Glad that clicked yeah, I’ve kept it internal since it’s usually tailored per setup. But I can definitely help you structure yours properly. I’d set you up with a simple logging layer that captures raw audio, transcripts, prompts, and model settings, plus a way to replay exact scenarios so you can debug consistently. If you want, we can take this to DM and I’ll break it down for your specific stack or even help you implement it end-to-end 👍