Notebook hygiene rituals our Gangwon cohort actually keeps

We stopped treating notebooks as scratchpads once teams began shipping models every sprint. The first paragraph here is about naming cells, pinning library versions, and exporting environment manifests before Friday stand-down.

Second paragraph covers the review board: two peers sign off on data sources, not on model accuracy promises. We log which snapshot IDs entered training and attach them to pull requests.

Third paragraph explains how Samcheok-si hybrid sessions use printed runbooks so field engineers can annotate paper copies, then photograph annotations back into the shared drive—low tech, high signal.

Fourth paragraph outlines what we intentionally omit: no automated “magic cleanup” scripts that hide messy cells; visibility beats polish for audit readiness.