Cold start (startup time)

Measurement comes before any optimization, and only on release builds. On Android, adb shell am start -W and the Displayed line in logcat give TTID; reportFullyDrawn (callable from a native module once the screen is ready) gives TTFD. On iOS, the App Launch template in Instruments splits pre-main (dynamic framework loading, static initializers) from post-main. On the JavaScript side, performance.mark and the Performance panel in React Native DevTools show the time between bundle execution and the first React commit. A real low-end Android device is the reference, not the simulator.

The usual causes, in the order I find them: module-level work (SDK initialization, storage reads, large image requires) executed before the first render; a bundle that loads every screen when inline require or lazy-loaded routes could defer them; a cascade of blocking network requests behind the splash; heavy Intl polyfills; and on iOS a dynamic use_frameworks! that multiplies the frameworks loaded at pre-main. Hermes and its bytecode solve the parsing part, not these five.

What I look at first in a project: what runs between index.js and the first screen, how many providers are nested at the root, and whether the splash is hidden on a precise event (data ready) rather than a timer.