A Performance Budget That Survives Contact With a Real Store
Performance work that is not enforced gets undone within a quarter. Here is the budget I use, the numbers behind each limit, and how to fail a build on it.
| Metric | Budget | Why this number |
|---|---|---|
| LCP, mobile p75 | 2,500 ms | Core Web Vitals "good" threshold |
| INP, mobile p75 | 200 ms | Core Web Vitals "good" threshold |
| CLS | 0.10 | Core Web Vitals "good" threshold |
| Theme JavaScript, compressed | 100 KB | Roughly 300 ms parse and execute on a mid range Android |
| Theme CSS, compressed | 60 KB | Above this, critical CSS extraction becomes necessary |
| Third party JavaScript | 150 KB | Enough for four typical apps |
| Total page weight, product page | 1.2 MB | Around 4 seconds on a slow connection |
| DOM nodes | 1,500 | Chrome's own guidance for style recalculation cost |
| Render blocking requests | 2 | One stylesheet, one font, nothing else |
| Liquid render time, p75 | 200 ms | The part of server response you control |
| Long tasks after load | 0 over 200 ms | Anything longer is a visible stall |
Related reading: The Flame Graph That Showed the Regex Inside the JSON Parser explores a connected angle.
Related reading: What a Connection Pool Is For, and the Setting That Ruins It explores a connected angle.
Related reading: Why LLM Latency Is Bimodal and the Average Hides It explores a connected angle.
Published on debugly.dev by Rohit Bhadani | Founder of Krova Cloud