A weekly report had been flagging the homepage as failing a core speed metric for months. Measuring it properly showed the page was fine – the alarm was.

• The weekly site-health report graded the homepage as failing Google's Largest Contentful Paint standard, week after week. Acting on that flag meant a plan to rebuild the hero image and strip out stylesheets – real work, aimed at a real-sounding number.
• Measuring the live page directly told a different story. In a real browser the page finished painting in roughly four tenths of a second. The failing figure came from a simulation that models a slow phone on a slow network, and it swung between seven and sixteen seconds across runs – the kind of variance that is itself a clue you are looking at a model, not a measurement.
• The image the plan would have optimised turned out to be the wrong target entirely: it was already compressed, already prioritised, and arrived in about two milliseconds. Roughly 86% of the simulated delay came from elsewhere. Fixing the image would have changed nothing.
• The reporting now separates the two kinds of data it was quietly conflating. Real-user measurements decide the pass/fail grade; lab simulations are labelled as such and can no longer escalate into a recommended action. Where no real-user data exists yet – which is the case for a site still building traffic – the report says so plainly instead of showing a red cross.