In Chrome 77 Performance Observer introduces the "buffered" version, meaning we can instantiate the observer late in the page and still get the metrics. This works for Element Timings and LCP and LayoutShift (but not for Long tasks).
Layout Shift needs config in Chrome to work, so we will need to wait with that until it is released for real, but Element and LCP is something we can use.
We need to discuss what to use and how. I've tried them out (Element timings reported really close to the video metrics when we tried it on the Russian Wiki). Should we use LCP? When I tested it on the Obama page, the largest paint was a paragraph. Element timings is good because we can mark exactly which element to measure.
LCP is more complicated since we can measure different things on different pages (whatever LCP pickup as the largest paint). When I tested its hard to just get the render metric, you need to track the reported URL (if it is an image), the id or at least the element type. But if you only get the type, how do you know what we are measuring?