CC mlamouri. We should collaborate on this to make sure we are measuring this consistently with our other video UKM metrics.

To date, for pageload metrics, we've used the WebContentsObserver::MediaStartedPlaying callback to determine if media was played. Would it be sufficient to use this for UKM as well?

We currently have WatchTime UKM which will indirectly answer the original question. Though, we wouldn't know if it was autoplay'd or not.

We have added "Autoplay Initiated" to the WatchTime UKM. Should this be closed?

Yes, the Autoplay Initiated boolean combined with a Duration > 0 filter would provide the signal originally described in the bug.

Unfortunately, bengr meant to connect this UKM with the PLM UKM. Because watch time UKM uses GetNewSourceID() instead of the ID on the NavigationHandle, this wouldn't be possible to join. We can either use the WCO method for media played, or we can plumb the UKM ID into blink WebLocalFrame. I'd prefer to just use the PLM harness, unless there is a reason to think we want all of the extra video information tied to the PLM UKM.

I think this bug can be well scoped to the time when the first media element played on the page.

bmcquade: is it sufficient to take the TimeTicks::Now() in the page load metrics harness, or should that measurement occur in the renderer to be comparable  to the other timing events?

We could do a bool instead if we think timing is overkill.

This isn't needed. We've decided to measure data savings more holistically with the data saver header.