In the talk, Lars mentions that they often rely on self-reported anonymous data. But in this case, Google is large enough that teams have developed similar systems and/or literally re-written things, and so this claim comes from analyzing projects before and after these re-writes, so you’re comparing like teams and like projects. Timestamped: https://youtu.be/6mZRWFQRvmw?t=27012

Some additional context on these two specific claims:

Google found that porting Go to Rust "it takes about the same sized team about the same time to build it, so that's no loss of productivity" and "we do see some benefits from it, we see reduced memory usage [...] and we also see a decreased defect rate over time"

On re-writing C++ into Rust: "in every case, we've seen a decrease by more than 2x in the amount of effort required to both build the services written in Rust, as well as maintain and update those services. [...] C++ is very expensive for us to maintain."