Global browser usage percentages are poor justification for any browser-feature decision. What a browser does across the whole web says little about its share of traffic on a given site, and that share can vary enormously from one site to the next. The figure worth acting on is what the people visiting your own site actually use.
The same reasoning extends past feature support. A generic analytics snippet yields generic answers: which pages are popular, how long people stay, which countries send the most traffic. Those are real user numbers, but they don't reveal how your site is actually doing. To learn how often Feature X gets used, how many files were uploaded this week, how many messages were sent, or how many times the star button was clicked, you have to write some JavaScript to capture and report each event. That takes deliberate thought up front and a bit of effort, but it produces the analytics that matter to you.
Synthetic Tests and Their Limits
That distinction matters just as much on the performance side. Most performance testing is synthetic: an automated browser loads the site and records timings, asset sizes, request counts and similar data. It's the kind of information that encourages micro-optimization — shaving one request, squeezing one asset a little smaller. The tools report bundle size, Largest Contentful Paint and Lighthouse scores readily enough, but these measure the loading process, not the experience of an actual user.
The alternative is Real User Monitoring: tracking real users on their actual devices rather than leaning on a single automated test. Beyond making logical sense, RUM unlocks data that synthetic testing can't produce.
Custom Metrics on Your Own Terms
First Input Delay, one of Google's Web Core Vitals that will soon affect SEO, is a case in point: collecting it requires JavaScript running on the page. Largest Contentful Paint is likewise an interesting attempt at a more meaningful metric. A start-render or first-paint measurement only tells you that something — probably — is appearing; that something may be nothing like the content the user came for. LCP guesses which element is the useful one and measures it.
Google has to guess, because it measures across a huge number of sites and has to stay generic. You don't. On your own site you can tell the tooling exactly which elements matter and record when they render. On this site, that would be the article body itself. With SpeedCurve's hero rendering time, for example:
<main elementtiming="article"></main>
<!-- or focus on the top of the page, like the "hero" timing suggests -->
<header elementtiming="hero"></header>
The result is a measurement specific to your site instead of a generic number.
FID deserves the same treatment. Rather than settle for it alone, fire a JavaScript event whenever something important to your site happens — on CodePen, when the editor is ready to use. This is User Timing, and it's a W3C spec:
Editors.init();
performance.mark("Editors are initialized.");
Effort-intensive analytics of this kind beat the standard fare. A performance budget that warns when JavaScript exceeds 200KB is useful; one that warns when a core feature of your app isn't ready until 1.4 seconds against a 1.1-second budget is substantially more so.
A practical note: charting the three Web Core Vitals on SpeedCurve required LUX, their RUM product, before FID could be added.



