LCP is way more than the total time LH took

[thany]

I let LH process a website (local only, can't share it) and it was done doing its thing in about 10 seconds. Fairly normal.

And then it manages to report a LCP of 82 seconds.

So the question is hopefully quite clear. But to be extra verbose: how does it arrive at a LCP that is much more than the total time taken to analyse a website? Where is this LCP magiced up? How could it possibly (incorrectly, I believe) know that it's 82 seconds, after being done in 10?

It wasn't a fluke either, it reproduces this weird result every time.
I'm on version 12, and I've done a fresh npm update -g as well, just minutes ago.

[adamraine]

Lighthouse will use simulated throttling by default which loads the page normally and then estimates what performance would look like under certain network and CPU conditions after. This means that the performance metrics are not expected to match what you observe locally.

Regardless, 82 seconds does seem pretty high. Can you try using DevTools throttling (i.e. --throttling-method=devtools) to see if an LCP that high is repeatable with different throttling strategies?

[thany]

I don't even have to. I can see in the "comic" that after the second screenshot everything is loaded and stays the same, so the LCP has already happened by then.

I can see if the throttling method makes for a more realistic measurement, but realistically I'm improving the performance, so it doesn't matter what the number is or means. I guess if I can get 82 seconds down to, say, 30 seconds, it'll be a website as fast as lightning 😀