Break LCP into 4 sub-parts before optimizing

Break LCP into 4 sub-parts before optimizing
LCP (largest contentful paint) is a sum, not a single event. Profiling the wrong sub-part wastes effort.
— TTFB: 0 -> 240ms (server)
— Resource load delay: 240 -> 900ms (when the browser discovered the image)
— Resource load time: 900 -> 1300ms (actual download)
— Render delay: 1300 -> 1450ms (main thread busy)
The biggest slice here was load delay (660ms) — a discovery problem, fixed by preload, not by compressing the image.
Takeaway: measure the 4 sub-parts; the largest one is your only target.