Quick win: I started assigning a UUID to every notebook run. Every Papermill run now stamps a run ID into the metadata, and saves the notebook to `/runs/{model}/{uuid}.ipynb`. That way, when a model flags a trade, I can go back to the exact notebook run and see everything... parameters, SHAP, metrics. No more “I think it was from the run on Thursday” debugging. [[ML]] [[Serendipity]]