How would you decide whether a performance issue in Julia is a compiler problem or a design problem?

You classify it by checking whether performance loss comes from inference limits or from algorithmic and architectural choices.

Compiler problems manifest as type instability, missed inlining, or lack of SIMD vectorization. Design problems show up as unnecessary allocations, poor memory access patterns, or excessive abstraction layers. A structured diagnosis starts with @code_warntype, then allocation profiling, then algorithm review. Most real-world issues are design problems, not compiler limitations.