--- name: pangzhenying2025/automotive-sotif-audit source: https://app.decimal.ai/s/pangzhenying2025-automotive-sotif-audit@1/SKILL.md source_sha256: d06e497e970c --- # Automotive Sotif Audit ## Sotif Audit # SOTIF Audit & Assessment — Process Maturity and Compliance Evaluation ## Overview Systematic audit and assessment framework for evaluating SOTIF (ISO 21448) implementation maturity across the product development lifecycle. Provides audit checklists, maturity models, and gap assessment tools for ADAS/ADS projects. ## SOTIF Maturity Model ``` SOTIF实施成熟度模型（5级） ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Level 1: Initial (初始级) ├── 无系统性SOTIF流程 ├── 依赖个人经验识别风险 ├── 测试以功能验证为主 └── 无SOTIF专项文档 Level 2: Managed (管理级) ├── 基本SOTIF分析流程建立 ├── 触发条件清单初步建立 ├── 部分场景覆盖仿真 ├── SOTIF角色分配 └── 基础文档体系 Level 3: Defined (定义级) ├── 完整SOTIF流程贯穿V模型 ├── 系统化触发条件识别方法 ├── 场景库建立（>500场景） ├── 量化覆盖度指标 ├── 跨部门协作机制 └── 与功能安全/网络安全集成 Level 4: Quantitatively Managed (量化管理级) ├── 基于数据的残余风险量化 ├── 自然驾驶数据驱动场景扩展 ├── 持续监控与闭环反馈 ├── 仿真-实车-运营数据链路 └── 预测性风险评估 Level 5: Optimizing (优化级) ├── AI辅助场景发现 ├── 持续学习与自适应安全 ├── 行业基准对标 ├── 前瞻性标准影响 └── 跨OEM经验共享 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ``` ## Audit Checklist ### Phase 1: Specification & Design (ISO 21448 Clause 5-7) ```markdown ## SOTIF Specification & Design Audit ### 5. Functional and System Specification - [ ] System functions clearly defined with intended behavior - [ ] ODD boundaries explicitly specified - [ ] Performance limitations documented per sensor/algorithm - [ ] Known functional insufficiencies identified - [ ] System architecture supports SOTIF analysis ### 6. Identification of Triggering Conditions & Hazardous Behaviors - [ ] Systematic method for triggering condition identification (STPA/HAZOP/brainstorming) - [ ] Triggering conditions categorized by source (sensor/algorithm/human/infrastructure) - [ ] Functional insufficiency chain analysis (TC → FI → HB → Harm) - [ ] Severity/Exposure/Controllability assessment per hazardous behavior - [ ] Coverage of China-specific triggering conditions (if applicable) - [ ] Triggering condition catalog maintained and version-controlled ### 7. Evaluation of Known Hazardous Scenarios - [ ] Hazardous scenario catalog established - [ ] Risk assessment performed (S/E/C or similar) - [ ] Acceptance criteria defined for residual risk - [ ] Scenarios prioritized for V&V - [ ] Traceability from TC to test cases ``` ### Phase 2: Design Measures (ISO 21448 Clause 8-9) ```markdown ## SOTIF Design Measures Audit ### 8. Functional Modifications to Reduce Risk - [ ] Design measures identified for each known hazardous scenario - [ ] Sensor fusion strategy addresses individual sensor limitations - [ ] Algorithm robustness measures (redundancy, plausibility checks) - [ ] ODD monitoring and graceful degradation implemented - [ ] DMS integration for human factor mitigation - [ ] Effectiveness of measures verified (before/after risk comparison) ### 9. Risk Evaluation of Residual Scenarios - [ ] Residual risk quantification method defined - [ ] Acceptance criteria for residual risk established - [ ] Comparison with societal acceptance levels - [ ] Statistical evidence for residual risk claims - [ ] Gap between required and achieved evidence documented ``` ### Phase 3: Verification & Validation (ISO 21448 Clause 10-11) ```markdown ## SOTIF V&V Audit ### 10. Verification of Known Scenarios - [ ] Simulation test plan for known hazardous scenarios - [ ] Simulation platform qualified (model validation) - [ ] Track test plan with scenario coverage mapping - [ ] Public road test plan with exposure tracking - [ ] Pass/fail criteria defined per scenario - [ ] Test coverage metrics tracked and reported - [ ] Failed scenarios → design iteration loop documented ### 11. Validation of Unknown Scenarios - [ ] Strategy for discovering unknown unsafe scenarios - [ ] Naturalistic driving data analysis methodology - [ ] Exploratory testing methods (fuzzing, adversarial testing) - [ ] Field monitoring and incident analysis process - [ ] Statistical argument for absence of unreasonable risk - [ ] Confidence level in residual risk assessment ``` ### Phase 4: Operations & Post-Market (ISO 21448 Clause 12) ```markdown ## SOTIF Operations Audit ### 12. Post-Development Activities - [ ] Field monitoring system deployed - [ ] Incident reporting and analysis process - [ ] Near-miss detection capability - [ ] OTA update process for SOTIF improvements - [ ] Customer complaint analysis for SOTIF issues - [ ] Periodic SOTIF reassessment trigger criteria - [ ] Regulatory reporting compliance (if applicable) ``` ## Assessment Scoring Framework ```python # SOTIF Audit Scoring audit_categories = { "specification": { "weight": 0.20, "items": [ ("Function specification completeness", 10), ("ODD definition clarity", 10), ("Performance limitation documentation", 10), ], }, "hazard_identification": { "weight": 0.25, "items": [ ("Triggering condition identification method", 10), ("TC catalog completeness", 10), ("Hazardous behavior analysis depth", 10), ("Risk assessment quality", 10), ], }, "design_measures": { "weight": 0.20, "items": [ ("Measure effectiveness", 10), ("Residual risk quantification", 10), ("Acceptance criteria appropriateness", 10), ], }, "verification_validation": { "weight": 0.25, "items": [ ("Simulation test coverage", 10), ("Track test execution", 10), ("Public road evidence", 10), ("Unknown scenario exploration", 10), ], }, "operations": { "weight": 0.10, "items": [ ("Field monitoring", 10), ("Incident analysis process", 10), ("Continuous improvement", 10), ], }, } def calculate_sotif_score(scores: dict) -> dict: """Calculate weighted SOTIF audit score""" category_scores = {} total = 0 for cat, config in audit_categories.items(): max_points = sum(item[1] for item in config["items"]) actual = sum(scores.get(cat, {}).get(item[0], 0) for item in config["items"]) pct = actual / max_points * 100 category_scores[cat] = pct total += pct * config["weight"] return { "category_scores": category_scores, "total_score": total, "maturity_level": ( 5 if total >= 90 else 4 if total >= 75 else 3 if total >= 60 else 2 if total >= 40 else 1 ), } ``` ## Common Audit Findings ``` 常见SOTIF审核发现 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ Finding 1: ODD定义不充分 ├── 问题：ODD仅列出道路类型，未定义环境/天气/基础设施边界 └── 建议：使用6层场景模型系统定义ODD Finding 2: 触发条件识别不系统 ├── 问题：依赖头脑风暴，缺少系统方法 └── 建议：采用STPA+HAZOP+历史事故分析组合方法 Finding 3: 仿真与实车测试脱节 ├── 问题：仿真场景与实车测试场景不对应 └── 建议：建立统一场景库，仿真-实车-数据三位一体 Finding 4: 残余风险论证不足 ├── 问题：缺少定量证据支撑"风险可接受"结论 └── 建议：基于自然驾驶暴露数据量化残余风险 Finding 5: 未知不安全场景探索缺失 ├── 问题：仅测试已知场景，未探索未知风险 └── 建议：引入NDD分析、对抗测试、随机化测试 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ``` ## Deliverables 1. **Audit Report**: Structured findings with severity and recommendations 2. **Maturity Assessment**: Current level and improvement roadmap 3. **Gap Analysis**: Requirements vs. implementation status 4. **Improvement Plan**: Prioritized actions with timeline 5. **Benchmark Report**: Comparison with industry best practices ## Related Skills - `automotive-sotif-hazard-scenario` — Scenario identification methodology - `automotive-sotif-highway-testing` — Highway SOTIF testing specifics - `iso-26262-overview` — Functional safety baseline