The performance budget that actually survived three product teams

A performance budget expressed as a single number — "LCP under 2.5s" — sounds precise and is usually useless in a room with a product deadline, because there is nothing to negotiate against. Somebody always wins by ignoring it.

What worked was tying the budget to specific, ownable line items: bundle size per route, third-party script weight, image payload per breakpoint. When a feature threatened the budget, the conversation became "which of these three things do we spend the headroom on" instead of "do we care about performance this sprint."

The LCP and TTI improvements came from that structure, not from one optimization pass. Budgets that are enforced automatically in CI, with a clear owner per line item, survive roadmap pressure. Budgets that live in a slide deck do not.