[Intermediate] — the gatekeepers, and why mobile release engineering is cautious by necessity.
The kid version first
You can’t just deploy. You submit a build, someone reviews it, and then users decide whether to install it.
So a bug can be live for days before a fix reaches anyone — and old versions live on phones for years. Every release decision in mobile follows from those two facts.
The pipeline
build → sign → upload → REVIEW → staged rollout → monitor → (halt or continue)
Signing is where iOS newcomers lose days. Certificates, provisioning profiles, capabilities and identifiers must all agree. Xcode’s automatic signing works; use it, and use fastlane match for teams so certificates aren’t passed around manually.
Android is simpler: one upload key, and Play App Signing holds the real key. Turn that on — losing your signing key without it means you can never update your app again and must publish a new listing.
Review
| Apple | ||
|---|---|---|
| Time | Hours to a few days | Usually hours |
| Strictness | High, and human | Lower, more automated |
| First submission | Slowest, most scrutiny | Slower than updates |
What actually gets rejected:
- Crashes on the reviewer’s device — the most common cause, and usually avoidable
- A login wall with no way in. Provide a demo account in review notes, always
- Broken or missing features relative to your description
- Payments outside the store’s system for digital goods — the rule with the most litigation attached, and still enforced
- Privacy declarations that don’t match behaviour → privacy
- Missing account deletion if you support account creation
- Permissions requested without a clear purpose string
When rejected: read the actual guideline cited, fix it, and reply in Resolution Center. You can appeal, and appeals do succeed — reviewers make mistakes. Don’t resubmit unchanged.
Budget for review. Never promise a launch date that assumes same-day approval.
Staged rollout — the most important habit
Never release to 100% at once.
1% → watch crash-free rate → 5% → 20% → 50% → 100%
Google Play supports this natively and lets you halt a rollout. Apple’s phased release rolls out over 7 days and can be paused.
Watch crash-free users and ANR rate at each step. A regression caught at 1% affects a hundredth of the damage. This is the closest thing mobile has to a rollback, because there isn’t one — you can only halt and ship a fix.
The two switches you need before you need them
1. Feature flags / remote config. The only way to turn off a broken feature without a release. Set this up before launch — Firebase Remote Config or equivalent. It converts “emergency release, three days” into “toggle, thirty seconds.”
2. Forced update. An endpoint returning a minimum supported version, and a blocking screen below it. You will need this — a security fix, a breaking API change, a broken build. Retrofitting it is impossible, because the users you need to reach are on the version that lacks it.
Use it sparingly. Blocking users is hostile; a “recommended update” prompt should be the default.
Versioning and API compatibility
Old versions never die. A meaningful share of users won’t update for months, some never.
- Never break your API for old clients. Add fields, don’t remove them → networking
- Clients must ignore unknown fields
- Know your actual version distribution — both consoles show it, and it’s usually worse than you’d guess
- Database migrations must chain across skipped versions. A failed migration on launch is a crash loop that permanently bricks the app for that user — test upgrades from several old versions, not just the last
CI/CD
Automate it — mobile release is too error-prone by hand:
- Fastlane is the standard for both platforms: build, sign, screenshot, upload, release notes
- CI: GitHub Actions, Bitrise, Codemagic. iOS builds need macOS runners, which is a real cost
- Distribute test builds — TestFlight (iOS), Play Internal Testing / Firebase App Distribution
- On every PR: build, unit test, lint. On main: upload to internal testing automatically
- Automate version bumps and changelogs → CI/CD
Monitoring — the part that’s easy to skip
You cannot SSH into a user’s phone. Instrumentation is all you get:
- Crash reporting — Crashlytics or Sentry, with symbol upload in the release pipeline or your stack traces are unreadable
- The metric to watch: crash-free users. Both stores rank on it, and it’s the clearest health signal
- ANRs (Android) — the app frozen more than 5s. Tracked by Play and penalised
- Android Vitals / Xcode Organizer — real-device data on startup, battery and rendering → performance
- Alert on crash-rate regressions per version, and halt the rollout automatically if you can
Store listing
The part engineers ignore and that determines whether anyone installs:
- Screenshots are the main conversion lever — far more than the description
- The first two lines of the description are what’s read
- Ratings prompts: use the native API, and ask after a moment of success, never on launch or after an error
- Respond to reviews. It measurably improves ratings, and it’s where you find bugs nobody reported
- Localise the listing at minimum, before the app itself
Key insight
Mobile has no rollback, so the entire discipline shifts to prevention and containment: staged rollouts to limit blast radius, feature flags to disable without shipping, forced update as the last resort, and never breaking an API that old clients depend on. Set up the flags and the version endpoint before your first release — they’re the two things you cannot add later, exactly when you need them.
Related
- permissions and privacy — declarations checked at review
- performance — the metrics the stores rank on
- CI/CD — the general discipline
- delivery — staged rollout in the wider sense
Source: [reference] — Aug 2026.
Software engineering
Infrastructure & security
Computer science
Hardware & engineering