Database Design Decisions

This page captures rationale for normalization, indexing, and storage choices.

• Service-owned schemas prevent cross-service coupling and keep ownership clear
• Normalization for market observations avoids repeated raw payloads
• Partitioning is planned for time-series tables at scale
• Large model files stay in DVC/MLflow (or S3); DB stores metadata only
• Job trace tables are append-only for auditability