Design a compliant multi-tenant analytics platform

## Prompt Design a **multi-tenant data analysis platform** for **government and enterprise** customers. The interviewer emphasizes that the key challenge is **not scale**, but: - **Extremely complex permissions / access control** - **Compliance requirements** (data handling, retention, approvals) - **Audit logging** (tamper-evident, explainable) - **Data lineage** (trace data origins and transformations) They repeatedly stress: > “Assume mistakes will happen. How do we detect and recover?” ## Requirements (make reasonable assumptions) 1. Multiple tenants (organizations) share the platform. Some customers may require strict isolation. 2. Users include analysts, data engineers, auditors, admins, and external reviewers. 3. Data sources: batch ingestion and/or uploaded datasets; transformations and query/BI-style analysis. 4. Authorization must support complex real-world rules (attributes, projects, clearance, purpose, time, geography, approvals, etc.). 5. Platform must provide: - Strong authentication + authorization - Fine-grained data access enforcement (dataset/table/column/row and API-level) - Immutable audit trails and reporting - End-to-end lineage across ingestion → transforms → derived datasets → queries - Detection and recovery mechanisms for inevitable human/config mistakes ## Deliverables Walk through: - Architecture (control plane vs data plane) - Tenant isolation model - Permission model (how policies are expressed, evaluated, and enforced) - Audit logging + evidence generation - Lineage design - “Mistakes will happen”: detection signals, blast-radius control, rollback/recovery, and operational processes - Key edge cases and trade-offs