{% endif %}
        > **解析页数**：共 {{ pages }} 页

        ## 1. 一句话核心思想与知识思维导图

        > **核心突破**：{{ thesis.one_sentence_insight }}

        ![论文全景思维导图]({{ mm_png }})

        ## 2. 研究问题与核心论点链条

        ### 2.1 既有瓶颈与问题定义

        {{ thesis.problem_and_bottleneck }}

        ### 2.2 核心论点与原文证据映射

        | 序号 | 核心论点 (Claim) | 论文实验/理论证据支撑 | 原文位置 |
        | --- | --- | --- | --- |
        {% for c in thesis.core_claims %}| {{ loop.index }} | **{{ c.claim }}** | {{ c.evidence_in_paper }} | `{{ c.source_section }}` |
        {% endfor %}

        ## 3. 核心算法与架构机制深度拆解

        ![核心方法架构图]({{ method_png }})

        {{ thesis.method_mechanics }}

        ## 4. 详细实验设置与定量数据提取

        ### 4.1 实验设置与基准说明

        {{ exp.experimental_setup }}

        ### 4.2 定量实验结果对比总表

        | 来源图表 | 任务 / 数据集 | 模型 / 变体 | 指标 | 数值 | 相对基线变化 | 支撑论点 |
        | --- | --- | --- | --- | --- | --- | --- |
        {% for r in exp.metrics_rows %}| `{{ r.table_or_figure }}` | {{ r.task_or_dataset }} | `{{ r.model_or_variant }}` | {{ r.metric_name }} | **{{ r.metric_value }}** | {{ r.delta_vs_baseline }} | {{ r.supports_claim }} |
        {% endfor %}

        ### 4.3 消融实验与关键规律解读

        {{ exp.ablation_findings }}

        ## 5. 摘要双语对照与核心章节精译

        ### 5.1 摘要双语对照

        {{ trans.abstract_bilingual }}

        {% for sec in trans.section_translations %}
        ### 5.{{ loop.index + 1 }} {{ sec.section_heading }}

        {{ sec.translated_markdown }}
        {% if sec.key_terms | length > 0 %}
        - **关键术语对照**：{{ sec.key_terms | join(sep=" · ") }}
        {% endif %}
        {% endfor %}

        ## 6. 局限性与后续研究启发

        {{ thesis.limitations_and_future }}

  - id: dossier_md
    type: render_markdown
    dependencies: ["assemble_dossier"]
    input_mapping:
      markdown: "{{ nodes.assemble_dossier.outputs.output }}"
    parameters:
      format: md

  - id: dossier_html
    type: render_markdown
    dependencies: ["assemble_dossier"]
    input_mapping:
      markdown: "{{ nodes.assemble_dossier.outputs.output }}"

  - id: pdf_flag
    type: template
    parameters:
      output_format: json
      template: '{"want_pdf": {{ export_pdf }}}'

  - id: dossier_pdf
    type: render_document
    dependencies: ["assemble_dossier", "pdf_flag"]
    run_if: "{{ nodes.pdf_flag.outputs.want_pdf }}"
    input_mapping:
      markdown: "{{ nodes.assemble_dossier.outputs.output }}"
    parameters:
      format: pdf
      toc: true

  - id: deliverable
    type: template
    dependencies:
      - translate_paper
      - analyze_thesis
      - extract_experiments
      - experiments_xlsx
      - experiments_csv
      - method_diagram
      - mindmap_svg
      - mindmap_html
      - dossier_md
      - dossier_html
      - dossier_pdf
    input_mapping:
      trans: "{{ nodes.translate_paper.outputs.output }}"
      thesis: "{{ nodes.analyze_thesis.outputs.output }}"
      exp: "{{ nodes.extract_experiments.outputs.output }}"
      xlsx: "{{ nodes.experiments_xlsx.outputs.experiments_xlsx_output }}"
      csv: "{{ nodes.experiments_csv.outputs.experiments_csv_output }}"
      fig_svg: "{{ nodes.method_diagram.outputs.method_diagram_svg }}"
      mm_svg: "{{ nodes.mindmap_svg.outputs.mindmap_svg_output }}"
      mm_html: "{{ nodes.mindmap_html.outputs.mindmap_html_output }}"
      md: "{{ nodes.dossier_md.outputs.dossier_md_output }}"
      html: "{{ nodes.dossier_html.outputs.dossier_html_output }}"
      pdf: "{{ nodes.dossier_pdf.outputs.dossier_pdf_output }}"
    parameters:
      output_format: json
      template: |
        {"deliverable": {
          "kind": "paper_reader",
          "title": {{ trans.paper_title_translated | json_encode() }},
          "content": {"thesis": {{ thesis | json_encode() }}, "experiments": {{ exp | json_encode() }}},
          "citations": [],
          "files": [
            {"node_id": "dossier_html", "name": {{ html | split(pat="/") | last | json_encode() }}, "mime_type": "text/html", "role": "preview"},
            {"node_id": "dossier_md", "name": {{ md | split(pat="/") | last | json_encode() }}, "mime_type": "text/markdown", "role": "asset"},
            {"node_id": "experiments_xlsx", "name": {{ xlsx | split(pat="/") | last | json_encode() }}, "mime_type": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet", "role": "primary"},
            {"node_id": "experiments_csv", "name": {{ csv | split(pat="/") | last | json_encode() }}, "mime_type": "text/csv", "role": "asset"},
            {"node_id": "method_diagram", "name": {{ fig_svg | split(pat="/") | last | json_encode() }}, "mime_type": "image/svg+xml", "role": "asset"},
            {"node_id": "mindmap_svg", "name": {{ mm_svg | split(pat="/") | last | json_encode() }}, "mime_type": "image/svg+xml", "role": "asset"},
            {"node_id": "mindmap_html", "name": {{ mm_html | split(pat="/") | last | json_encode() }}, "mime_type": "text/html", "role": "preview"}{% if pdf %},
            {"node_id": "dossier_pdf", "name": {{ pdf | split(pat="/") | last | json_encode() }}, "mime_type": "application/pdf", "role": "primary"}{% endif %}
          ]
        }}
