Milestone timing compatibility with CrUX data

[mattzeunert]

Lighthouse sometimes reports different data from what would be collected in CrUX:

• CrUX would use largestContentfulPaintAllFrames for LCP, but the Lighthouse report currently doesn't surface that value
• CrUX counts redirect time as part of LCP, but with Lighthouse I get the same LCP for youtube.com and m.youtube.com

(I'm a bit fuzzy on how it works exactly, so let me know if I'm getting something wrong here.)

I was wondering if it would make sense to bring these more in line with each other.

[connorjclark]

Hi Matt! 👋

re: all frames

We have a lot of lantern complications here, and it would be a lot of work. We should at least write down what all would need to change to Lantern to make this feasible, even if the priority of it is unclear ATM.

re: redirects

ref #8984

We determined awhile ago that we want to consider redirects for LCP in both simulated and observed. We haven't gotten around to it yet. @mattzeunert you must be using observed/devtools throttling, right?

[mattzeunert]

Yeah we use observed throttling for DebugBear.

When CrUX added the TTFB metric to the API we decided to surface an equivalent "Full TTFB" value including redirects, directly based on the request list. But then we realized that FCP was smaller than TTFB if there was a redirect.

We also have a customer who uses AEM Experience Fragments on their pages, so the LCP is often in an iframe. And no lab tool really surfaces that (DevTools also only shows main frame I think), so it's confusing for them to see a bad CrUX LCP.