We've chatted a bit about this - I think the right thing to do is probably to figure out how often this happens, and dig into some pages where it occurs frequently.

Intuitively, when a user does this, it probably means you aren't painting fast enough, so we'll have identified that the site is bad via paint metrics. Painting slowly could result in a FID that's not super representative of user experience, but a site that's blocked on resource loads for a long time is more likely to have a low FID in a kinda unrepresentative way in general.

I'm leaning towards just keeping the definition simple, but we should gather the data.

> I think the right thing to do is probably to figure out how often this happens

sg. If less than 0.05% of the time, we can ignore it for now, but IMO if its 1% of the time, then it doesn't seem fair to ask anyone to consider the 99pctl of FID.

> Intuitively, when a user does this, it probably means you aren't painting fast enough

Aye. I think at this point the signal isn't what FID is designed to represent; it's communicating a different "impatience" signal.

I guess we might actually care both about how often these occur, and what the FID is in cases where they do occur.

My intuition was that we'll tend to underestimate true FID in these cases - we'll just be blocked on network, and the main thread will be responsive.

Sounds like you're thinking of cases where the main thread is so busy we haven't had time to paint yet?

Nah i have same hunch as you: main thread will be free before FP. 

Gotcha, yeah. The 99th pctl shouldn't be affected too much by this. 
If there were happening 10% of the time, our 99th pctl would be 10% untrustworthy. 

I suspect the real number is a bit lower than 10%, so yeah it seems okay to dig into this later on. :)

Any update tdresser@?

The real number is just over 1%, but a bunch of the cases where this happens are clearly broken (404's, super slow pages, cases where FP doesn't fire because the page is just flash).

I'm still digging in here.

Ping! This bug has been flagged as stale by Chrome Speed Metrics bug triage, and it has an owner. Any update on this issue?

I think we're fine to include these cases.