| benchmark defaults | match durability and batching |
| average latency only | p99 during failure |
| ignore retention | replay and disk pressure |
| ignore operations | upgrade, rebalance, restore |
| one synthetic workload | representative payload and skew |
| Evaluate ordering scope, backpressure, redelivery, partition/queue limits, client maturity, and team competence. | |
| ## Test yourself | |
| 1. Why can batching distort latency comparisons? | |
| 2. What failure recovery should be timed? | |
| 3. Which workload facts dominate broker choice? | |
Continue to professional.md. | |