perf-barlisted

# perf-bar ## Overview My performance/algorithmic bar. Usable standalone on a diff, fed into the **SPIKE** sub-phase of `eskills:task-flow` (approach selection), and invoked by its per-task review. It earns its own lens because bottlenecks are a recurring core challenge. (This is the 4th lens — the hard cap; a new lens must displace one.) ## Hot-path checklist - **Big-O where it matters.** Profile or reason about complexity on hot paths; an O(n²) loop on a small list is fine, on a hot path it isn't. Don't micro-optimize cold code. - **N+1 queries.** A query inside a loop over rows → batch it, join it, or prefetch. - **Payload size, not just row count.** What crosses the wire per operation? Big blobs, over-fetching, sending a whole row when a version number would do. ## Per-stack hot-spots Same lens, applied to the stack in play — the footgun I hit most in each: - **React / Vite:** wasted re-renders → memoize (`memo`/`useMemo`/`useCallback`), stable list keys, split context; virtualize long lists. Profile with the React DevTools profiler, don't guess. Rule-level detail (derive in render, derived-boolean subscriptions, `content-visibility`, `localStorage` discipline) in `references/react.md`; chunking, pre-bundling and what hashed chunks mean at deploy time in `references/vite-build.md`. - **Django:** ORM **N+1** → `select_related` (FK) / `prefetch_related` (M2M); fetch only needed columns (`.only()`/`.values()`); never run a query inside a template or loop. - **Flas