We had a locked launch date and engineering flagged three edge cases two days out. I called them low-frequency, low-severity, and pushed to ship. Marketing went ahead, we launched on time, and support tickets spiked 40% that first week.
Shipped on assertion alone; the support spike reveals the severity framing was motivated, not analytical.
I pushed to delay. The edge cases hit roughly 15 percent of first-time users, and I knew that was the retention cliff. I made the call, absorbed the marketing rescheduling cost, and the fixed version shipped six weeks later with a 23 percent better day-seven retention.
Identified a real retention risk and held the line, but accepted the full six-week cost without evidence of triage or alternatives.
I pulled the engineering lead and marketing director into a room and mapped out exactly which edge cases hit our highest-traffic user paths. Three of them did. We delayed two weeks, fixed those three, shipped the rest. Marketing adjusted. Nobody loved it.
Diagnosed which edge cases actually mattered, negotiated a proportionate delay, and shipped a defensible middle path.