[Layout Shift Metrics]

[Josh Deltener]

Of the 3 techniques described, the one that seems to make the most sense is Session Windows.  CLS is painful, but once it's started it's just something you have to endure for 1-3s until the page is settled.  It's much worse when you have significant CLS over and over on the same page, so it seems those experiences should be penalized even more.

I'm not sure how the techniques will be applied to SPAs, but I would assume at some point SPAs would have all metrics reset once the history is updated, similar to how classic websites are treated.  Then, hopefully we can also see the correct URLs in Google Console that relate to the CLS that actually happened within that page instead of the whole website.

[Michal Mocny]

On Tue, Jan 26, 2021 at 9:55 AM Josh Deltener <heckt...@gmail.com> wrote:

Of the 3 techniques described, the one that seems to make the most sense is Session Windows.  CLS is painful, but once it's started it's just something you have to endure for 1-3s until the page is settled.  It's much worse when you have significant CLS over and over on the same page, so it seems those experiences should be penalized even more.

Thanks for the feedback!

Note that the 3 broad strategies are:

• Avg of Long Session window (5s gap between windows)
• Max of Medium Session Window (2 variants: w/ and w/o overall duration limit)
• Max of Sliding Window (2 variants: 1s and 300ms)

Based on your summary, I think you meant the first strategy of "avg of long session window" and not the second?  We'd love to see examples where one strategy is clearly better than the other (if possible)!

I'm not sure how the techniques will be applied to SPAs, but I would assume at some point SPAs would have all metrics reset once the history is updated, similar to how classic websites are treated.  Then, hopefully we can also see the correct URLs in Google Console that relate to the CLS that actually happened within that page instead of the whole website.

Indeed.  We expect all of the strategies listed above to do much better with arbitrary time durations than CLS, and that should indirectly improve measurement for SPA sites (and other long-lived pages).  However, for metric slicing and URL route attribution, we would need something like you suggest.

--
You received this message because you are subscribed to the Google Groups "web-vitals-feedback" group.
To unsubscribe from this group and stop receiving emails from it, send an email to web-vitals-feed...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/web-vitals-feedback/11a27460-7322-4581-bbe5-7fe0c670fd40n%40googlegroups.com.

[Josh Deltener]

What would this look like as a deliverable?  I ran a couple of the queries against some of my pages, before/after navigating to other pages (SPA).  

I saw a bunch of shifts (most very small), but I'm not sure how to quantify what "better" looks like from the output :) 

[Michal Mocny]

What a great question!

I think ideally you would compare the CLS value of the page to the current web vital CLS threshold (0.1), then also compare the new potential normalized LS scores to their own respective thresholds.  That we would give you a label marking the experience as "Good", "Needs Improvement" or "Bad" based on each score.  Then, you could compare those labels to see which strategy most often matches expectations.  Based on our own results, I would expect that the labels agree 95%+ of the time, so it's really only in the extreme cases where it gets interesting.

Unfortunately, we do not yet have guidance on expected thresholds for the experimental metrics.  This makes it tougher to compare strategies, since it's not reasonable to just compare the scores directly for a single experience (i.e. we of course expect max-sliding-1000 to be higher than max-sliding-300 in the aggregate...)

We've been focusing on analyzing a larger corpus of data, but we'll brainstorm how best to evaluate individual experiences with what we have now, and I'll get back to you.  Thanks for raising the question :)

To view this discussion on the web visit https://groups.google.com/d/msgid/web-vitals-feedback/d0e38382-0935-4102-b7ef-59e1d1c66cdfn%40googlegroups.com.