On a project to revamp our checkout flow, we were two weeks from launch and QA flagged that the new UI was slower on older mobile devices. My manager wanted to ship anyway and address it post-launch, but I pushed to delay by three days so we could optimize the rendering path. I profiled the bottleneck, made the fix, and we shipped on the revised date. Customer satisfaction scores for the flow improved after launch, and we didn't get the spike in support tickets we had seen with previous releases.
Six days before a major feature launch, our data pipeline silently started dropping roughly twelve percent of order-status events. Customers weren't seeing real-time updates. My manager wanted to ship and patch post-launch to protect the date. I disagreed — I pulled the error logs, quantified the drop rate, and walked him through what that meant for a customer waiting on a delivery. He pushed back hard. I held the position, documented the customer impact in writing, and escalated to the team lead with data. We delayed four days. I wrote a detection runbook so the same failure mode could never silently reach launch again.