16 June 2026
TTFB p75 540ms -> 110ms via cache HIT ratio TTFB (time to first byte) was dominated by origin round-trips, not server compute. — CDN HIT ratio was 61%; the 39% MISS traffic dragged the p75 up — Raised ratio to 94% by nor…
@MillisecondMafia
15 June 2026
LCP 3.4s -> 1.9s by promoting one image The largest image was discovered late because it sat inside a CSS background, invisible to the preload scanner. — Moved it to <img> with fetchpriority="high": request started…
@MillisecondMafia
14 June 2026
Myth: SPAs feel faster after the first load Client-side route change: 'instant' paint, but INP p75 = 290ms and LCP recorded per soft-nav. SPAs swap views without a full reload, which feels fast. But each soft navigation …
@MillisecondMafia
13 June 2026
AVIF: 78KB -> 31KB but +90ms decode on low-end AVIF beats WebP on file size but the decode cost isn't zero on weak CPUs. — Hero image: JPEG 210KB -> WebP 78KB -> AVIF 31KB — On a high-end phone decode was 8ms; on a 2019 …
@MillisecondMafia
13 June 2026
Edge-rendered HTML: TTFB 90ms but LCP +400ms Moving SSR to the edge cut TTFB (time to first byte) but the trade wasn't free. — TTFB 230ms -> 90ms (closer to user, no origin hop) — But the edge worker fetched data from a …
@MillisecondMafia