伤害修复模型建立自我边界的智能决策系统一、实际应用场景描述场景职场/亲密关系中的情感勒索- 同事A经常推卸责任给你事后说都是为团队好你大度点- 伴侣B忘记重要纪念日却说真正爱你的人不会计较这些- 朋友C借钱不还反而指责你不够义气现状传统道德观要求原谅大度导致受害者陷入自责循环边界感持续被侵蚀。二、引入痛点1. 道德绑架陷阱施害者将不原谅等同于心胸狭窄2. 自我PUA机制受害者内化他人标准产生是我太计较的认知扭曲3. 边界模糊化缺乏量化评估体系难以判断何时该坚持立场4. 情绪劳动过载反复纠结消耗心理能量影响决策质量三、核心逻辑讲解伤害修复模型三大支柱┌─────────────────────────────────────────────────────────┐│ 伤害修复模型 │├─────────────┬─────────────┬─────────────┤│ 伤害评估器 │ 修复验证器 │ 边界守护器 │├─────────────┼─────────────┼─────────────┤│ 量化伤害程度 │ 验证修复诚意 │ 动态边界调整 ││ 识别模式类型 │ 评估行为改变 │ 拒绝二次伤害 ││ 计算情绪成本 │ 确认责任承担 │ 自我保护策略 │└─────────────┴─────────────┴─────────────┘核心算法- 伤害指数 (频率×强度) 权力不对等系数 - 历史修复记录- 修复阈值 基础阈值 × (1 关系重要性权重)- 边界强度 伤害指数 / 修复进度 × 自我价值系数四、代码模块化实现项目结构boundary_guardian/├── core/│ ├── __init__.py│ ├── harm_assessor.py # 伤害评估器│ ├── repair_validator.py # 修复验证器│ └── boundary_keeper.py # 边界守护器├── models/│ ├── __init__.py│ └── relationship.py # 关系数据模型├── utils/│ ├── __init__.py│ └── decision_matrix.py # 决策矩阵工具├── main.py # 主程序入口├── config.py # 配置文件└── README.md # 项目说明1. 配置文件 (config.py)配置文件定义模型参数和阈值from dataclasses import dataclass, fieldfrom typing import Dict, Listdataclassclass ModelConfig:模型核心配置# 伤害评估参数HARM_FREQUENCY_WEIGHT: float 0.3 # 频率权重HARM_INTENSITY_WEIGHT: float 0.5 # 强度权重POWER_IMBALANCE_MULTIPLIER: float 1.5 # 权力不对等乘数# 修复验证参数MIN_REPAIR_ACTIONS: int 3 # 最低修复行动数REPAIR_TIME_WINDOW_DAYS: int 30 # 修复观察期(天)BEHAVIOR_CHANGE_THRESHOLD: float 0.7 # 行为改变阈值# 边界守护参数BASE_BOUNDARY_STRENGTH: float 50.0 # 基础边界强度SELF_VALUE_COEFFICIENT: float 1.2 # 自我价值系数MAX_FORGIVENESS_ATTEMPTS: int 2 # 最大原谅尝试次数dataclassclass RelationshipWeights:关系类型权重配置STRANGER: float 0.2COLLEAGUE: float 0.4FRIEND: float 0.6PARTNER: float 0.8FAMILY: float 0.9# 伤害类型定义HARM_TYPES {emotional_abuse: 情感虐待,responsibility_dumping: 责任转嫁,boundary_violation: 边界侵犯,manipulation: 情感操控,neglect: 情感忽视,betrayal: 背叛失信}# 修复行动类型REPAIR_ACTIONS {apology_sincerity: 真诚道歉,accountability_taking: 承担责任,behavior_change: 行为改变,compensation: 合理补偿,prevention_plan: 预防方案,time_proven: 时间证明}2. 数据模型 (models/relationship.py)关系数据模型存储和管理关系信息from dataclasses import dataclass, fieldfrom datetime import datetime, timedeltafrom typing import List, Optional, Dictimport uuiddataclassclass HarmEvent:伤害事件数据类用于记录单次伤害事件的详细信息event_id: str field(default_factorylambda: str(uuid.uuid4()))harm_type: str # 伤害类型description: str # 事件描述intensity: int 1 # 伤害强度(1-10)frequency: int 1 # 发生频率(1-10)power_imbalance: float 1.0 # 权力不对等系数occurred_at: datetime field(default_factorydatetime.now)is_repeated: bool False # 是否重复发生has_previous_warning: bool False # 是否有过预警def calculate_harm_score(self, config: ModelConfig) - float:计算单个事件的伤害分数公式: (频率×强度) 权力不对等系数Returns:float: 伤害分数base_score (self.frequency * config.HARM_INTENSITY_WEIGHT self.intensity * config.HARM_FREQUENCY_WEIGHT)return base_score * self.power_imbalancedataclassclass RepairAction:修复行动数据类用于记录施害者的修复行为和验证结果action_id: str field(default_factorylambda: str(uuid.uuid4()))action_type: str # 行动类型description: str # 行动描述taken_at: datetime field(default_factorydatetime.now)verified: bool False # 是否已验证verification_notes: str # 验证备注effectiveness_score: float 0.0 # 有效性评分(0-1)def to_dict(self) - Dict:转换为字典格式return {action_id: self.action_id,action_type: self.action_type,description: self.description,taken_at: self.taken_at.isoformat(),verified: self.verified,effectiveness_score: self.effectiveness_score}dataclassclass Relationship:关系数据类管理完整的关系信息和伤害/修复历史relationship_id: str field(default_factorylambda: str(uuid.uuid4()))name: str # 关系对象名称relationship_type: str colleague # 关系类型importance_weight: float 0.5 # 关系重要性权重created_at: datetime field(default_factorydatetime.now)# 历史记录harm_events: List[HarmEvent] field(default_factorylist)repair_actions: List[RepairAction] field(default_factorylist)# 状态追踪total_harm_score: float 0.0 # 累计伤害分数forgiveness_attempts: int 0 # 原谅尝试次数current_boundary_strength: float 50.0 # 当前边界强度last_interaction: datetime field(default_factorydatetime.now)def add_harm_event(self, event: HarmEvent) - None:添加伤害事件并更新累计分数self.harm_events.append(event)self.total_harm_score event.calculate_harm_score(ModelConfig())self.last_interaction datetime.now()def add_repair_action(self, action: RepairAction) - None:添加修复行动self.repair_actions.append(action)self.last_interaction datetime.now()def get_harm_by_type(self) - Dict[str, float]:按类型统计伤害分数type_scores {}for event in self.harm_events:if event.harm_type not in type_scores:type_scores[event.harm_type] 0.0type_scores[event.harm_type] event.calculate_harm_score(ModelConfig())return type_scoresdef get_recent_harms(self, days: int 90) - List[HarmEvent]:获取最近N天的伤害事件cutoff_date datetime.now() - timedelta(daysdays)return [e for e in self.harm_events if e.occurred_at cutoff_date]def to_dict(self) - Dict:转换为可序列化的字典return {relationship_id: self.relationship_id,name: self.name,relationship_type: self.relationship_type,importance_weight: self.importance_weight,total_harm_score: round(self.total_harm_score, 2),forgiveness_attempts: self.forgiveness_attempts,current_boundary_strength: round(self.current_boundary_strength, 2),harm_count: len(self.harm_events),repair_count: len(self.repair_actions)}3. 伤害评估器 (core/harm_assessor.py)伤害评估器模块负责量化分析伤害程度识别伤害模式计算情绪成本from dataclasses import dataclassfrom datetime import datetime, timedeltafrom typing import List, Dict, Tuple, Optionalimport statisticsfrom . import configdataclassclass HarmAssessmentResult:伤害评估结果包含完整的评估指标和分析建议total_harm_index: float # 总伤害指数pattern_analysis: Dict[str, any] # 模式分析结果emotional_cost: float # 情绪成本risk_level: str # 风险等级assessment_time: datetime field(default_factorydatetime.now)def to_dict(self) - Dict:return {total_harm_index: round(self.total_harm_index, 2),pattern_analysis: self.pattern_analysis,emotional_cost: round(self.emotional_cost, 2),risk_level: self.risk_level,assessment_time: self.assessment_time.isoformat()}class HarmAssessor:伤害评估器主类核心功能1. 计算综合伤害指数2. 识别伤害模式和趋势3. 评估情绪劳动成本4. 生成风险评估报告def __init__(self, model_config: config.ModelConfig None):self.config model_config or config.ModelConfig()self.assessment_history: List[HarmAssessmentResult] []def assess_relationship(self,relationship: models.Relationship) - HarmAssessmentResult:对关系进行全面伤害评估Args:relationship: 关系对象Returns:HarmAssessmentResult: 评估结果# 1. 计算总伤害指数total_harm self._calculate_total_harm_index(relationship)# 2. 分析伤害模式pattern_analysis self._analyze_patterns(relationship)# 3. 计算情绪成本emotional_cost self._calculate_emotional_cost(relationship, total_harm)# 4. 确定风险等级risk_level self._determine_risk_level(total_harm, pattern_analysis)result HarmAssessmentResult(total_harm_indextotal_harm,pattern_analysispattern_analysis,emotional_costemotional_cost,risk_levelrisk_level)self.assessment_history.append(result)return resultdef _calculate_total_harm_index(self,relationship: models.Relationship) - float:计算综合伤害指数公式: Σ(单次伤害分数) × 关系重要性权重 ÷ (修复记录 1)Args:relationship: 关系对象Returns:float: 伤害指数if not relationship.harm_events:return 0.0# 计算所有伤害事件的总分total_raw_score sum(event.calculate_harm_score(self.config)for event in relationship.harm_events)# 获取修复记录的积极影响repair_bonus self._calculate_repair_bonus(relationship)# 应用关系重要性权重weighted_score total_raw_score * relationship.importance_weight# 考虑原谅尝试次数的负面影响forgiveness_penalty relationship.forgiveness_attempts * 5.0# 最终伤害指数harm_index (weighted_score - repair_bonus forgiveness_penalty)return max(0.0, harm_index)def _calculate_repair_bonus(self,relationship: models.Relationship) - float:计算修复记录的积极减分项有效修复行为可以降低伤害指数if not relationship.repair_actions:return 0.0# 只计算已验证的有效修复verified_actions [a for a in relationship.repair_actionsif a.verified and a.effectiveness_score 0.5]if not verified_actions:return 0.0# 根据有效性和数量计算减分total_effectiveness sum(a.effectiveness_score for a in verified_actions)repair_bonus total_effectiveness * 3.0return min(repair_bonus, 20.0) # 设置上限def _analyze_patterns(self,relationship: models.Relationship) - Dict[str, any]:分析伤害模式识别- 高频伤害类型- 伤害升级趋势- 重复伤害模式- 权力滥用模式if not relationship.harm_events:return {status: no_harm_events}events relationship.harm_events# 1. 按类型统计type_counts {}for event in events:if event.harm_type not in type_counts:type_counts[event.harm_type] 0type_counts[event.harm_type] 1most_common_type max(type_counts, keytype_counts.get) if type_counts else None# 2. 分析时间趋势if len(events) 3:recent_intensities [e.intensity for e in events[-3:]]earlier_intensities [e.intensity for e in events[:-3]] if len(events) 3 else []if earlier_intensities:trend escalating if (statistics.mean(recent_intensities) statistics.mean(earlier_intensities) 1) else stable_or_improvingelse:trend insufficient_dataelse:trend insufficient_data# 3. 检查重复模式repeated_types [t for t, c in type_counts.items() if c 2]# 4. 权力不对等分析avg_power_imbalance statistics.mean([e.power_imbalance for e in events])high_power_events [e for e in events if e.power_imbalance 1.5]return {most_common_harm_type: most_common_type,harm_type_distribution: type_counts,intensity_trend: trend,repeated_harm_types: repeated_types,avg_power_imbalance: round(avg_power_imbalance, 2),high_power_incidents: len(high_power_events),total_incidents: len(events)}def _calculate_emotional_cost(self,relationship: models.Relationship,harm_index: float) - float:计算情绪劳动成本考虑因素- 伤害指数- 关系重要性- 处理时间- 心理影响# 基础情绪成本base_cost harm_index * 0.8# 关系重要性加成importance_multiplier 0.5 (relationship.importance_weight * 0.5)# 处理时间成本假设每次伤害需要3-7天恢复processing_days len(relationship.harm_events) * 5time_cost min(processing_days * 0.5, 15.0)# 计算总情绪成本emotional_cost (base_cost * importance_multiplier) time_costreturn round(emotional_cost, 2)def _determine_risk_level(self,harm_index: float,pattern_analysis: Dict) - str:确定关系风险等级基于伤害指数和模式分析的综合判断if harm_index 10:return LOWelif harm_index 25:if pattern_analysis.get(intensity_trend) escalating:return MEDIUM_HIGHreturn MEDIUMelif harm_index 40:if pattern_analysis.get(repeated_harm_types):return HIGHreturn MEDIUM_HIGHelse:if pattern_analysis.get(high_power_incidents, 0) 0:return CRITICALreturn HIGHdef get_improvement_suggestions(self,assessment: HarmAssessmentResult,relationship: models.Relationship) - List[str]:基于评估结果生成改善建议suggestions []if assessment.risk_level in [HIGH, CRITICAL]:suggestions.append(⚠️ 建议立即加强边界保护限制与该关系的深度接触)if assessment.pattern_analysis.get(intensity_trend) escalating:suggestions.append( 检测到伤害升级趋势建议制定明确的退出策略)if assessment.pattern_analysis.get(high_power_incidents, 0) 0:suggestions.append( 存在权力不对等情况建议寻求第三方支持或专业咨询)if assessment.emotional_cost 30:suggestions.append( 情绪成本过高建议暂停接触并进行自我关怀)if not suggestions:suggestions.append(✅ 当前状况可控继续保持观察和适度边界即可)return suggestions4. 修复验证器 (core/repair_validator.py)修复验证器模块负责验证施害者的修复诚意和行为改变颠覆口头道歉问题解决的错误认知from dataclasses import dataclass, fieldfrom datetime import datetime, timedeltafrom typing import List, Dict, Tuple, Optionalfrom enum import Enumimport statisticsclass RepairStatus(Enum):修复状态枚举INSUFFICIENT insufficient # 修复不足PROGRESSING progressing # 进行中SUBSTANTIAL substantial # 实质性进展COMPLETE complete # 基本完成RELAPSED relapsed # 复发dataclassclass RepairValidationResult:修复验证结果包含验证状态、详细分析和建议status: RepairStatusoverall_score: float # 总体修复分数(0-100)action_analysis: List[Dict] # 各行动分析behavior_change_score: float # 行为改变分数accountability_score: float # 责任承担分数consistency_score: float # 一致性分数validation_time: datetime field(default_factorydatetime.now)recommendations: List[str] field(default_factorylist)def to_dict(self) - Dict:return {status: self.status.value,overall_score: round(self.overall_score, 1),action_analysis: self.action_analysis,behavior_change_score: round(self.behavior_change_score, 1),accountability_score: round(self.accountability_score, 1),consistency_score: round(self.consistency_score, 1),validation_time: self.validation_time.isoformat(),recommendations: self.recommendations}class RepairValidator:修复验证器主类核心验证逻辑1. 修复必须超越口头道歉2. 必须有持续的行为改变证据3. 必须承认具体责任和伤害4. 必须通过时间考验def __init__(self, model_config: config.ModelConfig None):self.config model_config or config.ModelConfig()self.validation_history: List[RepairValidationResult] []def validate_repair_progress(self,relationship: models.Relationship) - RepairValidationResult:验证关系的修复进展Args:relationship: 关系对象Returns:RepairValidationResult: 验证结果actions relationship.repair_actionsif not actions:return self._create_insufficient_result(无修复行动记录)# 1. 分析各修复行动action_analysis self._analyze_actions(actions)# 2. 评估行为改变behavior_score self._evaluate_behavior_change(relationship, actions)# 3. 评估责任承担accountability_score self._evaluate_accountability(actions)# 4. 评估一致性consistency_score self._evaluate_consistency(relationship, actions)# 5. 计算总体分数overall_score self._calculate_overall_score(behavior_score,accountability_score,consistency_score,len(actions))# 6. 确定状态status self._determine_status(overall_score, behavior_score, consistency_score)# 7. 生成建议recommendations self._generate_recommendations(status, overall_score, action_analysis)result RepairValidationResult(statusstatus,overall_scoreoverall_score,action_analysisaction_analysis,behavior_change_scorebehavior_score,accountability_scoreaccountability_score,consistency_scoreconsistency_score,recommendationsrecommendations)self.validation_history.append(result)return resultdef _analyze_actions(self, actions: List[models.RepairAction]) - List[Dict]:分析每个修复行动的质量analysis []for action in actions:# 根据行动类型评估其价值type_value {apology_sincerity: 15,accountability_taking: 25,behavior_change: 30,compensation: 20,prevention_plan: 25,time_proven: 35}.get(action.action_type, 10)# 有效性加成effectiveness_bonus action.effectiveness_score * 20analysis.append({action_id: action.action_id,action_type: action.action_type,type_value: type_value,effectiveness_score: round(action.effectiveness_score, 2),combined_score: round(type_value effectiveness_bonus, 1),verified: action.verified,notes: actio利用AI解决实际问题如果你觉得这个工具好用欢迎关注长安牧笛