How do you keep project context next to the model experiments you run?

A Hugging Face forum discussion on keeping project context attached to model experiments recommends making each run carry its own record rather than keeping notes beside it: when a run starts, the dataset split, prompt file, model revision and eval command are saved automatically into that run's folder, with three lines added at the moment of each choice stating what was picked, what was rejected and what result would change the author's mind, and scores appended to the same folder when the eval finishes. The proposed test is reopening one run two weeks later with notebooks and chat logs closed — if the folder alone explains what was tried, why, and what the numbers meant, the record is complete, and any unanswerable question names the next field to add.

When I’m iterating on a small model or RAG setup, the hard part isn’t the training loop — it’s remembering why I chose a particular dataset split, which prompt variants I already tried, and what the eval notes meant two weeks later. Right now I bounce between notebooks, a scratch doc, and chat logs, and the “why” of a run gets lost. How do you keep experiment notes, data decisions, and eval commentary attached to the actual work so a future-you or a teammate can reopen it without archaeology? Practical workflows welcome — folders, Spaces habits, local tools, whatever has stuck for you. nootxlm https://discuss.huggingface.co/u/nootxlm 2 Make each run carry its own record instead of keeping notes beside it: when a run starts, save the dataset split, the prompt file, the model revision, and the eval command into that run’s folder automatically, then add only the part a machine cannot capture — three lines written at the moment of the choice saying what you picked, what you rejected, and what result would change your mind — with the scores appended to that same folder when the eval finishes. Test it by reopening one run two weeks later with the notebooks and chat logs closed: if you can say exactly what was tried, why, and what the numbers meant from that folder alone, the record is complete — and whichever question you cannot answer names the one field to add next time.