TL;DR of the paper, generated automatically.

A new benchmark tests whether LLMs can spot when simulator limits break a design pick.

• Benchmark: Siris pairs workloads where a simulator limit does or doesn't change the right design.
• Result: Naming a limitation doesn't guarantee the right choice. Even top models sometimes drop valid picks.
• Self-correction: A review turn fixes some choices but introduces new errors in others.