Ask HN: How do you find out if the LLM API is giving degraded responses (opens in new tab)

If you are building on top of multiple LLM APIs or even a single one amongst OpenAI, Claude, Gemini, etc. what do you do when the API starts degrading (slow TTFT, elevated error rates, timeouts). Or even worse, when there are responses but the model is drifting. How do you find this out? I'm trying to understand if this is a widespread pain or just something I've been unlucky with.