@lannka Great! Is it worth find a way to add a sampled CSI ping as well?

@dvoytenko that's what I'm hesitating. So far I don't know what to do with the data. how to normalize the data is also a challenge.

I'm think in-line with your "probability of bad frames"

frames = animation frames of course.

@dvoytenko sure it can be another "bad frame" metric, we can send a the number of long tasks at the time we send the Good Frame Probability. We expect this metric should be strongly correlated to the existing metric. In that sense, it's not providing new stuff, but maybe a good back up to the existing metric due to its neutrality, since it has no dependency on runtime code

On the other hands, I'm more interested about the "attribution" info from LongTask API. We can find out which 3P iframes are exhausting resources. However, I'm still thinking what we can do with that.

Cool. Let's merge this PR and re-open a task for LT API when ready.

To make this useful for all AMP developers, maybe a better place to load this script is when #development=1, just like how /v0/validator.js is loaded. Or, we can even compile this little piece of code into /v0/validator.js?

This way, developers can see what 3p frame is slowing their AMP pages down. This is our most conservative way to use such info.

@dvoytenko @Gregable thoughts?

Talked to @Gregable validator.js is also used in non-Browser env. so the code here won't run since it requires browser API.

will introduce it as a separate binary, like https://cdn.ampproject.org/v0/examiner.js

@dvoytenko I moved the code to a separate binary. Once deployed, I will get it loaded while #development=1