The crash-free rate is the first number we open after a launch, before downloads and before ratings. A user who crashes once forgives you. Twice, they leave.

Most crashes are not exotic. They cluster around a handful of causes: a null value from an api that changed, a permission the user denied, a device model nobody tested. Good crash reporting turns each one from a mystery into a line of code.

We instrument crash reporting before the first store build, not after the first bad review. That way the data is waiting when we need it.

Past ninety-nine and a half percent, the rest are edge cases worth chasing one by one. Below it, something structural is wrong, and no feature is worth shipping until it is fixed.