IMAGE DELIVERY COMPARISON / NIALL ARNFIELD
23 September 2026

18 recorded loads: three interleaved fixed/responsive pairs at each of390px DPR1,390px DPR2,1440px DPR1. Chrome153 Windows, DevTools Slow4G and4x CPU slowdown. Loopback read-only server, no-cache responses; unique page query per run; navigation requested ignoreCache. A pilot load is excluded. Same HTML/CSS/composition, only srcset and sizes differ. Rendered image width342mobile/1120desktop. No external fonts, production theme, analytics or third-party scripts.

CSV reports selected resource payload, browser transfer size (includes overhead), resource duration and last observed hero LCP after loading/decode/two animation frames. Later interactions were not tested. Mobile responsive runs also exposed a zero-byte fallback-image resource entry; that entry is not counted as a completed extra download. The full selected payload equals the saved source file in every run.

Source variants are existing public Soinvogue case-study graphics; see asset-provenance.json for URLs, dimensions, bytes and hashes. Images are not included in this download. example.html.txt is readable HTML source, not a production plug-in. It requires the named files alongside it to run locally; adapt the layout and filenames to your project.

KB in the article means1,000bytes. The file reduction is84.4% atmobile1x and65.7% atmobile2x, zero atdesktop. Resource and LCP medians come from three observations pervariant/profile, not a robust population estimate. No significance, live speed, revenue, ranking, Lighthouse or real-user Core Web Vitals claim. The live case study already had responsive delivery; fixed-source is an illustrative comparison, not its old implementation.

No website writes, tracking, credentials or dependencies in this pack.
