Failure-mode review
Here is our architecture: {{doc}}. List the ten most likely failure scenarios under partitions, retries, clock skew and partial outages. For each: trigger, blast radius, detection, and the smallest design change that prevents it.
Build software
Best agent for distributed systems
Concurrency, consistency, queues, caching, scaling and failure handling across services.
Full ranking
Task index, 0 to 100. Bars are scaled to the field, so small gaps look small.
- 1
- 2
- 3
- 4
- 5
Key takeaways
- 1Fable 5.1 leads: these problems reward deep, careful reasoning.
- 2Agents are great at analysis and design reviews; be cautious with autonomous changes.
- 3Always ask for failure scenarios explicitly.
Recommended setup
Use a frontier model as a design reviewer with architecture docs, traces and incident history; apply changes through normal PR review.