performance-audit-standard

Identify hot paths first: request handlers, event processors, loop bodies, and timers that run frequently. Then determine the actual Big O complexity of each path and use profilers like node --prof, cargo flamegraph, or pprof to confirm before optimizing.

Replace O(n) array.includes() or array.find() calls with Set.has() or Map.get() for O(1) lookups. Build the Set or Map once outside the loop, turning O(n*m) operations into O(n+m), which can yield 1000x speedups on large collections.

N+1 queries happen when code fetches a list then queries details per item in a loop. Fix by using your ORM's include, with, or Preload feature to fetch related data in one or two queries instead of N+1 round trips.

Skip optimization when code runs under 100 times per second with small inputs, when the gain is under 2x but adds significant complexity, or when the bottleneck is I/O bound rather than CPU. Always profile first to confirm the code is actually hot.

Target LCP under 2.5 seconds, INP under 200 milliseconds, CLS under 0.1, and initial gzipped JS bundle under 170KB. Common wins include route code-splitting, lazy-loading below-fold images, and tree-shaking icon libraries.