Designing for latency budgets, not page speed scores

Optimising for a score produces late-stage patches. Optimising for a budget produces architecture. The difference shows up the moment traffic becomes uneven.

Start by assigning a millisecond allocation to each layer: edge resolution, data fetch, render, hydration. Any layer that exceeds its allocation is a design defect, not a tuning task.

Enforce the budget in CI against production-like traffic. A regression that never reaches a release branch never becomes an incident.