DOCUMENTATION

Data Modeling

Evaluate an agent's ability to model data and evolve schemas safely.

What this competency is

Designing data models and schemas that match business entities, preserve meaning, and support analytics and operational use cases over time.

Why it matters

Poor modeling creates brittle pipelines, ambiguous metrics, and costly downstream rework. Strong modeling improves correctness, discoverability, and long-term maintainability.

What to evaluate in agents

  • Ability to select appropriate model patterns (normalized, dimensional, wide tables, event models).
  • Handling of keys, grain, cardinality, and slowly changing dimensions.
  • Strategy for schema versioning and backward-compatible evolution.
  • Explicit assumptions about nullability, defaults, and constraints.

Strong signals

  • States table grain and primary keys clearly.
  • Separates raw, staged, and curated semantics.
  • Calls out schema evolution risks and mitigation.
  • Aligns model design with query patterns and serving needs.

Weak signals

  • Mixes unrelated entities into one table without rationale.
  • Ignores key constraints and deduplication concerns.
  • Treats schema changes as no-risk refactors.
  • Uses vague terms without defining business meaning.

Example evaluation prompts

  • "Design a warehouse model for orders, refunds, and subscriptions with historical tracking."
  • "Propose a migration plan from denormalized JSON events to curated analytics tables."