I'm not sure what it has to do with translating the code to Rust.
The first is that the translation function (from C to Rust) has itself been formally verified. In other words: if you give it a C program that obeys certain necessary pre-conditions, it will give you a safe Rust program. I believe the title of the paper is primarily using "Formalized" with regard to this characteristic.
The second is that the set of programs the authors evaluate their tool on are C programs that have themselves been formally verified. I only barely skimmed the introduction and didn't see it addressed directly there, but I would assume that this is because the pre-conditions necessary for their translation to work are most easily (only?) met by formally verified C programs, where of course the verification has been rendered with respect to properties that would be relevant here (e.g., memory use).