Post

Interesting! Etsy implemented speculation rules and got these results: 👍 Time to First Byte on the listing page improved 23.6% 👍 First Contentful Paint improved 20.7% 👍 Largest Contentful Paint improved 21.1% 👍 DOMContentLoaded improved 20.4% #PerfNow #webperf www.etsy.com/codeascraft/...

I'm not sure how I feel about this part of Etsy's results: "About 1 in 14 prefetch requests was navigated to by the user." The performance improvements are superb, but a 7% hit rate doesn't sound great for server usage and bandwidth. What do you think, #PerfNow? What's the right balance?

I thought the same and questioned @stoyan.me on this! It also sounds like a LOT so wonder why do high if accurate. It means users are on average hovering over 14 questions and clicking on one. Or 14 users are hovering over links and only one is clicking. Both sounds suspiciously low to me.

On other hand, if users are hovering over items and then not clicking, then they may be clicking later (e.g. Search for "shoes", deciding between two—or more—search results, click one, then go back, and click the other). That may still show as an "unused prefetch" even if it was eventually used.

Either way I'd love to know more about this stat if @stoyan.me , @cdaveross.bsky.social or @paulcalvano.bsky.social have any more details?

Thanks Barry, I was curious to hear your thoughts on this. Did you have an 'ideal' activation rate in mind while setting the behaviour for the likes of moderate vs eager? (Side note: very cool to see the mobile viewport heuristics in your talk.)