What crash-free rate actually tells you
Crash-free rate is the share of app sessions that finished without crashing: (sessions − crashes) ÷ sessions × 100. It exists because a raw crash count is close to meaningless on its own — 500 crashes sounds bad until you learn it happened across 50 million sessions, and sounds catastrophic if it happened across 5,000. The rate normalizes for traffic so you can compare builds, releases, and platforms on equal footing.
What counts as good
We run production monitoring and crash-ingestion ourselves, so here is the honest version, not a marketing number: most teams treat 99.5%+ crash-free sessions as healthy and ship-worthy. Between 98% and 99.5% is a gray zone — plenty of shipped apps live there, but it is usually worth knowing exactly what is crashing before you roll out to everyone. Below 98% is where most teams start holding releases, because at real scale that is thousands of users hitting a crash. These are general guidelines from how the industry talks about release health (Play Vitals, Crashlytics-style dashboards) — not a certification, and a niche or early-access app may reasonably tolerate a lower number than a payments app.
Sessions vs. users — they tell different stories
Crash-free sessions answers “how often does a crash interrupt usage.” Crash-free users answers “what share of my install base has hit a crash at all.” The user number is almost always lower and scarier, because one unlucky user who crashes every time they open a specific screen drags the user rate down even if the vast majority of their other sessions were fine. Look at both if you can — sessions for day-to-day release health, users for how widespread the pain actually is.
Why this is the same math CleverQA runs on a real release
Crash-free rate is a release-health signal that tells you whether a build is safe to roll out further — a dropping crash-free rate is reason to hold, independent of anything else looking fine. This calculator runs the exact same crash-free sessions formula CleverQA computes from your connected Sentry session data for a real release, so what you see here is not a simplified stand-in — it is the calculation, minus the automatic data feed.