My guess is they did testing but the build they tested was not the build released to customers. That could have been because of poor deployment and testing practices, or it could have been malicious.
Such software would be a juicy target for bad actors.
If an intern can release to prod without extensive testing there are bigger issues.
Given the scope of the potential impact, if anyone can release to prod and have that deploy to all customers without some form of a canary release strategy, then there are still issues.
The relative number here might be more useful as long as it's understood that Google already has significant emissions. It's also sufficient to convey that they're headed in the wrong direction relative to their goal of net zero. A number like 14.3 million tCO₂e isn't as clear IMO.
"Up to 20%" is meaningless for a headline and is pure click bait. It could be any number between 0% and 20%. Or put another way, any number from no time at all to a horrifying more than an entire day per week.
Why not just state the average from what is probably a statistically irrelevant study and move on?
I think the key point is ownership. If the house is owned by an equity firm, even if it's occupied it still counts as a house which could instead be owned by, well, homeowners.
It's incredibly hard to help people who actively vote against their own interests.