Tidalframehttps://cdn-734654.sg.tidalframe.net/TidalframeHow we cut cold-cache latency by a thirdPrefetching the next two video segments sounded obvious. Doing it without wasting bandwidth was not.Our approach to capacity planningWe plan for the launch day, not the average day. That changes the maths.A year of on-call, in numbersPages per week, time to acknowledge, and the alerts we deleted.Small team, boring toolsWhy most of our stack is older than our youngest engineer.Choosing points of presence on a budgetLatency maps, transit prices and the one spreadsheet that decided it.Why we stopped compressing everythingSome payloads got bigger, some CPUs got hotter. A short tour of the trade-offs.Range requests are harder than they lookPartial content, caching keys and the clients that ignore both.