Performance budgets that actually hold

Every studio claims to care about performance. The test is what happens when performance and a beautiful idea collide at 4pm on a Thursday, three weeks before launch. If beauty always wins, you don't have a performance budget, you have a performance aspiration.

The budget is a contract

Ours is written into the statement of work: Lighthouse ≥ 95 in all four categories on a mid-range phone over throttled 4G, interaction latency under 200ms, no layout shift after first paint. Because it's contractual, it has standing in every design debate. The budget can't be charmed.

What we've learned the hard way

• Fonts are the silent killer: two families, variable where possible, subset always.
• The hero can be expensive if everything else is cheap, spend your bytes where the eyes are.
• Lazy-load below the fold, but never lazy-load the fold.
• Third-party scripts are guests: they arrive after the party is set up, or not at all.
• Measure on the £180 phone your users actually own, not the laptop you do.

A performance budget only works if someone can lose an argument to it.

The budget has killed some ideas we loved. Every one of those funerals was cheaper than the slow site would have been.