Independent web performance consultant. Co-chair of the performance.now() conference. Taller than my photo looks.
Let's work together: https://timkadlec.com/me/
Friendly reminder that when a third-party provider requires you to load their script in a render-blocking manner, their problems become your problems.
Got an email about a massive performance regression from a past client yesterday. Turns out a third-party provider was (still is) having server issues causing an absolutely massive regression in LCP. Thankfully an easy and quick fix: remove the script.
Always build with resilience in mind.
Started working with these folks in mid-May and not gonna lie, pretty pleased with the progress they've made on INP in such a short span of time.
From a p75 of 4.25s in April (90% of sessions having a poor experience) to a p75 of 225ms in June (9% of sessions have a poor experience).
Phew. What a difference a `:last` makes in jQuery. Helping a client with very high Interaction to Next Paint metrics.
They use a third-party Shopify app with a long list of selectors in a click handler.
With the deprecated and much slower`:last` selector included in that list: ~616ms
Without: ~16ms
(Talked to the app devs and they're looking at it now so hopefully this gets fixed)
INP officially moving out of "experimental" mode with the intent to replace FID in March 2024 is very welcome.
Gonna be a harsh reality check though with regards to how we've been building for the web.
Chrome is starting to ignore low-entropy images (measured in bits per pixel) for Largest Contentful Paint.
If you're curious to better understand how it's calculated, I threw together a quick little page where you can upload an image and see the bpp as you adjust the size.
Small, but helpful, improvement in WebPageTest.
We now provide the request priority for Firefox tests. Interesting to compare to Chrome, and will become even more interesting if/when Firefox implements priority hints (https://bugzilla.mozilla.org/show_bug.cgi?id=1797715)
I get that there's potential here, but....I look at these limitations that ChatGPT lists and the number of inaccuracies I've seen playing with it (some small, some less so) and I am *floored *that companies are so eager to crank out features and products built around it.
It's not the things it gets wrong that people recognize as incorrect that worry me; it's all the times it gets things wrong and people don't realize it.
More "0kb client-side JavaScript" baselines please.
