This is an expected outcome of SlimmingPaintV175 :) Raster invalidation is moved from from PrePaint stage into Paint stage.

See go/spv175status for more details of performance data.

But SPv175 is not in beta yet - this affected the beta channel.

Sorry for not reading the report carefully.

In split version graphs
95%: https://uma.googleplex.com/p/chrome/timeline_v2/?sid=4279c0792850085d9d2efbb33df9e774
50%: https://uma.googleplex.com/p/chrome/timeline_v2/?sid=d5bf2be6e7371c3a565ba4e3f9579fc1,
all versions has the same trend in their own life spans and the curves of different versions are approximately continuous, so I guess the improvement is just a result of overall web access trend change.

Looks like dev and beta both went down at the same time, so web trend
seems the cause.

https://uma.googleplex.com/p/chrome/timeline_v2/?sid=5a871029d23b2ecb60b13921f08a0f99

Let's look at UKM to determine which site.

Hmm. The drop in dev channel about Feb 1, then going back up, then down
again a couple of weeks later, is due to SPv175 landing then revert then
re-land.

However, metrics for this as well as LayersUpdateTime happens as soon as
M66 beta is cut. For some reason, M66 is matching SPv175 performance
characteristics. Could we have accidentally launched this in M66??

https://uma.googleplex.com/p/chrome/timeline_v2/?sid=44292838e249812ebf70103867980bd8

I'm looking at the release branch code now.

"metrics for this ... change"

Update: I built the release branch and printed out the feature flags.
SPv175 is not in fact on.

These metrics are quite odd.

I think I made an error, and was comparing 50th percentile to 95th. All seems
ok.