Every hundred milliseconds of delay is a small tax on trust. It compounds across a session, and users pay it whether or not they can name it.
Ship less, sooner
Render at the edge, stream the parts that are ready, and defer what the first screen does not need. The fastest request is the one you never make.
Measure what users feel
We budget for the metrics that map to perception: time to first byte, largest contentful paint, interaction latency. If a change regresses them, it does not ship.
Performance is not a phase at the end. It is a constraint you design within.