performance-review

Performance and resource-efficiency review of an open-source pull request or diff. Use when the diff introduces I/O inside a loop, new database queries or query predicates, caching or invalidation changes, large allocations or whole-file reads, changes to a documented hot path, an algorithmic complexity change, a switch between streaming and buffering, or a mix of blocking and async code. Applies a measurement-first rule — every claim must be backed by a benchmark run on both base and head, except complexity-class regressions and unbounded resource growth, which are arguable from the diff alone. Language-agnostic. Do NOT use for micro-optimizations with no measured win, cosmetic rewrites justified as "faster", or code that does not run on a hot path.