行业知识 · v2.3.0 · 资料核对 2026-10-03
AI Product Requirements Document
在交互知识库中阅读下载 Markdown 原文
1. Executive decision
- Problem and opportunity:
- Proposed product:
- Decision requested:
- Evidence strength:
2. Users and current workflow
- Primary/secondary users:
- Buyer, administrator, beneficiary, and risk owner:
- Trigger, task, steps, handoffs, and outcome:
- Current baseline: time, cost, quality, failure, adoption:
- Research evidence and known gaps:
3. Goals and non-goals
Goals
Non-goals
Stop conditions
4. Why AI
5. Scope and user journey
- In scope / out of scope:
- Happy path:
- Ambiguous input:
- No-answer / refusal:
- Partial failure:
- Correction / undo:
- Human escalation:
6. Requirements
7. AI behavior contract
- Allowed inputs and source trust:
- Required context and freshness:
- Expected output schema:
- Evidence / citation behavior:
- Uncertainty and refusal:
- Tool actions and authorization:
- Human confirmation and maximum autonomy:
- Prohibited behavior:
8. Data, privacy, and governance
- Data sources, authority, purpose, and owner:
- Sensitive fields and minimization:
- Access, tenant isolation, retention, deletion, and audit:
- Training / feedback use:
- Fairness, transparency, copyright, or domain review:
9. Success and evaluation
- Evaluation dataset and sampling:
- High-risk / long-tail coverage:
- Online experiment:
- Regression plan:
10. Non-functional requirements
- Latency / throughput / availability:
- Capacity and budget:
- Observability and traceability:
- Accessibility / localization:
- Fallback, recovery, and portability:
11. Release and operations
- MVP learning goal:
- Phase gates and rollout stages:
- Go/No-Go approvers:
- Monitoring, alerts, and sampling:
- Rollback triggers and mechanism:
- Support, incident, and operations owner:
12. Dependencies, risks, and decisions
13. Traceability
14. Appendix
- Confirmed facts:
- Labeled assumptions:
- Open questions:
- Decisions and rejected alternatives: