REAL PRODUCT TASK
Build, test, and improve an AI workflow.
Use a realistic product task, let the system choose a viable model strategy, then inspect the quality, latency, fallback behavior, and evidence behind the result.
Task type
Investigation
What the system optimizes
Correctness
Solve the actual task, not just produce plausible text.
Reliability
Recover from model/provider failures without leaking infrastructure errors.
Efficiency
Prefer the least expensive viable strategy within the task constraints.
Evidence
Keep the routing reason, evaluation result, and execution trace reproducible.
No model selection. The system owns candidate selection and fallback.