I've been switching a lot of generated files to being checked in (with CI verifying they haven't drifted from the source). The primary motivation has been performance. For example, in Rust code, it means I don't need to foist the code-gen process and all the dependencies needed for it on dependent crates. I've seen this play out similar in other build systems and circumstances. The key is the data needs to be independent of other factors (like the system doing the generation) and the rate of change of the code-gen source and generator has to be relatively low.