I’m seeing a lot of similar things during code reviews of substantially LLM-produced codebases now. Half-baked bad idea that probably leaked from training sets.