Crash-Free Sessions: What Founders Should Ask
Crash-free sessions is a stability ratio: sessions that did not crash, divided by sessions that ran. Founders should treat it as a question set, not as a trophy. Apple, a third-party SDK, and a hallway demo can all produce different numbers from the same week. This is the diligence list we use at Croncode before we call a binary ready.
Ask how crash-free sessions are counted, who owns crash symbols, and which build the number comes from. A percentage without a method is a slide, not a gate.
What does crash-free sessions actually measure?
A session usually starts when the app becomes active and ends after a period of background idle or a terminate. A crash is an unexpected exit the system or your reporter classifies as a crash. Force quit is not a crash. A watchdog kill might be. If two people argue about a percentage, they are often arguing about those definitions.
Ask the studio to write the definition. Then ask which reporter they trust for the store gate: Xcode Organizer, MetricKit, or a vendor dashboard. Mixing them in one slide hides the fight. Shipping culture: how Croncode ships.
Which questions should you ask before you sign?
- Which build number is in the chart, TestFlight or App Store?
- Are dSYMs uploaded automatically, and who has the Apple org that receives them?
- How fast can you reproduce the top crash group on a device we name?
- What is the kill rule for a crash that hits the first job?
- Who reads the report on Monday, not only after a one-star review?
Account ownership belongs in that list. If symbols sit in a personal Apple ID, you will lose them when a contractor leaves. Write the owner now. Related: who owns the App Store account. Hiring checks in this city: Istanbul diligence.
When is a crash a product bug, not a device story?
| Signal | Treat as product | Treat as noise first |
|---|---|---|
| Same stack on two OS versions | Yes | No |
| One jailbroken device | No | Yes, isolate |
| Crash on the first job path | Yes, block release | Never “later” |
| Crash in a rare locale string | Yes, if you ship that locale | Only if that locale is not-now |
Consumer jobs make this concrete. If YourStyle died while saving a garment photo, that is a first-job crash. If Luviport died while confirming a transfer window, same rule. A crash in a settings toggle you can live without is still a ticket. It is not a store blocker unless it is frequent.
How should a small studio report the number?
Report seven and twenty-eight days. Name the build and the session count so a 100% on twelve sessions cannot pose as a launch metric. We do not publish invented crash-free rates. Do not accept invented rates from a vendor. Catalog: what Croncode builds.
If the team is lean, one person owns the Monday read. Write the name. Hours: small team, multiple products. Founder: Goktug Unver. Co-founder: Yusufhan Sokur.
What mistakes look like stability work?
- Screenshot of a green dashboard with no build number: refuse it.
- Turning off crash reporting to “look clean”: that is fraud against yourself.
- Ignoring memory jetsam because it is not labeled crash: users still lose the session.
- Shipping an AI path with no timeout UI: a hang is a failed session even when the process lives. Fail-safe scoping: AI features that fail safe.
Quote packets should mention the reporter and the Apple org. Use what to send for a quote. If you are choosing a vendor, add this page to how to choose an iOS studio. Custom work: custom software. Hire path: hire the studio.
What does a release gate look like in practice?
Write a kill rule as counts plus a ratio so a quiet week cannot fake health. Example: do not promote a build if the top first-job crash group has more than N users in seven days, or if you cannot reproduce it. A founder who only hears about crashes from reviews is late. Reviews: read reviews as a team. Support: support email. Journal: /en/journel/. Contact: contact.
Frequently asked questions
Is 99.9% crash-free a requirement?
No universal number is a requirement. Apple publishes ranges that change. Ask for method, build, and first-job impact instead of a slogan percentage.
Do I need a paid crash tool on day one?
You need symbolicated reports and a weekly owner. Organizer can be enough for a small TestFlight. Add a vendor when volume or roles demand it.
What if the crash is in a third-party SDK?
It is still your listing. Update, isolate, or remove the SDK. Privacy labels also change when SDKs change.
Should I delay launch for a rare crash?
If it hits the first job, yes. If it is a one-device stack you cannot reproduce and the job completes, ship and watch the next seven days.
Who at Croncode reads crash reports?
Engineering in Istanbul owns the weekly read. Names on the team pages: Goktug Unver, Yusufhan Sokur, Ozgur Ismet Ozkan.
