mm-review

Stage 3.5: Global Quality Review

Purpose

After all tasks are solved (Stage 3), perform a comprehensive quality review of the entire modeling output. This review is conducted by an independent subagent that does not inherit the solving context, ensuring unbiased evaluation.

The goal is to answer: Is the overall result good enough, or does it need rework?

Input

• mm-workspace/01_analysis.json — problem analysis
• mm-workspace/02_modeling.json — modeling solution and task decomposition
• mm-workspace/03_task_{id}.json — all task outputs (including verification results)

Process

Step 1: Collect All Outputs

Read all workspace files:

1. mm-workspace/01_analysis.json — extract the problem text and analysis
2. mm-workspace/02_modeling.json — extract task list and modeling approach
3. All mm-workspace/03_task_*.json files — extract results, answers, verification status

Step 2: Dispatch Independent Review Subagent

Use the Agent tool to dispatch a general-purpose subagent with the following prompt:

你是一名资深的数学建模竞赛评审专家。请对以下完整的建模结果进行全局质量审查。

## 赛题原文
{Insert problem text from 01_analysis.json}

## 建模方案概述
{Insert modeling_solution from 02_modeling.json, truncated to key points}

## 任务列表及要求
{Insert tasks array from 02_modeling.json}

## 各任务求解结果
{For each task, insert:
  - task_id, description
  - execution_success
  - verification.passed and checks summary
  - result_interpretation (key findings)
  - answer (main conclusions)
}

## 审查维度（每项给出 Pass / Fail + 具体说明）

### 1. 结果一致性（Cross-Task Consistency）
各任务结果之间是否存在矛盾？
- 例如：Task 1 得出"A 方案最优"，Task 3 排序中 A 排第二
- 例如：Task 1 输出的数值范围与 Task 2 输入假设不匹配
- 例如：不同任务对同一变量的定义或取值不一致

### 2. 数值可信度（Numerical Credibility）
核心数值结果是否在合理范围？
- 是否存在量级错误（如人口 10^15）
- 是否存在符号错误（如成本为负）
- 是否存在百分比/概率超出 [0,1] 或 [0%,100%]
- 关键参数是否有数据支撑而非随意取值

### 3. 需求覆盖度（Requirement Coverage）
是否完整回答了赛题的所有子问题？
- 对照赛题原文，逐条检查是否都有对应回答
- 每个子问题是否有具体的数值结论（而非只有定性描述）
- 赛题要求的图表是否都已生成

### 4. 模型连贯性（Model Coherence）
各子模型之间的方法选择是否自洽？整体建模故事是否通顺？
- 方法选择是否有矛盾（如前面假设线性关系，后面用非线性模型）
- 各任务的方法是否形成一个完整的故事线
- 建模假设体系是否一致（不冲突）

### 5. 论据充分性（Evidence Sufficiency）
关键结论是否有充分的数据/图表/推导支撑？
- 核心结论是否只靠定性论述，缺少定量支撑
- 图表是否清晰展示关键结果
- 灵敏度分析是否充分

## 输出格式

```json
{
  "dimensions": [
    {"name": "cross_task_consistency", "passed": true/false, "details": "具体说明"},
    {"name": "numerical_credibility", "passed": true/false, "details": "具体说明"},
    {"name": "requirement_coverage", "passed": true/false, "details": "具体说明"},
    {"name": "model_coherence", "passed": true/false, "details": "具体说明"},
    {"name": "evidence_sufficiency", "passed": true/false, "details": "具体说明"}
  ],
  "overall_passed": true/false,
  "rework_list": [
    {
      "task_id": 2,
      "reason": "具体问题描述",
      "improvement_direction": "改进方向（不是完整方案）"
    }
  ],
  "summary": "一段话总结评审结论"
}

如果 overall_passed 为 false，rework_list 中必须列出需要重修的任务及具体原因。 如果 overall_passed 为 true，rework_list 为空数组。

### Step 3: Process Review Results

Receive the subagent's review output. Parse the JSON response.

If **overall_passed = true**:
- All 5 dimensions passed
- Save review results and proceed

If **overall_passed = false**:
- At least one dimension failed
- Extract the `rework_list`
- Present findings to user

### Step 4: Save Review Output

Write to `mm-workspace/03.5_review.json`:

```json
{
  "iteration": 1,
  "overall_passed": true,
  "dimensions": [
    {"name": "cross_task_consistency", "passed": true, "details": "..."},
    {"name": "numerical_credibility", "passed": true, "details": "..."},
    {"name": "requirement_coverage", "passed": true, "details": "..."},
    {"name": "model_coherence", "passed": true, "details": "..."},
    {"name": "evidence_sufficiency", "passed": true, "details": "..."}
  ],
  "rework_list": [],
  "summary": "review summary text",
  "stage": "review_complete"
}

Then commit: cd mm-workspace && git add -A && git commit -m "feat(s3.5): global review - iteration {N}"