Twelve Items Before Wheels-Up: A Preflight Discipline for Software Releases
Photo: Lockheed Aircraft Company, CC0, via Wikimedia Commons
Commercial pilots in the United States complete a preflight inspection before every single flight, whether they are flying a Cessna out of a regional strip in Kansas or a widebody out of Atlanta. The routine exists not because pilots are careless but because the cost of a missed item is measured in lives. Software launches rarely carry that stakes profile, yet the pattern of preventable failure is strikingly similar: the disaster almost always traces back to something that could have been checked.
At GleegerTestFlight, we have reviewed enough troubled rollouts to recognize the pattern. What follows is a twelve-point release checklist modeled on aviation preflight discipline — each item maps to a specific failure mode we have watched sink otherwise solid products.
1. Rehearse the Rollback, Not Just the Release
Most teams can describe their rollback plan. Far fewer have executed it under pressure. A rollback that has never been rehearsed is a hypothesis, not a plan. Run the drill in staging with the same time pressure you would face at 2 a.m. on launch night. The failure it prevents: the frozen database migration that cannot be reversed while revenue bleeds.
2. Run Smoke Tests on Physical Devices
Emulators are convenient and consistently misleading. They cannot reproduce thermal throttling on a three-year-old Android phone or the memory pressure of a device with 40 browser tabs open. A real-device smoke test across your top ten US market devices catches the class of defect that lab environments structurally miss.
3. Verify Feature Flags Default to Safe States
Confirm that every flag ships in its most conservative configuration. The failure it prevents: the experimental checkout flow that reaches 100% of users because someone forgot to flip the default.
4. Confirm Observability Before Traffic
Dashboards, alerts, and log pipelines must be live before the first user arrives — not forty minutes after. If you cannot see the system during the first hour, you are flying without instruments.
5. Load-Test the Actual Launch Pattern
Synthetic traffic shaped like a steady stream tells you almost nothing. US launches arrive as a spike — a morning email blast, a press embargo lifting at 6 a.m. Eastern. Test against that shape.
6. Validate Third-Party Dependencies
Your payment processor, identity provider, and analytics vendor are all single points of failure you do not control. Confirm their status, their rate limits, and your graceful degradation paths.
7. Walk the Critical Path by Hand
Automation coverage is valuable, but a human should manually complete signup, purchase, and support contact on launch morning. This is the five-minute check that catches the broken email domain or expired API key that forty automated suites waved through.
8. Check the App Store and Compliance Clocks
Review cycles, DNS propagation, and certificate expirations do not negotiate. A launch date set without buffer against these external clocks is a launch date you do not actually control.
9. Brief the Support Team
Frontline support is your first incident-detection system. If they learn about the release from a customer tweet, you have lost your earliest warning signal.
10. Freeze the Code Early Enough to Matter
A code freeze three hours before launch is not a freeze; it is a countdown. Late changes bypass the very validation the checklist exists to protect.
11. Assign Explicit Launch Roles
Who makes the call to roll back? Who talks to customers? Who is watching the error budget? Ambiguity during an incident costs minutes you cannot recover.
12. Schedule the Debrief Before You Launch
Commit now to a post-launch review while the details are fresh. Teams that skip this step repeat the same failure modes release after release.
Discipline Is Cheaper Than Downtime
None of these items is technically difficult. That is precisely the point — aviation's insight was that safety comes from systematically eliminating the easy omissions. A team that runs all twelve checks before wheels-up will still encounter problems. But it will not be surprised by the ones it could have seen coming.
