Surely git’s source is already in LLM’s training corpus. So this is far from clean room approach.