LCP breakdown insight subparts can be confusing

[connorjclark]

The durations for each subpart in this insight are what was observed, not what was simulated. So the LCP reported by Lighthouse will be larger than the subparts.

It's not feasible for us to accurately simulate the subpart durations. So I think we should change the column label to "Duration (observed)" and explain a bit in the description.

[eric-michel]

I wonder if there are any other mechanisms that can be used to provide feedback on the simulated LCP score. We're running into issues with long LCP scores on sites for which we've completely removed all render blocking resources. All other scores are outstanding. Here's an example: https://pagespeed.web.dev/analysis/https-drreynolds-com/urzyth1xj3?form_factor=mobile.

Without feedback, we're unable to make any iterative improvement on our sites. I'm not sure if the tool itself is flawed (I can't seem to replicate these LCP times when using throttling in devtools) or if there's something significant I'm missing.