A 98 Lighthouse score does not mean your site is fast
Lab scores and field data disagree constantly. Here is what we measure instead, and why the difference matters for Nepali users.
Sneha Karki
CT Tech Nepal
Lighthouse runs on a fast connection with a simulated throttle. Your users are on a mid-range Android phone with two bars of Ncell signal on a moving bus. Those are not the same test, and the gap between them is where real performance lives.
For Nepali audiences we track three field metrics: Largest Contentful Paint on 3G-class connections, Time to Interactive on a throttled mid-range device, and total transferred bytes on first load. The third one is the most predictive of perceived speed in our experience, because it correlates directly with how long the blank screen lasts.
The techniques that move those numbers are unglamorous: ship fewer bytes of JavaScript, defer everything below the fold, serve correctly sized images in modern formats, and avoid client-side rendering for content that could be server-rendered. We routinely cut first-load payload by 60% just by auditing what was actually shipped versus what was assumed to be needed.
Use Lighthouse as a smoke test, not a goal. If it is below 90, something is wrong. If it is 100, you still do not know how the site performs for the people paying you.
Key takeaways
- Lab scores and field performance measure different things
- Track transferred bytes — it predicts perceived speed
- Server-render content that does not need to be client-side
- Use Lighthouse as a floor, not a finish line