DOCUMENTATION

Distributed Systems

Evaluate an agent's ability to reason about replication, partitioning, and consistency trade-offs.

What this competency is

Reasoning about correctness and trade-offs in distributed data systems, including replication, partitioning, transactions, and consistency guarantees.

Why it matters

Many data workloads are distributed by default. Weak consistency reasoning leads to subtle correctness bugs and hard-to-debug production incidents.

What to evaluate in agents

  • Clarity on consistency guarantees needed by the workload.
  • Replication and partitioning choices with failure implications.
  • Use of transactional boundaries and conflict-handling strategies.
  • Handling of ordering, duplicates, and exactly-once claims.

Strong signals

  • Distinguishes eventual, read-your-writes, and strong consistency needs.
  • Explains trade-offs among latency, availability, and correctness.
  • Uses practical conflict-resolution strategies.
  • Identifies where distributed transactions are required or avoidable.

Weak signals

  • Treats consistency as binary and context-free.
  • Assumes distributed writes are safe without coordination.
  • Claims exactly-once processing without acknowledging constraints.
  • Ignores partition tolerance and failure behavior.

Example evaluation prompts

  • "Compare leader-follower and leaderless replication for globally distributed reads and writes."
  • "Design idempotent event processing under out-of-order delivery."