Journal

Why a crash-free session number can lie

8 June 2026

Person reviewing papers and a laptop at a desk

Teams like a single percentage. Crash-free sessions are easy to put on a slide. They also ignore the shopper whose screen froze on pay, who force-quit, and who never generated a native crash.

We still look at the percentage, because a sudden drop after a version is a real signal. We refuse to call a release “stable” on that number alone. Freeze checks exist for this reason: they ask whether the main thread was busy, whether resume was killed, and whether those events line up with support tickets.

If your crash-free rate is calm and your one-star comments mention “hang,” bring the comments to the review. They are part of crash and stability analytics even when the console looks polite.