Experiments and Runs

Hierarchy: Experiment → Run

An Experiment is a folder. A Run is a single execution inside that folder. Every run gets:

• a unique run_id (UUID)
• a start + end timestamp
• the user that launched it
• the source: filename, git commit, entry point
• params, metrics, tags, and artefacts you log

Naming hygiene

Choose experiment names by business goal, not by model:

GOOD: 'churn-q3-retraining'
GOOD: 'fraud-baseline-2026'
BAD : 'xgboost-tests-final-v3'
BAD : 'untitled-experiment-7'

Then use tags for the model family. That way you can search later by goal or by tech and the UI stays organised.

An Experiment is a filing cabinet drawer, a Run is a single folder inside it. You name the drawer by the case (mortgage-claims-2026) and put many folders inside, one per iteration, each holding the photos, receipts and notes for that attempt.