--- Testing Score Stability (Terminology Typo) ---
DEBUG: Running Rule Engine...
DEBUG: Fetching RAG Context...
DEBUG: Performing similarity search...
DEBUG: Similarity search finished.
DEBUG: Starting LLM Analysis...
DEBUG: Invoking LLM (glm-4-flash)...
DEBUG: LLM (glm-4-flash) response received in 3.87s.
DEBUG: Total QC processing time: 6.30s.
### 超声报告质控分析报告

#### ⚠️ 自动规则检测（高可靠）：
- 【建议修正】术语错误：检测到'冥想异常'，应修正为'异常回声'。
- 【建议修正】术语错误：检测到'冥想灶'，应修正为'异常回声灶'。
- 【建议修正】术语错误：检测到'冥想'，应修正为'异常回声'。

#### 🤖 AI 智能深度分析：
```json
{
  "score": 97,
  "issues": [
    {
      "original": "冥想",
      "reason": "术语错误",
      "suggestion": "回声",
      "type": "术语瑕疵/错别字/标点规范"
    }
  ],
  "corrected_fields": {
    "冥想": "回声"
  }
}
```