Developer uncovers how minor JS differences caused 30-point PageSpeed score splits

A developer running PageSpeed Insights (PSI) on their static site lans.cloud discovered a puzzling split: five pages scored 97–100 while five nearly identical pages scored 70–79, despite all pages performing well in local Lighthouse tests. The root cause was not an LCP issue but a first-paint delay triggered when async JavaScript chunks arrived and began evaluating on the main thread before the browser could complete its initial paint. On fast local hardware the race was never lost, but on Google's deliberately slower test runners even marginally more JavaScript per page was enough to delay first paint by around 2.3 seconds. PSI's Lantern throttling model then interpreted this paint-blocking race condition as a structural dependency, projecting the delay out to a 5.5-second LCP and penalising the score heavily. The case highlights how simulated throttling can amplify small, hardware-sensitive timing races into large reported performance gaps that are invisible during local testing.