AI 扩展计划 / Playbooks
AI Expansion Master Index
这套 AI 扩展不是单纯学习 prompt，也不是堆工具清单。它训练的是企业 AI 时代更稀缺的一组复合能力：
809 行AI_EXPANSION_MASTER_INDEX.md
AI Expansion Master Index
目的：把当前仓库中新增的 AI、架构师、产品经理、BA 学习资产整理成一张长期地图，方便后续持续学习、复习、作品集转换和面试表达。 原则：这是新增索引，不替代旧 Web3、架构 120 天、Solidity/Rust/Move、AIPA、LLM、DSDB 等内容。旧内容保留，新内容只做连接和扩展。
1. 这套扩展到底在训练什么
这套 AI 扩展不是单纯学习 prompt，也不是堆工具清单。它训练的是企业 AI 时代更稀缺的一组复合能力：
| 能力层 | 你要形成的判断 | 典型证据 |
|---|---|---|
| AI 底层逻辑 | 能解释 Transformer、RAG、Agent、MoE、Scaling、RLHF、Eval 为什么影响产品和架构 | 论文一页纸、机制图、面试表达 |
| 业务问题定义 | 能判断一个 AI use case 是否值得做、适不适合做、应该先做哪一段 | Opportunity canvas、stakeholder map、business case |
| BA 需求工程 | 能把模糊需求转成流程、规则、异常、验收标准和 eval contract | BPMN、requirements-to-eval、data readiness |
| AI 产品判断 | 能定义 MVP、pilot、指标、adoption、ROI、stop rule | PRD、pilot memo、adoption dashboard |
| AI 架构设计 | 能设计 RAG、tool gateway、agent workflow、context system、security、observability | C4、ADR、sequence、architecture gate |
| EvalOps / RiskOps | 能证明系统在上线前后可评测、可监控、可回滚、可审计 | golden set、rubric、release gate、incident runbook |
| 组织落地 | 能设计 RACI、运营机制、vendor/build-buy、培训、变更管理 | operating model、vendor memo、executive memo |
| 面试和作品集 | 能把复杂案例讲成 BA、PM、Architect 三种语言 | portfolio evidence map、storyline pack |
最终目标不是“我学过 AI”，而是能在金融、零售、Web3 或企业软件场景中说清楚：
- 为什么这个问题适合或不适合 AI。
- 需要什么数据、知识、流程、控制和组织能力。
- 怎样从 discovery 走到 pilot，再走到 release / scale。
- 怎样证明它有效、可靠、合规、值得继续投资。
2. 推荐阅读入口
如果只能从一个文件开始，先读：
docs/AI_2026_EXPANSION_START_HERE.md
如果想看完整计划，再读：
docs/AI_BA_PRODUCT_ARCHITECT_180_PLAN.mddocs/AI_NEW_DEMANDS_2026_EXPANSION.mddocs/AI_FOUNDATIONS_CLASSIC_PAPERS_PLAN.md
如果想直接做作品集，先读：
docs/AI_BA_PM_PRACTICE_LAB.mddocs/FINANCIAL_RETAIL_AI_CASE_PORTFOLIO.mddocs/abpa/capstone-aml/AML_30_DAY_DEEPENING_PLAN.md
如果想准备面试，先读：
docs/AI_ROLE_COMPETENCY_MATRIX_2026.mddocs/AI_INTERVIEW_PORTFOLIO_STORYLINE_PLAYBOOK.mddocs/abpa/interview/AI_BA_PM_ARCHITECT_INTERVIEW_BANK.md
网站可见规则：
- 所有新增学习笔记、论文解读和 playbook 都要注册到
/papers使用的src/data/papers.ts。 - 网站会按原有卡片格式展示，并通过同一套数据进入搜索索引和 sitemap。
- 新增嵌套目录笔记时，优先使用
docs/.../*.md，再用稳定 slug 暴露为/papers/...。
3. 资产分层地图
3.1 战略与需求层
| 文件 | 适合解决的问题 | 输出 |
|---|---|---|
docs/AI_NEW_DEMANDS_2026_EXPANSION.md | 企业 AI 新需求到底变成了什么 | capability map、能力缺口 |
docs/AI_BA_PRODUCT_ARCHITECT_180_PLAN.md | 如何用 180 天系统训练 AI BA/PM/Architect | 长期路线、阶段目标 |
docs/AI_ROLE_COMPETENCY_MATRIX_2026.md | AI BA、AI PM、AI Architect、FDE、EvalOps 差异是什么 | 角色能力矩阵、证据标准 |
docs/AI_REQUIREMENTS_TO_EVAL_COOKBOOK.md | 如何把需求写成可测试、可发布、可监控 | eval contract、golden set、release gate |
docs/AI_VENDOR_BUILD_BUY_ADOPTION_PLAYBOOK.md | 买、建、合作、混合方案如何决策 | vendor scorecard、build-buy memo、pilot gate |
docs/AI_PLATFORM_PM_PLAYBOOK.md | 如何把 AI 能力平台化，而不是每个团队重复做 POC | model gateway、RAG platform、EvalOps、成本、权限、adoption |
docs/AI_CAPABILITY_ASSESSMENT_RUBRIC.md | 如何判断学习是否已经变成能力证据 | C1-C14 能力评分、角色画像、evidence map、30 天自评节奏 |
3.2 底层论文与技术机制层
| 文件 | 核心机制 | 对 PM/BA/架构师的价值 |
|---|---|---|
docs/AI_FOUNDATIONS_CLASSIC_PAPERS_PLAN.md | 经典论文路线 | 避免只会工具，不懂底层 |
docs/ai-foundations/README.md | 论文索引 | 管理长期精读顺序 |
docs/ai-foundations/papers/01-attention-is-all-you-need.md | Transformer / attention | 解释 LLM 为什么能处理上下文 |
docs/ai-foundations/papers/02-retrieval-augmented-generation.md | RAG | 解释企业知识治理和引用 |
docs/ai-foundations/papers/03-react-toolformer-agent-foundations.md | Agent / tools | 解释从回答到行动的系统边界 |
docs/ai-foundations/papers/04-instructgpt-rlhf-alignment.md | RLHF / alignment | 解释 helpful/honest/harmless 与产品控制 |
docs/ai-foundations/papers/05-chain-of-thought-self-consistency.md | CoT / self-consistency | 解释推理产品与证据边界 |
docs/ai-foundations/papers/06-lora-peft-adaptation.md | LoRA / PEFT | 解释 RAG、微调、adapter 治理取舍 |
docs/ai-foundations/papers/07-inference-optimization-kv-cache-flashattention-speculative.md | inference optimization | 解释延迟、成本、吞吐和 SLO |
docs/ai-foundations/papers/08-llm-as-judge-evaluation.md | LLM-as-Judge | 解释自动评测、rubric、human review 边界 |
docs/ai-foundations/papers/09-mixture-of-experts-sparse-scaling.md | MoE / sparse scaling | 解释稀疏激活、router、expert capacity、load balancing 与企业成本/SLO |
docs/ai-foundations/papers/10-scaling-laws-pretraining-bert-gpt-t5.md | Scaling laws / pretraining | 解释 BERT、GPT、T5、预训练目标、RAG/微调/自训模型取舍 |
docs/ai-foundations/papers/11-dpo-constitutional-ai-preference-optimization.md | DPO / Constitutional AI | 解释 preference data、DPO、RLAIF、reward hacking、alignment tax 和行为治理 |
docs/ai-foundations/papers/12-tool-use-security-prompt-injection.md | Tool use security | 解释间接 prompt injection、data exfiltration、confused deputy、tool gateway 和 kill switch |
docs/ai-foundations/papers/13-rag-evaluation-ragas-retrieval-metrics.md | RAG evaluation / RAGAS | 解释 retrieval metrics、faithfulness、citation support、permission/version tests 和 release gate |
docs/ai-foundations/papers/14-graphrag-knowledge-graph-rag.md | GraphRAG / Knowledge Graph RAG | 解释 entity、relation、path、community summary、多跳推理和 graph eval |
docs/ai-foundations/papers/15-generative-agents-memory-reflection-planning.md | Generative Agents / Memory | 解释 memory stream、reflection、planning、state boundary、retention/deletion 和 memory eval |
docs/ai-foundations/papers/16-autogen-multi-agent-orchestration.md | AutoGen / Multi-Agent | 解释 multi-agent conversation、role charter、handoff contract、shared state、HITL 和 orchestration eval |
docs/ai-foundations/papers/17-helm-holistic-evaluation-models.md | HELM / holistic evaluation | 解释 scenario、metric、adapter、透明评测、模型选择、leaderboard 边界和 release gate |
docs/ai-foundations/papers/18-model-cards-datasheets-ai-documentation.md | Model Cards / Datasheets | 解释模型卡、数据集说明书、适用边界、风险披露、证据链和 AI governance 文档化 |
docs/ai-foundations/papers/19-tree-of-thoughts-planning-search.md | Tree of Thoughts / Planning Search | 解释多路径 thought、search controller、evaluator、搜索预算、人工选择点和复杂任务规划 |
docs/ai-foundations/papers/20-self-rag-crag-agentic-retrieval.md | Self-RAG / CRAG / Agentic Retrieval | 解释 retrieval need、context quality gate、纠错检索、refuse/escalate 和 RAG control plane |
docs/ai-foundations/papers/21-agentbench-taubench-agent-evaluation.md | AgentBench / τ-bench / Agent Eval | 解释 tool-agent-user interaction、tool sandbox、policy oracle、trace 和 agent release gate |
docs/ai-foundations/papers/22-mechanistic-interpretability-transformer-circuits-sae.md | Mechanistic Interpretability / SAE | 解释 Transformer circuits、sparse autoencoder、解释性证据、模型风险和治理边界 |
docs/ai-foundations/papers/23-long-context-lost-in-the-middle-ruler.md | Long Context / Lost in the Middle / RULER | 解释标称上下文长度、有效上下文长度、position robustness、长上下文/RAG 混合架构 |
docs/ai-foundations/papers/24-dspy-opro-automatic-prompt-optimization.md | DSPy / OPRO / Prompt Optimization | 解释 LM program、task signature、prompt registry、optimizer、eval-driven prompt release |
docs/ai-foundations/papers/25-reflexion-self-refine-agent-feedback-loops.md | Reflexion / Self-Refine / Feedback Loops | 解释 feedback object、refinement policy、reflection memory、agent 改进循环和停止条件 |
docs/ai-foundations/papers/26-process-supervision-step-by-step-verification.md | Process Supervision / Step Verification | 解释 outcome vs process supervision、step-level verifier、critical step gate 和审计证据 |
docs/ai-foundations/papers/27-structured-output-constrained-decoding-lmql-guidance.md | Structured Output / Constrained Decoding | 解释 schema contract、约束解码、LM program、validator、tool payload 和结构化输出治理 |
docs/ai-foundations/papers/28-model-routing-semantic-cache-frugal-ai.md | Model Routing / Semantic Cache / Frugal AI | 解释 FrugalGPT、RouteLLM、semantic cache、cascade、成本/质量/SLO 路由 |
docs/ai-foundations/papers/29-swe-bench-webarena-agent-benchmarks.md | SWE-bench / WebArena / OSWorld / GAIA | 解释真实环境 Agent benchmark、tool sandbox、policy oracle、state verifier 和 release gate |
docs/ai-foundations/papers/30-mamba-state-space-models-efficient-sequence.md | Mamba / State Space Models | 解释 S4、Mamba、selective state space、长序列效率和模型架构 frontier |
docs/ai-foundations/papers/31-embeddings-ann-vector-search-faiss-hnsw.md | Embeddings / ANN / Vector Search | 解释 Word2Vec、Sentence-BERT、FAISS、HNSW、bi-encoder、rerank 和 RAG 检索底座 |
docs/ai-foundations/papers/32-clip-multimodal-embeddings-product-architecture.md | CLIP / Multimodal Embeddings | 解释图文对比学习、zero-shot、多模态检索、文搜图/图搜图和多模态产品架构 |
docs/ai-foundations/papers/33-diffusion-latent-diffusion-generative-media.md | Diffusion / Latent Diffusion | 解释 DDPM、classifier-free guidance、Latent Diffusion、生成式媒体和品牌/版权/安全治理 |
docs/ai-foundations/papers/34-graph-neural-networks-gnn-fraud-risk-architecture.md | Graph Neural Networks / GNN | 解释 GCN、GraphSAGE、GAT、图学习、欺诈/AML 网络和风险架构 |
docs/ai-foundations/papers/35-recommender-systems-youtube-wide-deep-two-tower.md | Recommender Systems | 解释候选生成、召回、排序、重排、Two-Tower、Wide & Deep、YouTube DNN 和 next-best-action |
docs/ai-foundations/papers/36-learning-to-rank-lambdamart-neural-ranking.md | Learning to Rank / LambdaMART | 解释 pointwise、pairwise、listwise、NDCG、query group、搜索/推荐/告警排序和 neural reranking |
docs/ai-foundations/papers/37-feature-stores-real-time-ml-feast-michelangelo.md | Feature Stores / Real-Time ML | 解释 Feast、Michelangelo、offline/online store、point-in-time correctness、freshness SLO 和实时决策 |
docs/ai-foundations/papers/38-zanzibar-cedar-opa-authorization-policy-architecture.md | Zanzibar / Cedar / OPA Authorization | 解释 RBAC、ABAC、ReBAC、policy-as-code、PDP/PEP、RAG 权限过滤和 Agent 工具授权 |
docs/ai-foundations/papers/39-federated-learning-fedavg-cross-silo-ai.md | Federated Learning / FedAvg | 解释 cross-silo AI、local training、secure aggregation、参与方治理、跨机构风险协作和模型治理 |
docs/ai-foundations/papers/40-differential-privacy-dpsgd-ai-data-protection.md | Differential Privacy / DP-SGD | 解释 epsilon、delta、privacy budget、gradient clipping、noise addition、隐私-效用-公平取舍 |
docs/ai-foundations/papers/41-knowledge-distillation-small-models-quantization.md | Knowledge Distillation / Small Models | 解释 teacher-student、soft labels、DistilBERT、量化、小模型路由和模型组合策略 |
docs/ai-foundations/papers/42-durable-execution-agent-workflow-state-machines.md | Durable Execution / Agent Workflow | 解释 durable workflow、state machine、Saga、幂等、HITL 状态、workflow replay 和 Agent 生产架构 |
docs/ai-foundations/papers/43-digital-twin-agent-based-simulation-ai-decisioning.md | Digital Twin / Agent-Based Simulation | 解释 process twin、decision twin、agent-based simulation、校准、验证、敏感性分析和 AI 决策仿真 |
docs/ai-foundations/papers/44-online-experimentation-cuped-release-science-ai-products.md | Online Experimentation / CUPED | 解释 online controlled experiments、CUPED、guardrails、shadow launch、ramp 和 AI release science |
docs/ai-foundations/papers/45-data-lineage-contracts-openlineage-ai-data-quality.md | Data Lineage / Data Contracts | 解释 OpenLineage、DataHub、OpenMetadata、Great Expectations、AI data contract、quality SLO 和 lineage |
docs/ai-foundations/papers/46-ai-security-operations-mitre-atlas-owasp-csf.md | AI Security Operations | 解释 MITRE ATLAS、OWASP LLM Top 10、NIST CSF、AI telemetry、detection engineering 和 SOC runbook |
docs/ai-foundations/papers/47-time-series-forecasting-tft-deepar-foundation-models.md | Time-Series Forecasting | 解释 DeepAR、TFT、TimesFM、概率预测、预测区间、层级预测、rolling backtest 和 forecast-to-decision |
docs/ai-foundations/papers/48-anomaly-detection-isolation-forest-autoencoder-risk-monitoring.md | Anomaly Detection / Risk Monitoring | 解释 Isolation Forest、autoencoder、SPC、streaming anomaly detection、阈值校准、告警疲劳和反馈闭环 |
docs/ai-foundations/papers/49-causal-discovery-dowhy-econml-structural-causal-models.md | Causal Discovery / SCM | 解释 DAG、SCM、DoWhy、EconML、DAGitty、NOTEARS、混杂、可识别性和产品干预设计 |
docs/ai-foundations/papers/50-optimization-operations-research-or-tools-ai-decisioning.md | Optimization / Operations Research | 解释 LP/MIP、CP-SAT、OR-Tools、Gurobi、PuLP、目标/约束、多目标权衡和 solver decision service |
docs/ai-foundations/papers/51-contextual-bandits-linucb-thompson-online-learning.md | Contextual Bandits / Online Learning | 解释 LinUCB、Thompson Sampling、epsilon-greedy、propensity logging、IPS/DR 和 adaptive experimentation |
docs/ai-foundations/papers/52-reinforcement-learning-offline-rl-cql-policy-decisioning.md | Reinforcement Learning / Offline RL | 解释 MDP、policy、reward、DQN、CQL、offline RL、reward hacking、safe exploration 和策略治理 |
docs/ai-foundations/papers/53-bayesian-optimization-botorch-optuna-experiment-design.md | Bayesian Optimization / Experiment Design | 解释 GP surrogate、acquisition functions、BoTorch、Optuna、multi-objective/constrained BO 和实验预算 |
docs/ai-foundations/papers/54-calibration-conformal-prediction-uncertainty-governance.md | Calibration / Conformal Prediction | 解释 ECE、reliability diagram、Brier score、temperature scaling、coverage、abstention 和不确定性治理 |
docs/ai-foundations/papers/55-data-centric-ai-snorkel-programmatic-labeling.md | Data-Centric AI / Programmatic Labeling | 解释 Snorkel、weak supervision、labeling functions、label model、coverage/conflict 和标签治理 |
docs/ai-foundations/papers/56-active-learning-human-in-the-loop-labeling.md | Active Learning / HITL Labeling | 解释 uncertainty sampling、query-by-committee、SME review、label budget、feedback loop 和人工反馈运营 |
docs/ai-foundations/papers/57-dataset-shift-monitoring-model-performance.md | Dataset Shift / Model Monitoring | 解释 covariate shift、label shift、concept drift、training-serving skew、outcome lag 和漂移处置 |
docs/ai-foundations/papers/58-ai-management-system-iso42001-operating-model.md | AI Management System / ISO 42001 | 解释 AIMS、NIST AI RMF、AI inventory、risk tier、release gate、control library 和 operating model |
docs/ai-foundations/papers/59-hidden-technical-debt-ml-systems-ai-architecture.md | Hidden Technical Debt / AI Architecture Debt | 解释 CACE、entanglement、boundary erosion、hidden feedback loops、undeclared consumers 和配置/数据债 |
docs/ai-foundations/papers/60-cd4ml-mlops-continuous-delivery-ai-release.md | CD4ML / MLOps / AI Release Engineering | 解释 CI/CD/CT、release bundle、model/data/prompt registry、shadow/canary/ramp、rollback 和证据链 |
docs/ai-foundations/papers/61-human-ai-interaction-guidelines-product-design.md | Human-AI Interaction / Product Design | 解释 HAI guidelines、calibrated trust、automation bias、recoverability、feedback 和 escalation UX |
docs/ai-foundations/papers/62-ai-architecture-decision-records-governance.md | AI ADR / Decision Governance | 解释 ADR、architecture knowledge、risk tier、evidence link、reversal trigger 和 AI 决策治理 |
docs/ai-foundations/papers/63-ai-requirements-engineering-gqm-eval-contracts.md | AI Requirements Engineering / GQM / Eval Contracts | 解释 GQM、AI requirement taxonomy、eval contract、golden set、release gate、monitoring gate 和风险 owner |
docs/ai-foundations/papers/64-ai-quality-attributes-atam-architecture-tradeoff.md | AI Quality Attributes / ATAM / Architecture Tradeoff | 解释 quality attribute scenario、utility tree、tradeoff point、sensitivity point 和 AI 架构评审 |
docs/ai-foundations/papers/65-ai-safety-engineering-stpa-control-structure.md | AI Safety Engineering / STPA / Control Structure | 解释 loss、hazard、unsafe control action、control structure、safety constraint 和 agent 安全工程 |
docs/ai-foundations/papers/66-sociotechnical-ai-resilience-work-as-done.md | Sociotechnical AI / Resilience / Work-as-Done | 解释 work-as-imagined、work-as-done、human-AI collaboration、handoff、load 和 operating model |
docs/ai-foundations/papers/67-ai-capability-based-planning-business-architecture.md | AI Capability-Based Planning / Business Architecture | 解释 capability map、value stream、maturity、portfolio prioritization、architecture runway 和 funding gate |
docs/ai-foundations/papers/68-wardley-mapping-ai-product-platform-strategy.md | Wardley Mapping / AI Product & Platform Strategy | 解释 user need、value chain、evolution axis、build-buy-partner、platform boundary 和战略地图 |
docs/ai-foundations/papers/69-conway-team-topologies-ai-platform-operating-model.md | Conway's Law / Team Topologies / AI Platform Operating Model | 解释组织沟通结构、team topology、cognitive load、team API 和 AI 平台 operating model |
docs/ai-foundations/papers/70-dora-space-ai-sdlc-engineering-productivity.md | DORA / SPACE / AI SDLC Engineering Productivity | 解释 DORA、SPACE、AI code agent governance、AI-assisted PR gate、工程生产力和安全发布 |
docs/ai-foundations/papers/71-continuous-discovery-opportunity-solution-tree-ai-products.md | Continuous Discovery / Opportunity Solution Tree for AI Products | 解释 OST、AI opportunity taxonomy、assumption map、prototype/eval/pilot learning loop 和 discovery-to-delivery 追踪 |
docs/ai-foundations/papers/72-jtbd-outcome-driven-innovation-ai-use-case-selection.md | JTBD / Outcome-Driven Innovation for AI Use Case Selection | 解释 job map、outcome statements、underserved outcome、opportunity score、AI fit 和 automation boundary |
docs/ai-foundations/papers/73-north-star-ai-product-metrics-value-measurement.md | North Star Metrics / AI Product Value Measurement | 解释 AI metrics tree、guardrail metrics、risk-adjusted value、因果证据和收益兑现 |
docs/ai-foundations/papers/74-ai-product-operating-model-empowered-teams.md | AI Product Operating Model / Empowered Teams | 解释 AI product trio+、decision rights、discovery-delivery-governance cadence 和授权团队 guardrails |
docs/ai-foundations/papers/75-ai-portfolio-management-funding-governance.md | AI Portfolio Management / Funding Governance | 解释 AI portfolio kanban、funding gate、capacity allocation、risk-adjusted value 和 scale/stop 决策 |
docs/ai-foundations/papers/76-service-blueprint-ai-customer-journey-trust.md | AI Service Blueprint / Customer Journey / Trust | 解释 frontstage/backstage、trust moments、human handoff、correction、appeal 和服务证据 |
docs/ai-foundations/papers/77-ai-business-process-reengineering-bpmn-dmn.md | AI Business Process Reengineering / BPMN / DMN | 解释 AI BPR、BPMN、DMN、decision automation、eval traceability 和 control evidence |
docs/ai-foundations/papers/78-ai-risk-appetite-policy-product-management.md | AI Risk Appetite / Policy Product Management | 解释 risk appetite、policy lifecycle、risk-tiered guardrails、runtime control 和 stop rule |
docs/ai-foundations/papers/79-enterprise-ai-reference-architecture-control-plane.md | Enterprise AI Reference Architecture / Control Plane | 解释企业 AI 八层参考架构、control plane、model/tool gateway、eval/observability 和 evidence plane |
docs/ai-foundations/papers/80-ai-product-line-engineering-reuse-platform-assets.md | AI Product Line Engineering / Reusable Platform Assets | 解释 core assets、variation points、domain/application engineering、platform reuse 和 asset governance |
docs/ai-foundations/papers/81-ai-maturity-model-roadmap-capability-assessment.md | AI Maturity Model / Roadmap / Capability Assessment | 解释 capability domains、maturity levels、evidence standard、roadmap sequencing 和能力评估 |
docs/ai-foundations/papers/82-ai-control-library-assurance-evidence-graph.md | AI Control Library / Assurance Evidence Graph | 解释 control library、assurance case、claim-risk-control-evidence graph 和 release assurance |
docs/ai-foundations/papers/83-ai-domain-driven-design-ubiquitous-language.md | AI Domain-Driven Design / Ubiquitous Language | 解释 bounded context、ubiquitous language、RAG boundary、context map 和 eval vocabulary |
docs/ai-foundations/papers/84-event-storming-agent-workflow-design.md | AI EventStorming / Agent Workflow Discovery | 解释 domain event、command、policy、actor、tool call、HITL、compensation 和 workflow trace |
docs/ai-foundations/papers/85-ai-knowledge-work-redesign-role-task-architecture.md | AI Knowledge Work Redesign / Role-Task Architecture | 解释 role-task decomposition、human-AI teaming、accountability、override 和 workload-risk metrics |
docs/ai-foundations/papers/86-ai-platform-service-catalog-golden-paths.md | AI Platform Service Catalog / Golden Paths | 解释 service catalog、golden paths、self-service with guardrails、platform metrics 和 adoption |
docs/ai-foundations/papers/87-ai-architecture-views-c4-arc42-42010.md | AI Architecture Views / C4 / arc42 / 42010 | 解释 stakeholder concern、viewpoint、C4/arc42 架构表达、AI control/evidence view |
docs/ai-foundations/papers/88-ai-architecture-fitness-functions-continuous-governance.md | AI Architecture Fitness Functions / Continuous Governance | 解释把质量属性、eval、SLO、策略和证据转成持续可检查架构约束 |
docs/ai-foundations/papers/89-contract-first-ai-tool-api-design-openapi-asyncapi.md | Contract-First AI Tool/API Design | 解释 OpenAPI、AsyncAPI、JSON Schema、tool/event contract、side effect、approval 和 audit |
docs/ai-foundations/papers/90-ai-traceability-requirements-eval-control-graph.md | AI Traceability Graph | 解释 business outcome、requirement、eval、control、ADR、telemetry 和 evidence 的追踪图 |
docs/ai-foundations/papers/91-ai-enterprise-architecture-togaf-archimate-adm.md | AI Enterprise Architecture / TOGAF / ArchiMate / ADM | 解释 TOGAF ADM、ArchiMate、architecture repository 和 AI governance 如何支撑能力演进 |
docs/ai-foundations/papers/92-banking-ai-reference-models-bian-fibo-iso20022.md | Banking AI Reference Models / BIAN / FIBO / ISO 20022 | 解释银行能力边界、金融本体和消息语义如何支撑 AI 需求、RAG、工具和审计 |
docs/ai-foundations/papers/93-ai-semantic-interoperability-rdf-owl-shacl.md | AI Semantic Interoperability / RDF / OWL / SHACL | 解释 semantic contract、ontology slice、SHACL 约束和语义评测如何减少 AI 语义风险 |
docs/ai-foundations/papers/94-ai-value-stream-management-flow-metrics.md | AI Value Stream Management / Flow Metrics | 解释 AI 从 idea 到 safe release、adoption 和 value realization 的 flow management |
docs/ai-foundations/papers/95-ai-regulatory-architecture-eu-ai-act-nist-iso42001.md | AI Regulatory Architecture / EU AI Act / NIST / ISO42001 | 解释法规、框架、管理体系如何转成 AI inventory、risk tier、control、gate 和 evidence architecture |
docs/ai-foundations/papers/96-ai-model-risk-validation-independent-challenge.md | AI Model Risk Validation / Independent Challenge | 解释 GenAI system validation、independent challenge、validation evidence 和 revalidation trigger |
docs/ai-foundations/papers/97-ai-third-party-vendor-contract-exit-architecture.md | AI Third-Party Vendor Contract / Exit Architecture | 解释 AI vendor 采购、合同控制、model update notice、audit rights 和 exit architecture |
docs/ai-foundations/papers/98-ai-data-lifecycle-governance-provenance-retention.md | AI Data Lifecycle Governance / Provenance / Retention | 解释 AI 数据来源、用途、保留、删除、血缘和审计证据如何支撑生产级 AI |
docs/ai-foundations/papers/99-ai-agent-autonomy-delegation-architecture.md | AI Agent Autonomy / Delegation Architecture | 解释 agent 自主权如何被拆成委派边界、工具权限、人工升级、kill switch 和 evidence |
docs/ai-foundations/papers/100-ai-agent-identity-delegated-authorization.md | AI Agent Identity / Delegated Authorization | 解释 agent 身份、on-behalf-of、OAuth token exchange、scope、consent 和 audit claims |
docs/ai-foundations/papers/101-ai-runtime-evidence-observability-architecture.md | AI Runtime Evidence / Observability Architecture | 解释 prompt、RAG、tool、policy、approval、output、feedback 和 incident 的运行证据链 |
docs/ai-foundations/papers/102-ai-portfolio-systemic-risk-dependency-architecture.md | AI Portfolio Systemic Risk / Dependency Architecture | 解释 shared model/vendor/data/tool/HITL/evidence 依赖如何形成组合级系统性风险 |
docs/ai-foundations/papers/103-ai-customer-harm-redress-recovery-architecture.md | AI Customer Harm / Redress / Recovery Architecture | 解释客户伤害、投诉/申诉、救济、补偿、恢复和防复发如何成为 AI 产品控制 |
docs/ai-foundations/papers/104-ai-fairness-fair-lending-bias-control-architecture.md | AI Fairness / Fair Lending / Bias Control Architecture | 解释公平信贷、proxy 风险、segment eval、human review calibration 和 bias evidence binder |
docs/ai-foundations/papers/105-ai-explainability-contestability-adverse-action-architecture.md | AI Explainability / Contestability / Adverse Action Architecture | 解释 reason code、adverse action、用户可争议路径、申诉证据和可解释决策接口 |
docs/ai-foundations/papers/106-ai-change-impact-release-governance.md | AI Change Impact / Release Governance | 解释 model/prompt/RAG/tool/policy/eval/vendor/workflow 变更如何进入 impact graph 和 release gate |
docs/ai-foundations/papers/107-ai-continuous-control-monitoring-assurance-architecture.md | AI Continuous Control Monitoring / Assurance Architecture | 解释 control test、exception、KRI、sampling、management action 和 control effectiveness 如何持续运行 |
docs/ai-foundations/papers/108-ai-operational-resilience-bcp-degraded-mode-architecture.md | AI Operational Resilience / BCP / Degraded Mode Architecture | 解释 AI 关键操作在模型/RAG/工具/身份/供应商/HITL 降级时如何保持安全连续 |
docs/ai-foundations/papers/109-ai-management-information-board-reporting-architecture.md | AI Management Information / Board Reporting Architecture | 解释 AI telemetry、价值、风险、控制、客户伤害和集中度如何转成有 lineage 的 MI |
docs/ai-foundations/papers/110-ai-closed-loop-learning-corrective-action-architecture.md | AI Closed-Loop Learning / Corrective Action Architecture | 解释反馈、投诉、人工覆盖、eval 失败、漂移和审计发现如何转成 CAPA 闭环 |
docs/ai-foundations/papers/111-ai-regulatory-horizon-obligation-intelligence-architecture.md | AI Regulatory Horizon / Obligation Intelligence Architecture | 解释法律、监管指引、监督重点和标准如何转成 obligation-to-control/eval/change 情报系统 |
docs/ai-foundations/papers/112-ai-exception-risk-acceptance-waiver-architecture.md | AI Exception / Risk Acceptance / Waiver Architecture | 解释策略例外、临时豁免、残余风险接受、补偿控制和 hard stop 如何治理 |
docs/ai-foundations/papers/113-ai-supply-chain-ai-bom-provenance-architecture.md | AI Supply Chain / AI BOM / Provenance Architecture | 解释模型、数据、RAG、prompt、tool、MCP、eval、人审和 telemetry 如何进入 AI BOM |
docs/ai-foundations/papers/114-ai-human-review-operations-capacity-architecture.md | AI Human Review Operations / Capacity Architecture | 解释人审队列、技能路由、容量、校准、质量、升级和 surge mode 如何成为运营架构 |
docs/ai-foundations/papers/115-ai-segregation-of-duties-dual-control-architecture.md | AI Segregation of Duties / Dual Control Architecture | 解释 maker-checker、four-eyes、incompatible duties、approval-before-action 和审计证据如何约束 AI 工作流 |
docs/ai-foundations/papers/116-ai-consent-preference-purpose-bound-data-architecture.md | AI Consent / Preference / Purpose-Bound Data Architecture | 解释同意、偏好、目的限制、撤回、重新同意和 runtime enforcement 如何控制 AI 数据使用 |
docs/ai-foundations/papers/117-ai-shadow-ai-citizen-development-governance-architecture.md | AI Shadow AI / Citizen Development Governance Architecture | 解释未授权 AI 使用和公民开发如何转成发现、分级、批准路径和平台迁移 |
docs/ai-foundations/papers/118-ai-conduct-risk-suitability-sales-guardrails-architecture.md | AI Conduct Risk / Suitability / Sales Guardrails Architecture | 解释适当性、销售行为、approved claims、offer guardrails、监控和投诉整改如何约束金融 AI |
docs/ai-foundations/papers/119-ai-records-retention-legal-hold-ediscovery-architecture.md | AI Records / Retention / Legal Hold / eDiscovery Architecture | 解释 prompt、RAG、tool、approval、output、eval、incident 记录如何进入留存、法律保全和调取 |
docs/ai-foundations/papers/120-ai-data-residency-cross-border-sovereign-architecture.md | AI Data Residency / Cross-Border / Sovereign AI Architecture | 解释地域、司法辖区、供应商、模型路由、日志、密钥和 transfer review 如何约束 AI 数据路径 |
docs/ai-foundations/papers/121-ai-customer-communications-regulated-content-lifecycle.md | AI Customer Communications / Regulated Content Lifecycle Architecture | 解释客户沟通内容如何经过 approved claims、pre-use review、surveillance、disclosure 和 complaint linkage |
docs/ai-foundations/papers/122-ai-financial-crime-typology-scenario-coverage-architecture.md | AI Financial Crime Typology / Scenario Coverage Architecture | 解释 AML typology、red flag、scenario coverage、SAR evidence bundle 和 alert-to-SAR traceability 如何治理 |
docs/ai-foundations/papers/123-ai-intellectual-property-content-rights-provenance-architecture.md | AI Intellectual Property / Content Rights / Provenance Architecture | 解释输入权利、RAG 语料许可、生成内容、C2PA provenance、rights clearance 和 takedown 如何治理 |
docs/ai-foundations/papers/124-ai-deepfake-synthetic-identity-authentication-fraud-architecture.md | AI Deepfake / Synthetic Identity / Authentication Fraud Architecture | 解释 deepfake、synthetic identity、liveness/PAD、step-up authentication、fraud evidence 和客户摩擦如何联动 |
docs/ai-foundations/papers/125-ai-workforce-hr-decision-employee-monitoring-governance-architecture.md | AI Workforce / HR Decision / Employee Monitoring Governance Architecture | 解释招聘、排班、绩效、员工监控、adverse impact、worker data minimization 和 human review 如何治理 |
docs/ai-foundations/papers/126-ai-incident-disclosure-liability-risk-transfer-architecture.md | AI Incident Disclosure / Liability / Risk Transfer Architecture | 解释 AI incident、materiality triage、通知、责任边界、保险映射、供应商赔偿和损失量化如何组织 |
docs/ai-foundations/papers/127-ai-post-quantum-cryptographic-agility-ai-architecture.md | AI Post-Quantum / Cryptographic Agility Architecture | 解释 AI 系统里的长期证据、RAG、工具、签名、密钥、证书和供应商如何进入 PQC 迁移与密码敏捷架构 |
docs/ai-foundations/papers/128-ai-authorized-push-payment-scam-intervention-architecture.md | AI Authorized Push Payment / Scam Intervention Architecture | 解释授权支付诈骗、社工诱导、收款人风险、客户意图、step-up friction 和救济证据如何组织 |
docs/ai-foundations/papers/129-ai-agent-marketplace-tool-certification-governance-architecture.md | AI Agent Marketplace / Tool Certification Governance Architecture | 解释内部 agent/tool marketplace 如何通过 capability card、认证、权限、签名包、监控和退出治理降低 agentic risk |
docs/ai-foundations/papers/130-ai-customer-vulnerability-accessibility-inclusive-ai-architecture.md | AI Customer Vulnerability / Accessibility / Inclusive AI Architecture | 解释弱势客户、可访问性、包容性 UX、plain language、人工升级、投诉和模型风险如何成为金融 AI 产品护栏 |
docs/ai-foundations/papers/131-ai-payment-dispute-chargeback-claims-evidence-architecture.md | AI Payment Dispute / Chargeback / Claims Evidence Architecture | 解释支付争议、拒付、EFT error claim、billing error、证据、SLA、临时贷记和客户沟通如何组织 |
docs/ai-foundations/papers/132-ai-collections-hardship-delinquency-treatment-architecture.md | AI Collections / Hardship / Delinquency Treatment Architecture | 解释逾期预测、困难客户处理、联系策略、可访问渠道、投诉链接和 fair treatment 如何治理 |
docs/ai-foundations/papers/133-ai-voice-ai-contact-center-agent-assist-governance-architecture.md | AI Voice AI / Contact Center / Agent Assist Governance Architecture | 解释 voice bot、实时转写、坐席辅助、call summary、QA、披露边界和投诉证据如何治理 |
docs/ai-foundations/papers/134-ai-digital-identity-wallet-verifiable-credentials-trust-architecture.md | AI Digital Identity Wallet / Verifiable Credentials / Trust Architecture | 解释 digital wallet、VC、DID、WebAuthn、selective disclosure、revocation 和 trust policy 如何支撑 AI 身份信任 |
docs/ai-foundations/papers/135-ai-open-banking-open-finance-consented-data-sharing-architecture.md | AI Open Banking / Open Finance / Consented Data Sharing Architecture | 解释开放银行/开放金融、客户授权、数据最小化、撤回、API 契约、第三方风险和 AI 使用边界如何治理 |
docs/ai-foundations/papers/136-ai-personalized-pricing-offer-decisioning-governance-architecture.md | AI Personalized Pricing / Offer Decisioning Governance Architecture | 解释个性化价格、费率、额度、offer、实验、解释、投诉和 surveillance pricing 风险如何治理 |
docs/ai-foundations/papers/137-ai-document-intelligence-unstructured-data-evidence-quality-architecture.md | AI Document Intelligence / Unstructured Data / Evidence Quality Architecture | 解释 OCR、layout、字段抽取、置信度、人工复核、记录留存、篡改检测和 workflow evidence 如何组织 |
docs/ai-foundations/papers/138-ai-privacy-clean-room-data-collaboration-measurement-architecture.md | AI Privacy Clean Room / Data Collaboration / Measurement Architecture | 解释 clean room、PEC、聚合、差分隐私、合成数据、合作方测量、输出审查和目的限制如何成为数据协作产品 |
docs/ai-foundations/papers/139-ai-credit-lifecycle-underwriting-line-management-governance-architecture.md | AI Credit Lifecycle / Underwriting / Line Management Governance Architecture | 解释授信、审批、额度增减、账户管理、adverse action、fair lending、组合监控和投诉证据如何治理 |
docs/ai-foundations/papers/140-ai-wealth-advice-robo-advisor-best-interest-boundary-architecture.md | AI Wealth Advice / Robo-Advisor / Best Interest Boundary Architecture | 解释财富建议、robo-advisor、教育/建议/执行边界、风险画像、人工升级和监督证据如何组织 |
docs/ai-foundations/papers/141-ai-treasury-liquidity-alm-forecasting-stress-evidence-architecture.md | AI Treasury / Liquidity / ALM Forecasting / Stress Evidence Architecture | 解释流动性预测、存款流失、ALM、压力测试、FTP、委员会决策和董事会 MI 如何形成证据架构 |
docs/ai-foundations/papers/142-ai-complaint-intelligence-root-cause-regulatory-response-architecture.md | AI Complaint Intelligence / Root Cause / Regulatory Response Architecture | 解释投诉分类、伤害识别、根因、产品缺陷、监管响应、CAPA 和整改证据如何组织 |
3.3 架构与工程层
| 文件 | 训练重点 | 输出 |
|---|---|---|
docs/AGENTIC_ENTERPRISE_ARCHITECTURE_90_PLAN.md | Agentic enterprise architecture | 9 层架构、agent workflow、tool governance |
docs/AI_ARCHITECTURE_DIAGRAM_PLAYBOOK.md | 架构图表达 | capability、C4、BPMN、RAG、Agent、Eval、Risk 图 |
docs/AI_ARCHITECTURE_REVIEW_GATE_CHECKLISTS.md | 架构评审门禁 | G0-G9 review gate、pilot/release/scale checklist |
docs/AI_CONTEXT_ENGINEERING_PLAYBOOK.md | 上下文工程 | prompt、RAG、tool observation、policy、schema、memory、eval |
docs/AI_ENGINEERING_METHODOLOGY.md | AI 工程方法论 | 生产级 AI system checklist |
docs/AI_RETRIEVAL_EVAL_GRAPH_RAG_PLAYBOOK.md | Retrieval / Eval / GraphRAG | retrieval eval stack、RAGAS 指标、GraphRAG ADR、金融零售知识系统 lab |
docs/AI_PLATFORM_SECURITY_GATEWAY_LAB.md | AI 平台安全网关 | prompt injection、tool gateway、policy engine、DLP、audit、kill switch、incident drill |
docs/AI_MEMORY_CONTEXT_STATE_PLAYBOOK.md | AI Memory / Context / State | memory taxonomy、state boundary、retention/deletion、privacy、memory eval、incident triage |
docs/AI_MULTI_AGENT_ORCHESTRATION_PLAYBOOK.md | Multi-Agent Orchestration / HITL | role charter、handoff contract、shared state、policy supervisor、human approval、eval scorecard |
docs/AI_OBSERVABILITY_COST_SLO_PLAYBOOK.md | AI Observability / Cost / SLO | GenAI trace、span、quality/cost/latency/safety SLO、unit economics、incident loop |
docs/AI_AGENT_PROTOCOLS_MCP_A2A_PLAYBOOK.md | Agent Protocols / MCP / A2A | MCP、tool contract、capability discovery、agent handoff、auth、audit、integration governance |
docs/AI_ASSURANCE_SAFETY_CASE_PLAYBOOK.md | AI Assurance / Safety Case | assurance case、claim-argument-evidence、confidence level、control evidence、release decision |
docs/AI_SYNTHETIC_EVAL_DATA_PLAYBOOK.md | Synthetic Eval Data | synthetic scenario generation、coverage matrix、quality controls、attack/edge/regulatory question sets |
docs/AI_HUMAN_OVERSIGHT_HITL_PLAYBOOK.md | Human Oversight / HITL | oversight taxonomy、handoff、override、kill switch、AI literacy、audit trail、workflow requirements |
docs/AI_THREAT_MODELING_RED_TEAM_PLAYBOOK.md | AI Threat Modeling / Red Team | LLM/RAG/Agent attack surface、OWASP/MITRE mapping、red-team tests、mitigation、incident tabletop |
docs/AI_PRIVACY_DATA_PROTECTION_PLAYBOOK.md | AI Privacy / Data Protection | privacy-by-design、PII boundary、DPIA/PIA、retention/deletion、prompt/RAG/memory/log privacy controls |
docs/AI_THIRD_PARTY_VENDOR_RISK_PLAYBOOK.md | AI Third-Party / Vendor Risk | vendor lifecycle、due diligence、contract clauses、model update notice、audit rights、exit plan |
docs/AI_PROCESS_MINING_WORKFLOW_INTELLIGENCE_PLAYBOOK.md | AI Process Mining / Workflow Intelligence | event log、process discovery、conformance、variant/bottleneck analysis、AI opportunity scoring |
docs/AI_KNOWLEDGE_GOVERNANCE_ONTOLOGY_PLAYBOOK.md | AI Knowledge Governance / Ontology | source authority、taxonomy/ontology、knowledge graph、freshness、permission、evidence lineage、GraphRAG fit |
docs/AI_SEMANTIC_LAYER_METRICS_ARCHITECTURE_PLAYBOOK.md | AI Semantic Layer / Metrics Architecture | semantic models、metric contracts、lineage、LLM-to-SQL guardrails、AI value/eval metrics |
docs/AI_PRODUCT_ARCHITECTURE_STRATEGY_PLAYBOOK.md | AI Product Architecture Strategy | capability-to-architecture、platform vs point solution、architecture runway、funding gate、scale/stop rules |
docs/AI_EVALOPS_PLATFORM_ARCHITECTURE_PLAYBOOK.md | AI EvalOps Platform Architecture | eval datasets、judge、human review、experiment comparison、release gate、production monitoring、evidence binder |
docs/AI_ENTERPRISE_INTEGRATION_EVENT_DRIVEN_AGENT_PLAYBOOK.md | AI Enterprise Integration / Event-Driven Agent | API/event/workflow pattern、CloudEvents、AsyncAPI、tool contract、idempotency、HITL queues、audit/replay |
docs/AI_POLICY_AS_CODE_DECISION_AUTOMATION_PLAYBOOK.md | AI Policy-as-Code / Decision Automation | DMN、decision service、OPA/Rego、Cedar、Zanzibar/ReBAC、PDP/PEP、policy testing、simulation、audit evidence |
docs/AI_REAL_TIME_FEATURE_STORE_DECISIONING_PLAYBOOK.md | AI Real-Time Feature Store / Decisioning | feature store、offline/online consistency、point-in-time correctness、freshness SLO、实时 fraud/credit/KYC/Agent 决策 |
docs/AI_ENGINEERING_PRODUCTIVITY_CODE_AGENT_OPERATING_SYSTEM_PLAYBOOK.md | AI Engineering Productivity / Code Agent Operating System | code agent、AI SDLC、DORA/SPACE、coding eval、secure coding、PR gate、工程效率平台 operating model |
docs/AI_DURABLE_AGENT_WORKFLOW_STATE_MACHINE_PLAYBOOK.md | AI Durable Agent Workflow / State Machine | durable execution、状态机、Saga/compensation、幂等、HITL、replay、DLQ、Agent 工具副作用控制 |
docs/AI_FINOPS_UNIT_ECONOMICS_CAPACITY_PLAYBOOK.md | AI FinOps / Unit Economics / Capacity | token/GPU/case 成本模型、capacity planning、routing/cache、SLO budget、showback/chargeback、预算门禁 |
docs/AI_FRONTIER_MODEL_STRATEGY_DISTILLATION_SMALL_MODELS_PLAYBOOK.md | AI Frontier Model Strategy / Distillation / Small Models | frontier vs small model、蒸馏、量化、specialist models、model cascade、teacher-student eval、生命周期治理 |
docs/AI_DIGITAL_TWIN_SIMULATION_PRODUCT_ARCHITECTURE_PLAYBOOK.md | AI Digital Twin / Simulation Product Architecture | digital twin、ABM/DES/system dynamics、scenario library、calibration、validation、policy simulation、decision memo |
docs/AI_EXPERIMENTATION_PLATFORM_RELEASE_SCIENCE_PLAYBOOK.md | AI Experimentation Platform / Release Science | A/B、CUPED、shadow/ramp、guardrail metrics、release gate、post-experiment decision、AI 发布治理 |
docs/AI_DATA_CONTRACTS_LINEAGE_QUALITY_PLAYBOOK.md | AI Data Contracts / Lineage / Quality | data contract、metadata、lineage、quality SLO、training/eval/RAG lineage、contract testing、data incident |
docs/AI_PROGRAMMATIC_LABELING_DATA_CENTRIC_AI_PLAYBOOK.md | AI Programmatic Labeling / Data-Centric AI | labeling functions、label model、SME workflow、coverage/conflict、label provenance、dataset card 和 LabelOps 平台 |
docs/AI_ACTIVE_LEARNING_HUMAN_FEEDBACK_OPERATIONS_PLAYBOOK.md | AI Active Learning / Human Feedback Operations | query strategy、HITL review queue、reviewer calibration、label budget、feedback governance 和 eval set protection |
docs/AI_ML_TECHNICAL_DEBT_ARCHITECTURE_PLAYBOOK.md | AI ML Technical Debt Architecture | CACE、entanglement、data/config debt、hidden feedback loops、consumer registry、debt register 和 paydown roadmap |
docs/AI_MLOPS_CONTINUOUS_DELIVERY_RELEASE_PLAYBOOK.md | AI MLOps Continuous Delivery / Release | CD4ML、release bundle、CI/CD/CT、registry、shadow/canary/ramp、rollback、release evidence 和 risk-tiered approval |
docs/AI_ARCHITECTURE_DECISION_RECORDS_GOVERNANCE_PLAYBOOK.md | AI Architecture Decision Records Governance | AI ADR taxonomy、risk tier、evidence links、reversal triggers、review workflow、supersede 和 architecture knowledge |
docs/AI_REQUIREMENTS_ENGINEERING_GQM_EVAL_CONTRACTS_PLAYBOOK.md | AI Requirements Engineering / GQM Eval Contracts | GQM、AI requirement taxonomy、eval contract、golden set、release gate、monitoring gate 和作品集证据 |
docs/AI_QUALITY_ATTRIBUTES_ATAM_TRADEOFF_PLAYBOOK.md | AI Quality Attributes / ATAM Tradeoff | quality attribute scenario、utility tree、tradeoff matrix、sensitivity/risk、ADR/eval/release gate 连接 |
docs/AI_CAPABILITY_BASED_PLANNING_BUSINESS_ARCHITECTURE_PLAYBOOK.md | AI Capability-Based Planning / Business Architecture | capability portfolio、value stream、business architecture、maturity model、architecture roadmap 和 funding gate |
docs/AI_WARDLEY_MAPPING_PRODUCT_STRATEGY_PLAYBOOK.md | AI Wardley Mapping / Product Strategy | user need、AI value chain、evolution mapping、build-buy-partner、platform boundary 和战略路线 |
docs/AI_CONTINUOUS_DISCOVERY_OPPORTUNITY_SOLUTION_TREE_PLAYBOOK.md | AI Continuous Discovery / Opportunity Solution Tree | outcome、opportunity、solution、assumption、eval、pilot decision 和金融零售 AI discovery portfolio |
docs/AI_JTBD_OUTCOME_DRIVEN_INNOVATION_PLAYBOOK.md | AI JTBD / Outcome-Driven Innovation | job map、desired outcome、opportunity score、automation boundary、evidence-to-architecture traceability |
docs/AI_PRODUCT_METRICS_NORTH_STAR_VALUE_MEASUREMENT_PLAYBOOK.md | AI Product Metrics / North Star Value Measurement | North Star、input metrics、guardrails、risk-adjusted value、benefits realization 和 product analytics governance |
docs/AI_PORTFOLIO_MANAGEMENT_FUNDING_GOVERNANCE_PLAYBOOK.md | AI Portfolio Management / Funding Governance | portfolio kanban、investment thesis、capacity allocation、funding gate、benefits realization 和 scale/stop governance |
docs/AI_SERVICE_BLUEPRINT_CUSTOMER_JOURNEY_TRUST_PLAYBOOK.md | AI Service Blueprint / Customer Journey / Trust | service blueprint、trust calibration、frontstage/backstage、handoff、appeal、controls/evidence 和 journey metrics |
docs/AI_BUSINESS_PROCESS_REENGINEERING_BPMN_DMN_PLAYBOOK.md | AI Business Process Reengineering / BPMN / DMN | process mining、BPMN orchestration、DMN decision service、AI insertion、eval traceability 和 audit trail |
docs/AI_ENTERPRISE_REFERENCE_ARCHITECTURE_CONTROL_PLANE_PLAYBOOK.md | AI Enterprise Reference Architecture / Control Plane | enterprise AI 八层架构、control plane、model/tool gateway、observability、evidence plane 和架构视图 |
docs/AI_PRODUCT_LINE_ENGINEERING_REUSABLE_PLATFORM_ASSETS_PLAYBOOK.md | AI Product Line Engineering / Reusable Platform Assets | software product line、core assets、variation points、domain architecture、asset governance、reuse ROI 和 funding model |
docs/AI_MATURITY_MODEL_ROADMAP_CAPABILITY_ASSESSMENT_PLAYBOOK.md | AI Maturity Model / Roadmap / Capability Assessment | maturity levels、capability domains、evidence standard、roadmap dependency 和个人/组织成长证据 |
docs/AI_DOMAIN_DRIVEN_DESIGN_UBIQUITOUS_LANGUAGE_PLAYBOOK.md | AI Domain-Driven Design / Ubiquitous Language | bounded context、ubiquitous language、domain model、RAG boundary、AI task boundary 和 eval taxonomy |
docs/AI_EVENT_STORMING_AGENT_WORKFLOW_DISCOVERY_PLAYBOOK.md | AI EventStorming / Agent Workflow Discovery | event storm board、agent workflow trace、hotspot-to-eval map、exception/compensation 和 tool/HITL 设计 |
docs/AI_KNOWLEDGE_WORK_REDESIGN_ROLE_TASK_ARCHITECTURE_PLAYBOOK.md | AI Knowledge Work Redesign / Role-Task Architecture | role-task matrix、human-AI responsibility、task allocation、training/adoption 和 workload-risk dashboard |
docs/AI_PLATFORM_SERVICE_CATALOG_GOLDEN_PATHS_PLAYBOOK.md | AI Platform Service Catalog / Golden Paths | model/RAG/eval/tool/policy/HITL/evidence service catalog、golden path checklist 和平台 adoption dashboard |
docs/AI_ARCHITECTURE_VIEWS_C4_ARC42_42010_PLAYBOOK.md | AI Architecture Views / C4 / arc42 / 42010 | stakeholder-concern matrix、viewpoint catalog、C4/arc42 documentation、runtime/control/evidence views |
docs/AI_ARCHITECTURE_FITNESS_FUNCTIONS_CONTINUOUS_GOVERNANCE_PLAYBOOK.md | AI Architecture Fitness Functions / Continuous Governance | fitness function catalog、gate matrix、exception memo、architecture governance dashboard |
docs/AI_CONTRACT_FIRST_TOOL_API_DESIGN_OPENAPI_ASYNCAPI_PLAYBOOK.md | AI Contract-First Tool/API Design / OpenAPI / AsyncAPI | tool contract card、event contract、schema review、compatibility policy、contract test |
docs/AI_TRACEABILITY_REQUIREMENTS_EVAL_CONTROL_GRAPH_PLAYBOOK.md | AI Traceability Requirements-Eval-Control Graph | traceability graph table、coverage matrix、evidence query、release decision memo 和 audit Q&A |
docs/AI_ENTERPRISE_ARCHITECTURE_TOGAF_ARCHIMATE_ADM_PLAYBOOK.md | AI Enterprise Architecture / TOGAF / ArchiMate / ADM | AI ADM canvas、ArchiMate layer map、architecture repository、transition roadmap 和 governance pack |
docs/AI_BANKING_REFERENCE_MODELS_BIAN_FIBO_ISO20022_PLAYBOOK.md | AI Banking Reference Models / BIAN / FIBO / ISO 20022 | reference model mapping、semantic gap log、domain service card 和金融 AI use case fit |
docs/AI_SEMANTIC_INTEROPERABILITY_RDF_OWL_SHACL_PLAYBOOK.md | AI Semantic Interoperability / RDF / OWL / SHACL | semantic contract、ontology slice、SHACL constraints、semantic drift log 和语义评测 |
docs/AI_VALUE_STREAM_MANAGEMENT_FLOW_METRICS_PLAYBOOK.md | AI Value Stream Management / Flow Metrics | AI value stream canvas、flow metrics dashboard、blocked work taxonomy 和 benefits realization loop |
docs/AI_REGULATORY_ARCHITECTURE_EU_AI_ACT_NIST_ISO42001_PLAYBOOK.md | AI Regulatory Architecture / EU AI Act / NIST / ISO 42001 | risk tier taxonomy、obligations-to-controls map、lifecycle gate、evidence architecture 和金融零售监管案例 |
docs/AI_MODEL_VALIDATION_INDEPENDENT_CHALLENGE_PLAYBOOK.md | AI Model Validation / Independent Challenge | AI system inventory、validation plan、independent challenge memo、finding log 和 revalidation dashboard |
docs/AI_THIRD_PARTY_VENDOR_CONTRACT_EXIT_ARCHITECTURE_PLAYBOOK.md | AI Third-Party Vendor Contract / Exit Architecture | vendor due diligence、contract clause map、model update impact、audit rights 和 exit runbook |
docs/AI_DATA_LIFECYCLE_GOVERNANCE_PROVENANCE_RETENTION_PLAYBOOK.md | AI Data Lifecycle Governance / Provenance / Retention | data lifecycle register、provenance card、retention matrix、deletion evidence 和 lineage control |
docs/AI_AGENT_AUTONOMY_DELEGATION_ARCHITECTURE_PLAYBOOK.md | AI Agent Autonomy / Delegation Architecture | autonomy levels、delegation contract、tool authority、human escalation、kill switch 和 monitoring dashboard |
docs/AI_AGENT_IDENTITY_DELEGATED_AUTHORIZATION_PLAYBOOK.md | AI Agent Identity / Delegated Authorization | agent identity card、OAuth token exchange、scope catalog、consent UX、audit claims 和 revocation runbook |
docs/AI_RUNTIME_EVIDENCE_OBSERVABILITY_ARCHITECTURE_PLAYBOOK.md | AI Runtime Evidence / Observability Architecture | AI span schema、evidence event contract、dashboard spec、incident evidence pack 和 audit query catalog |
docs/AI_PORTFOLIO_SYSTEMIC_RISK_DEPENDENCY_ARCHITECTURE_PLAYBOOK.md | AI Portfolio Systemic Risk / Dependency Architecture | AI dependency register、concentration heatmap、blast-radius map、scenario stress test 和 portfolio KRI |
docs/AI_CUSTOMER_HARM_RECOURSE_REMEDIATION_PLAYBOOK.md | AI Customer Harm / Recourse / Remediation | harm taxonomy、recourse workflow、remediation ledger、customer recovery KPI 和 prevention control |
docs/AI_FAIRNESS_FAIR_LENDING_BIAS_CONTROL_PLAYBOOK.md | AI Fairness / Fair Lending / Bias Control | fairness eval matrix、proxy risk register、segment guardrail、review calibration 和 evidence binder |
docs/AI_EXPLAINABILITY_CONTESTABILITY_ADVERSE_ACTION_PLAYBOOK.md | AI Explainability / Contestability / Adverse Action | reason-code catalog、adverse-action evidence packet、appeal SLA、human review checklist 和 explanation QA |
docs/AI_CHANGE_IMPACT_RELEASE_GOVERNANCE_PLAYBOOK.md | AI Change Impact / Release Governance | change classification、impact graph、regression gate、release evidence bundle 和 rollback runbook |
docs/AI_CONTINUOUS_CONTROL_MONITORING_ASSURANCE_PLAYBOOK.md | AI Continuous Control Monitoring / Assurance | control test catalog、exception schema、KRI dashboard、sampling plan、management action 和 assurance pack |
docs/AI_OPERATIONAL_RESILIENCE_BCP_DEGRADED_MODE_PLAYBOOK.md | AI Operational Resilience / BCP / Degraded Mode | critical operation map、dependency graph、degraded-mode matrix、manual fallback、RTO/RPO 和 exercise evidence |
docs/AI_MANAGEMENT_INFORMATION_BOARD_REPORTING_PLAYBOOK.md | AI Management Information / Board Reporting | metric contracts、MI lineage、risk appetite dashboard、board pack、action log 和 report validation |
docs/AI_CLOSED_LOOP_LEARNING_CORRECTIVE_ACTION_PLAYBOOK.md | AI Closed-Loop Learning / Corrective Action | feedback taxonomy、CAPA workflow、root cause、change linkage、effectiveness verification 和 closure evidence |
docs/AI_REGULATORY_HORIZON_OBLIGATION_INTELLIGENCE_PLAYBOOK.md | AI Regulatory Horizon / Obligation Intelligence | source registry、obligation ontology、applicability triage、impact graph、governance workflow 和 horizon dashboard |
docs/AI_EXCEPTION_RISK_ACCEPTANCE_WAIVER_PLAYBOOK.md | AI Exception / Risk Acceptance / Waiver | exception taxonomy、waiver lifecycle、risk acceptance memo、compensating controls、expiry 和 hard stop |
docs/AI_SUPPLY_CHAIN_AI_BOM_PROVENANCE_PLAYBOOK.md | AI Supply Chain / AI BOM / Provenance | AI BOM schema、component taxonomy、provenance graph、vulnerability response、rights 和 supplier mapping |
docs/AI_HUMAN_REVIEW_OPERATIONS_CAPACITY_PLAYBOOK.md | AI Human Review Operations / Capacity | queue taxonomy、skill/risk routing、capacity model、calibration、reviewer quality、surge mode 和 evidence |
docs/AI_SEGREGATION_OF_DUTIES_DUAL_CONTROL_PLAYBOOK.md | AI Segregation of Duties / Dual Control | incompatible duty matrix、maker-checker workflow、approval token、override ownership 和 evidence checklist |
docs/AI_CONSENT_PREFERENCE_PURPOSE_BOUND_DATA_PLAYBOOK.md | AI Consent / Preference / Purpose-Bound Data | purpose catalog、consent event schema、preference center、runtime enforcement、withdrawal/re-consent |
docs/AI_SHADOW_AI_CITIZEN_DEVELOPMENT_GOVERNANCE_PLAYBOOK.md | AI Shadow AI / Citizen Development Governance | discovery register、risk tiering、approved tool catalog、citizen developer guardrails 和 platform migration |
docs/AI_CONDUCT_RISK_SUITABILITY_SALES_GUARDRAILS_PLAYBOOK.md | AI Conduct Risk / Suitability / Sales Guardrails | conduct taxonomy、suitability gates、approved claims、surveillance KRI、complaint/remediation linkage |
docs/AI_RECORDS_RETENTION_LEGAL_HOLD_EDISCOVERY_PLAYBOOK.md | AI Records / Retention / Legal Hold / eDiscovery | AI record taxonomy、retention matrix、legal hold trigger、production package 和 export manifest |
docs/AI_DATA_RESIDENCY_CROSS_BORDER_SOVEREIGN_AI_PLAYBOOK.md | AI Data Residency / Cross-Border / Sovereign AI | residency decision tree、jurisdiction-purpose-vendor matrix、region routing、key residency 和 transfer review |
docs/AI_CUSTOMER_COMMUNICATIONS_REGULATED_CONTENT_LIFECYCLE_PLAYBOOK.md | AI Customer Communications / Regulated Content Lifecycle | approved claims、forbidden claims、pre-use review、post-use surveillance、disclosure versioning 和 complaint linkage |
docs/AI_FINANCIAL_CRIME_TYPOLOGY_SCENARIO_COVERAGE_PLAYBOOK.md | AI Financial Crime Typology / Scenario Coverage | typology object model、red-flag mapping、coverage matrix、SAR evidence bundle 和 alert-to-SAR traceability |
docs/AI_INTELLECTUAL_PROPERTY_CONTENT_RIGHTS_PROVENANCE_PLAYBOOK.md | AI Intellectual Property / Content Rights / Provenance | content object taxonomy、rights matrix、C2PA manifest、output clearance workflow、license evidence 和 takedown |
docs/AI_DEEPFAKE_SYNTHETIC_IDENTITY_AUTHENTICATION_FRAUD_PLAYBOOK.md | AI Deepfake / Synthetic Identity / Authentication Fraud | proofing control matrix、liveness/PAD、step-up policy、fraud evidence schema、red-team scenarios 和 customer friction |
docs/AI_WORKFORCE_HR_DECISION_EMPLOYEE_MONITORING_GOVERNANCE_PLAYBOOK.md | AI Workforce / HR Decision / Employee Monitoring Governance | workforce AI inventory、adverse impact test、employee notice、human review、data minimization 和 monitoring KRI |
docs/AI_INCIDENT_DISCLOSURE_LIABILITY_RISK_TRANSFER_PLAYBOOK.md | AI Incident Disclosure / Liability / Risk Transfer | incident taxonomy、materiality decision tree、liability boundary map、insurance notification 和 executive evidence pack |
docs/AI_POST_QUANTUM_CRYPTOGRAPHIC_AGILITY_PLAYBOOK.md | AI Post-Quantum / Cryptographic Agility | AI crypto inventory、long-lived evidence matrix、vendor readiness、crypto profile、evidence replay 和 migration roadmap |
docs/AI_AUTHORIZED_PUSH_PAYMENT_SCAM_INTERVENTION_PLAYBOOK.md | AI Authorized Push Payment / Scam Intervention | APP scam taxonomy、customer intent、beneficiary risk、intervention ladder、fraud escalation 和 remediation evidence |
docs/AI_AGENT_MARKETPLACE_TOOL_CERTIFICATION_GOVERNANCE_PLAYBOOK.md | AI Agent Marketplace / Tool Certification Governance | capability card、risk tier、tool/API certification、signed package、runtime permission、owner attestation 和 lifecycle |
docs/AI_CUSTOMER_VULNERABILITY_ACCESSIBILITY_INCLUSIVE_AI_PLAYBOOK.md | AI Customer Vulnerability / Accessibility / Inclusive AI | support-need taxonomy、accessibility gate、plain language、safe escalation、QA/eval、complaint linkage 和 CAPA |
docs/AI_PAYMENT_DISPUTE_CHARGEBACK_CLAIMS_EVIDENCE_PLAYBOOK.md | AI Payment Dispute / Chargeback / Claims Evidence | dispute taxonomy、case clock、evidence bundle、provisional credit logic、customer communication 和 complaint RCA |
docs/AI_COLLECTIONS_HARDSHIP_DELINQUENCY_TREATMENT_PLAYBOOK.md | AI Collections / Hardship / Delinquency Treatment | delinquency signals、hardship options、contact strategy、fair treatment controls、accessibility 和 complaints loop |
docs/AI_VOICE_AI_CONTACT_CENTER_AGENT_ASSIST_GOVERNANCE_PLAYBOOK.md | AI Voice AI / Contact Center / Agent Assist Governance | voice bot taxonomy、agent-assist guardrails、call summary QA、disclosure boundary、telemetry 和 complaint linkage |
docs/AI_DIGITAL_IDENTITY_WALLET_VERIFIABLE_CREDENTIALS_TRUST_PLAYBOOK.md | AI Digital Identity Wallet / Verifiable Credentials / Trust | wallet trust framework、VC verification、DID resolution、WebAuthn context、selective disclosure 和 revocation policy |
docs/AI_OPEN_BANKING_OPEN_FINANCE_CONSENTED_DATA_SHARING_PLAYBOOK.md | AI Open Banking / Open Finance / Consented Data Sharing | consented data taxonomy、authorization UX、revocation、API contract、third-party onboarding、AI use boundary |
docs/AI_PERSONALIZED_PRICING_OFFER_DECISIONING_GOVERNANCE_PLAYBOOK.md | AI Personalized Pricing / Offer Decisioning Governance | pricing taxonomy、feature boundary、offer policy、experiment guardrail、reason handoff、complaint monitoring |
docs/AI_DOCUMENT_INTELLIGENCE_UNSTRUCTURED_DATA_EVIDENCE_QUALITY_PLAYBOOK.md | AI Document Intelligence / Unstructured Data / Evidence Quality | document taxonomy、extraction schema、confidence/review policy、records mapping、tamper checks、evidence manifest |
docs/AI_PRIVACY_CLEAN_ROOM_DATA_COLLABORATION_MEASUREMENT_PLAYBOOK.md | AI Privacy Clean Room / Data Collaboration / Measurement | collaboration use-case intake、data contract、query/output controls、PET selection、partner risk、measurement evidence |
docs/AI_CREDIT_LIFECYCLE_UNDERWRITING_LINE_MANAGEMENT_GOVERNANCE_PLAYBOOK.md | AI Credit Lifecycle / Underwriting / Line Management Governance | credit lifecycle inventory、decision factory、line governance、reason architecture、portfolio and complaint monitoring |
docs/AI_WEALTH_ADVICE_ROBO_ADVISOR_BEST_INTEREST_BOUNDARY_PLAYBOOK.md | AI Wealth Advice / Robo-Advisor / Best Interest Boundary | advice boundary taxonomy、risk profile、approved universe、recommendation policy、human escalation、supervision evidence |
docs/AI_TREASURY_LIQUIDITY_ALM_FORECASTING_STRESS_EVIDENCE_PLAYBOOK.md | AI Treasury / Liquidity / ALM Forecasting / Stress Evidence | liquidity forecast object、deposit runoff、stress scenario、ALM committee workflow、contingency funding、board MI |
docs/AI_COMPLAINT_INTELLIGENCE_ROOT_CAUSE_REGULATORY_RESPONSE_PLAYBOOK.md | AI Complaint Intelligence / Root Cause / Regulatory Response | complaint ledger、harm taxonomy、RCA graph、regulatory response pack、CAPA workflow、board reporting |
docs/AI_FORECASTING_DEMAND_PLANNING_PRODUCT_ARCHITECTURE_PLAYBOOK.md | AI Forecasting / Demand Planning Product Architecture | time-series forecasting、demand planning、capacity forecasting、现金流、预测区间、backtesting、forecast governance |
docs/AI_OPTIMIZATION_OPERATIONS_RESEARCH_DECISION_PLAYBOOK.md | AI Optimization / Operations Research Decision | LP/MIP、CP-SAT、solver service、目标/约束、多目标权衡、forecast-to-optimization、例外工作流 |
docs/AI_BAYESIAN_OPTIMIZATION_EXPERIMENT_DESIGN_PLAYBOOK.md | AI Bayesian Optimization / Experiment Design | BoTorch、Optuna、surrogate/acquisition、prompt/RAG/model tuning、pricing/offer 参数优化、实验预算和 safe experimentation |
3.4 Governance / EvalOps / Operations 层
| 文件 | 训练重点 | 输出 |
|---|---|---|
docs/AI_GOVERNANCE_EVALOPS_RISK_90_PLAN.md | AI governance、EvalOps、RiskOps | risk control、eval gate、incident response |
docs/AI_OPERATING_MODEL_RACI_RUNBOOK.md | 上线后谁负责 | RACI、change control、incident runbook、adoption cadence |
docs/AI_SAFETY_ENGINEERING_STPA_PLAYBOOK.md | AI Safety Engineering / STPA | loss、hazard、unsafe control action、control structure、safety constraint、熔断和人工接管 |
docs/AI_SOCIO_TECHNICAL_RESILIENCE_OPERATING_MODEL_PLAYBOOK.md | AI Sociotechnical Resilience Operating Model | work-as-done、human-AI collaboration、handoff/exception/load、resilience metrics 和 operating model |
docs/AI_TEAM_TOPOLOGIES_CONWAY_PLATFORM_OPERATING_MODEL_PLAYBOOK.md | AI Team Topologies / Conway Platform Operating Model | team topology、Conway mirroring、cognitive load、team API、interaction modes 和平台组织模型 |
docs/AI_DORA_SPACE_ENGINEERING_PRODUCTIVITY_SDLC_PLAYBOOK.md | AI DORA / SPACE Engineering Productivity SDLC | DORA、SPACE、AI SDLC metrics、code agent governance、PR/eval/release gate 和 DevEx |
docs/AI_PRODUCT_OPERATING_MODEL_EMPOWERED_TEAMS_PLAYBOOK.md | AI Product Operating Model / Empowered Teams | product trio+、decision rights、discovery-delivery-governance cadence、product/platform/risk operating model |
docs/AI_RISK_APPETITE_POLICY_PRODUCT_MANAGEMENT_PLAYBOOK.md | AI Risk Appetite / Policy Product Management | risk appetite statement、policy-to-product matrix、risk-tiered controls、exception memo 和 stop rule |
docs/AI_CONTROL_LIBRARY_ASSURANCE_EVIDENCE_GRAPH_PLAYBOOK.md | AI Control Library / Assurance Evidence Graph | control catalog、assurance case、claim-risk-control-evidence graph、release gate 和 regulator Q&A map |
docs/AI_REGULATORY_RESPONSE_PLAYBOOK.md | AI 监管响应 | regulatory radar、applicability judgment、AI inventory、control mapping、evidence pack |
docs/AI_BOARD_AUDIT_COMMITTEE_GOVERNANCE_PACK.md | 董事会/审计委员会 AI 治理 | portfolio risk dashboard、three lines of defense、attestation、audit evidence |
docs/AI_REGULATOR_EXAM_SIMULATION_PACK.md | 监管检查 / 内审问询演练 | exam room scenario、50 个 examiner questions、evidence index、control effectiveness narrative |
docs/AI_MODEL_RISK_MANAGEMENT_PLAYBOOK.md | AI Model Risk Management | SR 11-7 到 GenAI system risk、inventory、validation、independent challenge、change control、monitoring |
docs/AI_AUDIT_EVIDENCE_BINDER_PLAYBOOK.md | AI Audit Evidence Binder | control evidence、model/system cards、dataset cards、evidence lifecycle、RACI、audit log、quality rubric |
docs/AI_INCIDENT_POSTMORTEM_RELIABILITY_PLAYBOOK.md | AI Incident / Postmortem / Reliability | AI incident taxonomy、severity、containment、rollback、postmortem、corrective action、reliability review |
docs/AI_TRUST_EXPERIENCE_PRODUCT_GOVERNANCE_PLAYBOOK.md | AI Trust Experience / Product Governance | trust calibration、transparency、uncertainty、refusal、escalation、complaint/appeal UX、over-reliance controls |
docs/AI_POLICY_AS_CODE_DECISION_AUTOMATION_PLAYBOOK.md | AI Policy-as-Code / Decision Automation | 把政策、授权、审批、解释、模拟、回滚和审计从 prompt 外置成可测试 runtime governance |
docs/AI_PRIVACY_ENHANCING_TECH_CONFIDENTIAL_AI_PLAYBOOK.md | AI Privacy-Enhancing Tech / Confidential AI | 把 DP、FL、TEE、FHE、secure aggregation、clean room、privacy budget 和 confidential inference 变成高阶隐私架构 |
docs/AI_SECURITY_OPERATIONS_SOC_PLAYBOOK.md | AI Security Operations / SOC | 把 MITRE ATLAS、OWASP、NIST CSF、AI telemetry、detection rules、SIEM/SOAR、incident runbook 和 purple team 变成持续安全运营 |
docs/AI_ANOMALY_DETECTION_RISK_MONITORING_PLAYBOOK.md | AI Anomaly Detection / Risk Monitoring | 把 fraud、AML、ops、model drift、security、cost anomaly 接入阈值、告警、triage、runbook 和反馈治理 |
docs/AI_UNCERTAINTY_CALIBRATION_CONFORMAL_PREDICTION_PLAYBOOK.md | AI Uncertainty / Calibration / Conformal Prediction | 把 calibration、coverage、confidence UX、abstention、risk-based routing 和人工升级变成不确定性控制面 |
docs/AI_DATASET_SHIFT_MONITORING_MODEL_PERFORMANCE_PLAYBOOK.md | AI Dataset Shift / Model Performance Operations | 把 feature/score/embedding drift、outcome lag、segment monitoring、alert runbook 和 model performance 变成生产运营控制面 |
docs/AI_MANAGEMENT_SYSTEM_ISO42001_OPERATING_MODEL_PLAYBOOK.md | AI Management System / ISO 42001 Operating Model | 把 AI inventory、risk tier、release gate、control library、management review 和 continual improvement 变成 AI 管理体系 |
docs/AI_HUMAN_AI_INTERACTION_PRODUCT_DESIGN_PLAYBOOK.md | AI Human-AI Interaction Product Design | 把 capability boundary、calibrated trust、recoverability、automation bias、feedback 和 human escalation 变成产品治理能力 |
docs/LEARNING_ASSET_GOVERNANCE.md | 学习资产治理 | 保留旧内容、复习、归档、作品集转换 |
docs/AI_LONG_TERM_KNOWLEDGE_GRAPH_AND_REVIEW_SYSTEM.md | 12-18 个月复习系统 | knowledge graph、spaced review、artifact conversion |
3.5 案例、作品集和面试层
| 文件 | 训练重点 | 输出 |
|---|---|---|
docs/AI_BA_PM_PRACTICE_LAB.md | BA/PM/Architect case drill | problem framing、workflow、eval、risk、ROI |
docs/FINANCIAL_RETAIL_AI_CASE_PORTFOLIO.md | 金融零售 AI 作品集案例 | 12 个 case backlog |
docs/AI_INTERVIEW_PORTFOLIO_STORYLINE_PLAYBOOK.md | 面试叙事 | 30 秒、2 分钟、deep dive、追问 |
docs/AI_CASE_DRILL_WORKBOOK_30_DAYS.md | 30 天金融零售 case drill | AML、KYC、客服、信贷、支付、欺诈、财富、零售、供应链、合规训练 |
docs/AI_ADVANCED_CASE_DRILL_WORKBOOK_60_DAYS.md | 60 天高阶 case drill | 复杂金融零售 AI case、architecture review、governance gate、flagship portfolio pack |
docs/AI_TRANSFORMATION_VALUE_OFFICE_PLAYBOOK.md | AI Transformation Value Office | use case portfolio、funding gate、benefits realization、platform reuse、scale/stop、value governance |
docs/AI_CAUSAL_DISCOVERY_STRUCTURAL_DECISION_PLAYBOOK.md | AI Causal Discovery / Structural Decision | causal question、DAG review、assumption register、identification、intervention、sensitivity analysis 和 funding/rollout evidence |
docs/AI_DECISION_INTELLIGENCE_CAUSAL_PRODUCT_PLAYBOOK.md | AI Decision Intelligence / Causal Product | causal DAG、uplift、experimentation、DiD、synthetic control、ROI attribution、portfolio funding gates |
docs/AI_CONTEXTUAL_BANDITS_ADAPTIVE_EXPERIMENTATION_PLAYBOOK.md | AI Contextual Bandits / Adaptive Experimentation | contextual bandit、exploration budget、propensity logging、OPE、next-best-action、offer/contact routing 和 kill switch |
docs/AI_REINFORCEMENT_LEARNING_POLICY_DECISION_PLAYBOOK.md | AI Reinforcement Learning / Policy Decision | MDP、reward design、offline RL、CQL、simulator/replay、policy guardrail、human approval 和 governed rollout |
docs/AI_PERSONALIZATION_RECOMMENDER_PRODUCT_ARCHITECTURE_PLAYBOOK.md | AI Personalization / Recommender Product Architecture | candidate generation、retrieval/ranking/re-ranking、feedback loop、suitability、consent、推荐策略和 next-best-action |
docs/AI_ADOPTION_CHANGE_MANAGEMENT_PLAYBOOK.md | AI Adoption / Change Management | AI literacy、role redesign、training、support model、feedback loop、adoption metrics、benefit realization |
docs/AI_CUSTOMER_FACING_REGULATED_PRODUCT_PLAYBOOK.md | Customer-Facing Regulated AI Product | disclosure、advice boundary、complaints、accessibility、human escalation、monitoring、incident response |
docs/AI_EXECUTIVE_COMMUNICATION_MEMO_PACK.md | 高管沟通与决策 memo | pilot funding、architecture、risk acceptance、vendor、scale/stop、incident、portfolio memo |
docs/AI_DATA_PRODUCT_MANAGEMENT_PLAYBOOK.md | AI 数据产品管理 | data product canvas、contract、metadata、lineage、quality SLO、golden set、feedback loop |
docs/abpa/README.md | ABPA starter kit | 模板入口和 capstone 入口 |
docs/abpa/capstone-aml/README.md | AML capstone | 第一个完整案例主线 |
4. 七条长期路线
Route A: AI Solutions Architect 路线
适合目标：转向 AI Solutions Architect、Enterprise AI Architect、Agentic RAG Architect。
推荐顺序：
docs/AI_2026_EXPANSION_START_HERE.mddocs/AI_FOUNDATIONS_CLASSIC_PAPERS_PLAN.mddocs/ai-foundations/README.mddocs/AGENTIC_ENTERPRISE_ARCHITECTURE_90_PLAN.mddocs/AI_ARCHITECTURE_DIAGRAM_PLAYBOOK.mddocs/AI_CONTEXT_ENGINEERING_PLAYBOOK.mddocs/AI_RETRIEVAL_EVAL_GRAPH_RAG_PLAYBOOK.mddocs/AI_PLATFORM_SECURITY_GATEWAY_LAB.mddocs/AI_ARCHITECTURE_REVIEW_GATE_CHECKLISTS.mddocs/AI_GOVERNANCE_EVALOPS_RISK_90_PLAN.mddocs/AI_REGULATORY_RESPONSE_PLAYBOOK.mddocs/AI_BOARD_AUDIT_COMMITTEE_GOVERNANCE_PACK.md
每周产出：
| 周期 | 必交资产 |
|---|---|
| Week 1 | Transformer/RAG/Agent 一页纸 |
| Week 2 | Enterprise RAG ADR |
| Week 3 | Agentic workflow sequence |
| Week 4 | Architecture review gate for one case |
| Week 5 | Eval architecture + release gate |
| Week 6 | Context engineering ADR |
面试表达：
我不把 AI 架构理解成“选一个模型”。我会先看业务能力和流程，再设计知识、上下文、工具、权限、eval、observability 和 human approval。模型只是其中一层，能否进入生产取决于整套系统是否可评估、可审计、可运营。
Route B: AI BA / Business Architect 路线
适合目标：AI BA、AI Business Architect、AI Transformation Consultant。
推荐顺序：
docs/AI_BA_PRODUCT_ARCHITECT_180_PLAN.mddocs/abpa/README.mddocs/AI_REQUIREMENTS_TO_EVAL_COOKBOOK.mddocs/AI_BA_PM_PRACTICE_LAB.mddocs/FINANCIAL_RETAIL_AI_CASE_PORTFOLIO.mddocs/AI_OPERATING_MODEL_RACI_RUNBOOK.mddocs/AI_DATA_PRODUCT_MANAGEMENT_PLAYBOOK.mddocs/AI_REGULATORY_RESPONSE_PLAYBOOK.mddocs/AI_INTERVIEW_PORTFOLIO_STORYLINE_PLAYBOOK.md
每周产出：
| 周期 | 必交资产 |
|---|---|
| Week 1 | AI Opportunity Canvas + stakeholder map |
| Week 2 | AS-IS / TO-BE workflow |
| Week 3 | requirements-to-eval matrix |
| Week 4 | data readiness pack |
| Week 5 | control pack + operating model |
| Week 6 | executive decision memo |
面试表达：
AI BA 的价值不是把业务需求翻译成 prompt，而是把业务目标、流程痛点、异常路径、数据约束和验收标准转换成可评测的 AI 系统需求。我的强项是把业务事实、技术可行性和风险控制连接起来。
Route C: AI Product Manager 路线
适合目标：AI PM、AI Product Lead、AI Platform PM、金融零售 AI 产品负责人。
推荐顺序：
docs/AI_NEW_DEMANDS_2026_EXPANSION.mddocs/AI_ROLE_COMPETENCY_MATRIX_2026.mddocs/AI_BA_PM_PRACTICE_LAB.mddocs/AI_CASE_DRILL_WORKBOOK_30_DAYS.mddocs/AI_PLATFORM_PM_PLAYBOOK.mddocs/AI_VENDOR_BUILD_BUY_ADOPTION_PLAYBOOK.mddocs/AI_REQUIREMENTS_TO_EVAL_COOKBOOK.mddocs/AI_OPERATING_MODEL_RACI_RUNBOOK.mddocs/AI_EXECUTIVE_COMMUNICATION_MEMO_PACK.mddocs/AI_DATA_PRODUCT_MANAGEMENT_PLAYBOOK.mddocs/AI_CAPABILITY_ASSESSMENT_RUBRIC.mddocs/AI_INTERVIEW_PORTFOLIO_STORYLINE_PLAYBOOK.md
每周产出：
| 周期 | 必交资产 |
|---|---|
| Week 1 | use case portfolio prioritization |
| Week 2 | MVP scope + not-in-scope |
| Week 3 | pilot metrics + stop rules |
| Week 4 | adoption dashboard |
| Week 5 | build/buy/vendor decision memo |
| Week 6 | scale/stop decision memo |
面试表达：
AI PM 不能只看用户说喜欢不喜欢，还要看模型质量、工作流改变、业务价值和风险控制是否同时成立。我会用 pilot gate 管理不确定性，用 eval 和 adoption 数据决定继续、调整还是停止。
Route D: AI Governance / EvalOps 路线
适合目标：AI Governance PM、EvalOps Lead、Model Risk / AI Risk、Responsible AI。
推荐顺序：
docs/AI_GOVERNANCE_EVALOPS_RISK_90_PLAN.mddocs/AI_REQUIREMENTS_TO_EVAL_COOKBOOK.mddocs/ai-foundations/papers/08-llm-as-judge-evaluation.mddocs/AI_OPERATING_MODEL_RACI_RUNBOOK.mddocs/AI_ARCHITECTURE_REVIEW_GATE_CHECKLISTS.mddocs/AI_CONTEXT_ENGINEERING_PLAYBOOK.mddocs/AI_REGULATORY_RESPONSE_PLAYBOOK.mddocs/AI_BOARD_AUDIT_COMMITTEE_GOVERNANCE_PACK.mddocs/AI_ASSURANCE_SAFETY_CASE_PLAYBOOK.mddocs/AI_MODEL_RISK_MANAGEMENT_PLAYBOOK.mddocs/AI_SYNTHETIC_EVAL_DATA_PLAYBOOK.md
每周产出：
| 周期 | 必交资产 |
|---|---|
| Week 1 | risk register |
| Week 2 | golden set + rubric |
| Week 3 | release gate |
| Week 4 | production monitoring plan |
| Week 5 | incident runbook |
| Week 6 | audit evidence pack |
| Week 7 | regulatory impact memo + AI inventory |
| Week 8 | board/audit committee governance dashboard |
| Week 9 | safety case + model risk validation plan |
| Week 10 | synthetic eval data pack + coverage report |
面试表达：
AI governance 的重点不是写原则，而是把原则落成上线门禁、测试集、监控、审批、日志、事故处理和责任人。只有这样，高风险 AI 才能进入真实业务流程。
Route E: AI Data Product / Platform Evidence 路线
适合目标：AI Data Product Manager、AI Platform PM、AI Architect、EvalOps Lead。
推荐顺序：
docs/AI_DATA_PRODUCT_MANAGEMENT_PLAYBOOK.mddocs/AI_CONTEXT_ENGINEERING_PLAYBOOK.mddocs/AI_REQUIREMENTS_TO_EVAL_COOKBOOK.mddocs/AI_PLATFORM_PM_PLAYBOOK.mddocs/AI_RETRIEVAL_EVAL_GRAPH_RAG_PLAYBOOK.mddocs/AI_PLATFORM_SECURITY_GATEWAY_LAB.mddocs/AI_CAPABILITY_ASSESSMENT_RUBRIC.md
每周产出：
| 周期 | 必交资产 |
|---|---|
| Week 1 | data product canvas |
| Week 2 | data contract + quality SLO |
| Week 3 | golden set register + labeling SOP |
| Week 4 | RAG knowledge product lifecycle |
| Week 5 | data quality dashboard + incident playbook |
| Week 6 | platform evidence map + capability self-score |
面试表达：
企业 AI 的质量上限经常不是模型，而是数据产品能力。我要证明的不只是“能拿到数据”，而是能把数据治理成可被 RAG、eval、feedback、audit 和 ROI 稳定消费的产品。
Route F: AI Advanced Case Portfolio / Exam Defense 路线
适合目标：AI Solutions Architect、AI PM、AI BA、Enterprise AI Consultant，希望把学习资产转成可面试、可审计、可被深挖的旗舰作品集。
推荐顺序：
docs/AI_CASE_DRILL_WORKBOOK_30_DAYS.mddocs/AI_ADVANCED_CASE_DRILL_WORKBOOK_60_DAYS.mddocs/AI_RETRIEVAL_EVAL_GRAPH_RAG_PLAYBOOK.mddocs/AI_PLATFORM_SECURITY_GATEWAY_LAB.mddocs/AI_REGULATOR_EXAM_SIMULATION_PACK.mddocs/AI_INTERVIEW_PORTFOLIO_STORYLINE_PLAYBOOK.mddocs/AI_CAPABILITY_ASSESSMENT_RUBRIC.md
每周产出：
| 周期 | 必交资产 |
|---|---|
| Week 1 | advanced case selection + baseline evidence |
| Week 2 | BA workflow + requirements-to-eval |
| Week 3 | RAG/GraphRAG eval report + ADR |
| Week 4 | security gateway threat model + tool permission matrix |
| Week 5 | regulator exam evidence pack + Q&A |
| Week 6 | flagship case story + capability self-score |
面试表达：
我会把一个 AI 案例准备到能被 CTO、CRO、审计、监管和业务负责人连续追问的程度。作品集不是截图，而是一组 evidence：业务问题、流程、需求、架构、eval、权限、安全、治理、上线门禁、事故处理和面试叙事。
Route G: AI Production Operations / Agent Platform Integration 路线
适合目标：AI Platform PM、AI Solutions Architect、AI Product Operations Lead、Enterprise Agent Architect。
推荐顺序：
docs/AI_MEMORY_CONTEXT_STATE_PLAYBOOK.mddocs/AI_MULTI_AGENT_ORCHESTRATION_PLAYBOOK.mddocs/AI_OBSERVABILITY_COST_SLO_PLAYBOOK.mddocs/AI_AGENT_PROTOCOLS_MCP_A2A_PLAYBOOK.mddocs/AI_PLATFORM_SECURITY_GATEWAY_LAB.mddocs/AI_PLATFORM_PM_PLAYBOOK.mddocs/AI_ARCHITECTURE_REVIEW_GATE_CHECKLISTS.md
每周产出：
| 周期 | 必交资产 |
|---|---|
| Week 1 | memory inventory + state boundary diagram |
| Week 2 | multi-agent role charter + handoff contract |
| Week 3 | AI trace schema + SLO/cost dashboard spec |
| Week 4 | MCP/tool contract catalog + protocol ADR |
| Week 5 | security gateway policy + incident drill |
| Week 6 | platform integration blueprint + scale gate memo |
面试表达：
我把生产级 AI 平台看成 memory、workflow、tool protocol、observability、cost、security 和 governance 的组合能力。模型只是其中一层；真正能规模化的是可复用的 tool contract、state boundary、trace schema、SLO、policy gateway 和 incident loop。
5. 从学习到作品集的转换规则
每读一个理论文件，都要转成至少一个 artifact。否则它只是笔记，不是能力证据。
| 学习内容 | 作品集转换 |
|---|---|
| Transformer / attention | 一页纸解释 + 金融客服类比 |
| RAG | Enterprise RAG ADR + eval set |
| Agent / tools | Agent workflow + tool risk catalog |
| RLHF / alignment | product policy + refusal/escalation examples |
| LoRA / PEFT | RAG vs fine-tuning decision memo |
| inference optimization | cost/latency/SLO architecture note |
| LLM-as-Judge | rubric + judge calibration plan |
| MoE / sparse scaling | dense vs MoE cost/SLO memo + enterprise model selection note |
| Scaling laws / BERT-GPT-T5 | pretraining vs RAG vs fine-tuning vs self-training decision memo |
| DPO / Constitutional AI | preference guideline + chosen/rejected samples + model upgrade eval |
| Tool use security | tool risk catalog + prompt injection red-team + tool gateway ADR |
| RAG evaluation / RAGAS | retrieval eval matrix + gold source register + release gate + failure triage |
| GraphRAG / Knowledge Graph RAG | GraphRAG fit assessment + entity/relationship ontology + path eval + ADR |
| Generative Agents / Memory | memory inventory + state boundary diagram + retention/deletion matrix + memory eval set |
| AutoGen / Multi-Agent | agent role charter + handoff contract + shared state schema + orchestration eval |
| AI governance | control pack + release gate |
| operating model | RACI + incident runbook |
| vendor decision | build/buy scorecard + recommendation memo |
| AI platform | platform MVP PRD + model gateway/RAG/eval/cost backlog |
| executive communication | funding/scale/stop memo + 30 秒/2 分钟/10 分钟表达 |
| regulatory response | regulatory impact memo + AI inventory + evidence binder + exam Q&A |
| AI data product | data product canvas + data contract + golden set register + incident playbook |
| board governance | portfolio risk dashboard + three lines of defense + management attestation |
| capability assessment | C1-C14 self-score + evidence map + next-action backlog |
| AI platform security gateway | tool permission matrix + prompt injection test pack + action risk tier + incident drill |
| regulator exam simulation | evidence index + examiner Q&A + control effectiveness narrative + remediation plan |
| advanced case drill | 3 flagship cases + 6 mini cases + weekly gate evidence + interview pack |
| AI memory/context/state | Memory Inventory + State Boundary + Retention Matrix + deletion test + privacy incident triage |
| multi-agent orchestration | Role Charter + Handoff Contract + Shared State Schema + Supervisor Policy + Eval Scorecard |
| AI observability/cost/SLO | Trace Schema + SLO Matrix + Cost Unit Economics + Dashboard Spec + Postmortem |
| MCP/A2A protocols | MCP Server Intake + Tool Contract Card + Capability Matrix + Protocol ADR + Integration Risk Checklist |
| HELM / holistic evaluation | scenario-metric matrix + model comparison memo + benchmark limitation note |
| Model Cards / Datasheets | model card + dataset datasheet + system evidence register + review cadence |
| AI assurance / safety case | claim-argument-evidence tree + release confidence memo + evidence gap backlog |
| model risk management | model/system inventory + validation plan + change control + independent challenge questions |
| synthetic eval data | synthetic scenario catalog + coverage matrix + data quality checks + leakage review |
| AI transformation value office | portfolio scorecard + funding gate memo + benefits realization dashboard + scale/stop decision |
| Tree of Thoughts / planning search | candidate plan review + thought space map + search budget ADR + human choice point |
| Self-RAG / CRAG | retrieval decision matrix + context quality gate + corrective retrieval ADR + refusal/escalation rule |
| AgentBench / τ-bench | agent scenario pack + tool sandbox spec + policy oracle + agent release gate dashboard |
| mechanistic interpretability | explainability layer map + model risk memo + safety case supporting evidence note |
| human oversight / HITL | intervention matrix + handoff UX + override log + kill switch procedure + AI literacy pack |
| threat modeling / red-team | attack surface inventory + red-team test suite + mitigation matrix + incident tabletop |
| audit evidence binder | evidence index + control-to-evidence matrix + model/system card + dataset card + audit log spec |
| adoption / change management | stakeholder segmentation + role redesign + training plan + adoption dashboard + feedback loop |
| long context / RULER | long-context vs RAG decision tree + position robustness eval + cost model + context strategy ADR |
| DSPy / prompt optimization | task signature + prompt registry card + eval split + optimizer release memo |
| Reflexion / Self-Refine | feedback taxonomy + refinement policy + reflection memory ADR + before/after eval report |
| process supervision | process eval map + step schema library + golden trace set + critical step release gate |
| AI privacy / data protection | DPIA/PIA + data minimization matrix + PII boundary map + retention/deletion tests |
| third-party AI vendor risk | vendor scorecard + contract clause checklist + model update/change notice + exit plan |
| process mining / workflow intelligence | event log inventory + variant/bottleneck report + baseline ROI + AI opportunity backlog |
| customer-facing regulated AI | disclosure script + advice boundary matrix + complaint/escalation workflow + monitoring dashboard |
| structured output / constrained decoding | schema registry + validator matrix + tool payload ADR + schema eval suite |
| model routing / semantic cache | routing policy + cache governance ADR + cost-quality frontier report + golden route eval |
| SWE-bench / WebArena / Agent benchmark | scenario library + tool sandbox spec + policy oracle + state verifier dashboard |
| Mamba / State Space Models | long sequence eval pack + model candidate scorecard + context strategy ADR |
| knowledge governance / ontology | source authority map + entity/relation ontology + freshness/permission SLO + GraphRAG fit memo |
| semantic layer / metrics architecture | metric contract catalog + lineage map + LLM-to-SQL guardrail + AI value dashboard |
| AI incident / postmortem / reliability | severity model + incident runbook + postmortem + corrective action register |
| AI product architecture strategy | capability-to-architecture map + platform strategy memo + funding gate + scale/stop review |
| EvalOps platform architecture | eval dataset registry + judge calibration report + release gate + production eval dashboard |
| decision intelligence / causal product | causal DAG + experiment/quasi-experiment design + uplift report + funding gate memo |
| enterprise integration / event-driven agent | OpenAPI/AsyncAPI contract + CloudEvents schema + idempotency ADR + HITL queue design |
| trust experience / product governance | trust calibration map + disclosure/refusal/escalation UX + complaint/appeal workflow + over-reliance controls |
| embeddings / vector search | embedding scorecard + vector index ADR + hard-negative retrieval eval + RAG release gate |
| CLIP / multimodal embeddings | multimodal taxonomy + image-text eval set + threshold calibration + privacy control map |
| diffusion / latent diffusion | generative media policy + asset registry + brand/safety/rights eval + human approval workflow |
| GNN / graph learning | graph schema ADR + temporal eval plan + risk workflow map + explanation design |
| case drill | PRD + workflow + eval + risk + business case |
6. 12 周执行节奏
| 周 | 主线 | 最低产出 |
|---|---|---|
| 1 | AI 2026+ 全景 + Transformer | capability map、Transformer 一页纸 |
| 2 | RAG + 企业知识架构 | RAG ADR、5 条 gold questions |
| 3 | Agent + tool governance | agent workflow、tool risk catalog |
| 4 | RLHF / alignment + product policy | alignment policy、拒答/升级示例 |
| 5 | CoT / eval thinking | reasoning task eval、rubric |
| 6 | LoRA / PEFT / adaptation | RAG vs fine-tuning memo |
| 7 | inference optimization | cost/latency/SLO map |
| 8 | LLM-as-Judge / EvalOps | judge rubric、human review loop |
| 9 | AI BA case drill | AS-IS/TO-BE、requirements-to-eval |
| 10 | AI architecture drill | C4、sequence、context engineering ADR |
| 11 | governance / operating model | control pack、RACI、runbook |
| 12 | portfolio / interview | executive memo、storyline、evidence map |
7. 单个案例的完整交付包
建议每个 flagship case 都形成下面 12 个文件或片段：
| 序号 | Artifact | 证明能力 |
|---|---|---|
| 1 | Problem framing | 能定义问题而不是直接套 AI |
| 2 | Stakeholder evidence map | 能识别真实使用者、审批者、反对者 |
| 3 | AS-IS workflow | 能画出现实流程 |
| 4 | TO-BE workflow | 能设计人机协作流程 |
| 5 | Requirements-to-eval matrix | 能把需求变成测试 |
| 6 | Data readiness pack | 能判断数据是否够用 |
| 7 | Architecture ADR | 能解释技术取舍 |
| 8 | Risk/control pack | 能控制越权、幻觉、隐私、合规风险 |
| 9 | Eval architecture | 能定义上线门禁 |
| 10 | Operating model / RACI | 能说明上线后谁负责 |
| 11 | Business case | 能解释投资价值 |
| 12 | Executive decision memo | 能推动决策 |
如果时间有限，先做 5 件：
- Problem framing。
- AS-IS / TO-BE workflow。
- Requirements-to-eval。
- Architecture ADR。
- Executive memo。
8. 金融零售案例优先级
建议优先做这些案例，因为它们最能发挥已有金融零售经验：
| 优先级 | Case | 为什么值得做 |
|---|---|---|
| 1 | AML alert triage copilot | 金融合规、高风险、流程复杂，能展示 BA/PM/架构综合能力 |
| 2 | KYC policy assistant | RAG、权限、引用、合规边界清晰 |
| 3 | Payment dispute assistant | 涉及证据聚合、工单、SLA、客户体验、风控 |
| 4 | Credit underwriting document assistant | 涉及信贷、审查、解释、人工审批 |
| 5 | Retail customer service knowledge copilot | 易解释、易 demo、可做 adoption 指标 |
| 6 | Fraud investigation workbench | 适合 agentic workflow 和 human approval |
| 7 | Compliance reporting assistant | 适合文档生成、审计、版本控制 |
| 8 | Retail inventory exception analyst | 跨到零售供应链，展示复合背景 |
9. 每周复习问题
每周结束时，用这些问题检查是否只是“看完”，还是已经变成能力：
- 我能否用 30 秒向业务负责人解释这个概念？
- 我能否用 2 分钟向 CTO 解释架构取舍？
- 我能否写出一个可执行的 eval？
- 我能否指出这个方案最可能失败在哪里？
- 我能否说明需要哪些数据和权限？
- 我能否定义上线前 stop rule？
- 我能否说明上线后谁负责？
- 我能否把它放进一个金融零售案例？
- 我能否把它转成一页 executive memo？
- 我能否在面试追问下 defend 这个方案？
10. 不要陷入的误区
| 误区 | 更好的做法 |
|---|---|
| 只学模型名字 | 学机制、边界、架构影响 |
| 只写总结 | 每篇笔记转成一个 artifact |
| 只做 chatbot | 训练 workflow、tool、eval、governance |
| 把 RAG 当搜索框 | 设计知识治理、权限、版本、引用、freshness |
| 把 Agent 当自动化万能工具 | 明确工具权限、状态、停止条件、human approval |
| 把 eval 当上线前测试 | 设计 offline、pilot、production monitoring 闭环 |
| 用 ROI 包装幻想 | 用 baseline、target、evidence、stop rule 管理不确定性 |
| 只准备技术答案 | 同一案例准备 BA、PM、Architect 三种表达 |
11. 未来继续扩展的方向
后续可以继续新增这些资产：
| 方向 | 可能文件 | 价值 |
|---|---|---|
| 经典论文继续精读 | mechanistic interpretability、world models、AI safety cases、agent benchmark design | 在 memory 和 multi-agent 之后继续补 AI 底层逻辑 |
| 更强案例训练 | docs/AI_ADVANCED_CASE_DRILL_WORKBOOK_60_DAYS.md 后续可扩展 flagship case artifacts | 把 60 天训练产出固化成可展示作品集 |
| 高管沟通 | docs/AI_REGULATOR_EXAM_SIMULATION_PACK.md 后续可扩展 incident board briefing | 训练监管问询、审计抽样和事故汇报 |
| 监管与合规 | AI regulatory change tracker | 形成持续法规雷达和 impact review |
| 数据产品 | AI data product portfolio case studies | 把数据产品能力做成 AML/KYC/客服/信贷作品集 |
| 平台产品 | AI platform PM playbook | 补平台、内部门户、shared service |
| 工具安全 | docs/AI_PLATFORM_SECURITY_GATEWAY_LAB.md 后续可扩展 red-team case library | 补 agent 工具调用、权限、沙箱和审计 |
| 生产运营 | docs/AI_OBSERVABILITY_COST_SLO_PLAYBOOK.md 后续可扩展 AI service review board pack | 补 trace、质量、成本、SLO、incident 和 adoption |
| 协议集成 | docs/AI_AGENT_PROTOCOLS_MCP_A2A_PLAYBOOK.md 后续可扩展 MCP server portfolio | 补 agent 工具生态、能力发现、集成治理和供应商风险 |
| 能力评估 | recurring capability review log | 把 C1-C14 rubric 变成月度复盘证据 |
12. 最小日常动作
每天只要做 60-90 分钟，也按这个顺序：
- 读 20 分钟：只读一个概念。
- 画 15 分钟：画机制图或流程图。
- 写 20 分钟：写一个 artifact 小片段。
- 讲 10 分钟：写 30 秒和 2 分钟表达。
- 复盘 10 分钟：记录一个风险、一个 eval、一个面试追问。
长期有效的学习不是“多看”，而是把每个概念都转成：图、决策、eval、case、表达。