The Stonemason's False Plumb: On the Over-Reliance on Lighthouse Metrics

We have, in our quest for performance, crafted for ourselves a new gospel. It is a clean, numeric doctrine, delivered by a single, all-knowing lighthouse. We run the audit, we receive the scores, and we know, with near-religious certainty, whether our site is “good” or “bad.” A green 95 is a blessing. A red 32 is a sin. We have taken the complex, human experience of loading a webpage and distilled it into a handful of numbers, and in doing so, I fear we have mistaken the blueprint for the building itself.

Don’t misunderstand me. Core Web Vitals and the tools that measure them are not the enemy. They are a necessary plumb line, a tool for alignment. But watch any master stonemason at work. They use the plumb line to check their initial setup, to ensure the wall is founded correctly. But the true craft lies in their hands, their eyes, the feel of the stone. They do not build the entire wall by staring only at the bob of the plumb; they build by understanding the grain of the material, the load of the structure, the play of light and shadow on the final form. To build solely by the plumb is to create something technically straight, yet utterly soulless and potentially fragile to real-world conditions.

The Numbers are a Controlled Environment

Our Lighthouse audits run in a simulated, clean, and consistent environment—a lab. It’s a sterile workshop with perfect lighting. The real web is a chaotic, windy cliffside. It is a user on a three-year-old phone riding a train through a tunnel, their cache already full of a dozen other single-page-app tabs. It is the third-party analytics script that finally decides to load, or the live-blog widget that churns unpredictably. The lab score tells you the potential of your structure; it does not tell you how it will stand in every storm.

Worse, this gospel of the score becomes a game. We learn to “trick” the plumb line. We preload, we defer, we lazy-load with such aggression that we create a new kind of jank—the jank of *nothingness*, where a user stares at a perfectly scored, semantically correct shell waiting for the content to finally decide to arrive. We optimize for First Contentful Paint by rendering a blank white page a millisecond faster, missing the point entirely. The metric becomes the goal, and the experience—the actual, felt experience of another person—becomes a secondary byproduct.

The true craft of performance lies not in worshiping the score, but in developing a feel for the material. It’s in watching a real user session on a slow connection and wincing at the hesitation they experience. It’s in understanding that a strategically eager-loaded hero image, while perhaps “penalizing” a lab metric, might be the very thing that makes a visitor feel welcomed and certain they are in the right place. It is knowing when to follow the rule and when the rule must serve a higher, human purpose.

Keep your Lighthouse. Consult it often. Let it guide your foundational work. But then, put it away. Test on real devices. Feel the friction. Listen to the pauses. The ultimate metric is not a number, but the absence of a thought—the moment a user’s intent flows seamlessly into action without a single, silent curse directed at the page. That is a feeling no audit can score, and no false plumb can ever guarantee.

Notes & further reading

A few pages I came back to while writing this: