The Archer's Unseen Anchor: On the Misplaced Faith in a Single Performance Score
In the quest for a faster web, we have, understandably, sought a single metric to crown our efforts. A number to present to stakeholders, a score to boast about, a definitive measure of ‘done’. For years, Lighthouse’s Performance Score has been that golden idol. We draw our bow, aim for a 90 or above, and release, believing that hitting that target means we have built something good. But this fixation is a dangerous simplification, akin to an archer judging their entire craft solely by where a single arrow lands, ignoring the wind, the bow’s tension, and the archer’s own steadying breath.
The Performance Score is an aggregate, a weighted average of several metrics. It is a useful high-level signal, but it is not the truth. It is a proxy, and like all proxies, it can be gamed. I have seen builds where a developer, desperate to push a score from an 89 to a 90, will spend hours micro-optimizing a tiny piece of JavaScript, while a dozen unoptimized, lazy-loaded images sit just below the fold, waiting to tank a user’s experience the moment they scroll. The score was green, but the reality was fragile.
This is the core of the critique: the score measures a synthetic, lab-based condition. It is the archer firing at a perfectly still target in a windless room. It cannot account for the real world—the shaky 3G connection on a commuter train, the CPU being throttled by a background app, the sheer diversity of devices and networks our users actually employ. A perfect score can offer a false sense of security, blinding us to the nuanced, jagged reality of human browsing.
The True Quiver of Metrics
Our focus should shift from worshipping the single score to understanding the individual metrics it comprises. Each one tells a specific part of the story. Largest Contentful Paint (LCP) tells us about perceived load time. Cumulative Layout Shift (CLS) tells us about visual stability. First Input Delay (FID) or its newer counterpart, Interaction to Next Paint (INP), tells us about responsiveness. These are the real, user-centric signals.
A great user experience isn’t a high score; it’s the combination of a fast LCP, a zero CLS, and a minimal INP. It is entirely possible to have a ‘good’ score with a terrible CLS because the other metrics compensated. But the user who had to chase a jumping ‘Buy Now’ button doesn’t care about the score. They care about the frustration. The archer’s true skill isn’t in one perfect shot, but in consistently grouping arrows in the bullseye, shot after shot, regardless of conditions.
Let us put down the bow of singular obsession. Instead, let’s build a richer understanding. Let’s use the score as a starting point for investigation, not a finishing line for celebration. Our craft demands we listen to the full chorus of metrics, not just the aggregate tune. The real victory is not in a number from a robot, but in the smooth, stable, and responsive experience delivered to a human.
Notes & further reading
A few pages I came back to while writing this: