We recently caught a candidate who had perfect syntax for a high-concurrency problem. However, our engine flagged them because they skipped the 'logical state' of addressing a potential race condition in their initial logic flow—something no Senior Engineer would ignore, but a standard LLM often glosses over.
How are you guys currently verifying 'Senior' logic when every candidate has access to GPT-5?