I'd guess this is the page-load time with the safe-browsing overhead.  Since safe-browsing self-evidently cannot speed up actual page loading, is there any possibility that instead it's preventing the actual page loading?  Like maybe a bunch of slow legit page loads are being shifted to a fast interstitial?

Of course, it's also possible that the metrics are broken, like for some reason some requests _only_ show the first-step value, or something.  Perhaps if something like a redirect were busted.

I don't think Linux even has a canary, so it wouldn't show up there until dev.  I'd also look closely at Mac before assuming it's platform-specific - Mac has many fewer users than Windows, and they may have some different characteristics (including some like DNS and TCP stacks), so it's possible the signal is there, just being overwhelmed by noise.

Ah yes, V4Only is only enabled on Canary, and linux has no Canary.

Looking at the UMA again today showed the graphs for PVer3 and PVer4 to be quite comparable: https://uma.googleplex.com/variations?sid=efbe13813c6b2313ce33cbe0eb4e931b

Will keep looking at this over the next 1-2 days since UMA is experiencing some issues right now.

Marking this as fixed now since the said difference has disappeared.
See: https://docs.google.com/document/d/1EwdQoBtCK2CJquq5ZczG7O102eUC4mClxf7JDyZwKvU/edit#heading=h.vym60wlwqhvk