You know how LLMs always imitate expertise and understanding? That isn't specific to English, it happens in code as well. It's harder for them to get away with it because even the best guess isn't likely to be approximately correct, but they still try and it still sometimes works.

One of cucumber scenario tests now contains a deadlock somehow. We run them one at a time. I need some rakia.