Google Developer Expert in web performance
CTO @ RUMvision.com
Also sharing over at https://linkedin.com/in/erwinhofman
I saw this on a production site today. Who's gonna tell them? π¬

@erwinhofman.bsky.social
Google Developer Expert in web performance
CTO @ RUMvision.com
Also sharing over at https://linkedin.com/in/erwinhofman
I saw this on a production site today. Who's gonna tell them? π¬
So, submit your own proposal or give a π to already listed/proposed features you'd like to see supported across browsers.
Starting with LoAF when unsure where to start upvoting π
github.com/web-platform...
RUM tools, developers and ultimately site visitors will all benefit!
3/3
I just proposed ππ¨π§π ππ§π’π¦πππ’π¨π§ π π«ππ¦ππ¬ aka #LoAF
An incredibly useful API for spotting performance & #INP regressions & improvements.. across real users!
Our recent Meta Pixel finding might be one of the best examples of that so far:
bsky.app/profile/erwi...
2/3
If you're a web-dev and there's a web feature you like ( #CSS, #JS or an #API) but isn't cross-browser yet: this is your time to shine β¨
Every year, the Web Platform Interop project (collaboration between browsers) to decide on which features to support cross-browser:
web.dev/blog/interop...
1/3
Just updated it :)
Thanks, Ian! I just updated the article with lab data. It was way more work than expected, as their JS time is spread across many tasks and has grown over time. So yes, let's hope they can start slicing away some of that work again. That would be a win for many sites!
Thanks Ethan! I ran out of time for the day after this, but gonna do a DevTools trace too and might end up including it in the article π€ (or separate one if it becomes too long π )
Do let us know if you're seeing anything!
I documented the RUM data, historic Pixel versions, rollout timeline and code changes here:
www.rumvision.com/blog/increas...
Correlation β causation, but that August rollout is difficult to ignore.
Meta also has separate AAM experiment gates for platforms like WooCommerce, BigCommerce, Magento and Klaviyo. So not every site necessarily follows the same matching path. My hypothesis: more work is happening around an existing click pipeline, rather than Meta suddenly adding click tracking.
August is where the strongest correlation appears. A large set of AAM experiments rapidly went from roughly: 1% β 10% β 100% This included Shadow DOM, forms, selectors, iframes and FormData extraction. And there was another interesting detail...
June brought changes to Metaβs parameter-extraction pipeline. By July, the Pixel had gained more automatic processing, including expanded Automatic Advanced Matching (AAM). Interesting timing, but nothing here was strong enough to call the cause of the #document.onclick regression.
Same trend across multiple sites. First finding: Metaβs document-level click handling isnβt new. SubscribedButtonClick, Inferred Events and geometry reads were already there before the regressions. So the question changed from βwhat was added?β to: what changed behind the existing click handler?
The regression wasnβt gradual. Our RUMvision data shows three distinct jumps in Meta JS time: β’ June: 34 β 39ms β’ July: 39 β 49ms β’ August: 49 β 59ms That looked a lot more like rollout steps than normal variance.
Then it got more interesting.
I shared the finding and @krzksz.bsky.social from #Shopify performance team confirmed theyβre seeing the same pattern across their platform.
So this wasnβt looking like a site-specific implementation issue anymore.
I checked a few to feed my own curiosity: mostly checked + unchecked checkbox images as well as other icons. But even their logo is in their, base64 encoded π¬
π₯²
Nice to meet you IRL! And I was already wondering, but this confirms it: one could definitely see my fried brains at the end of the day π
For their "Send prompt" and few other buttons, also if you'renot logged in π€
Also to be found if you're not logged in:
chatgpt.com