Working on reproducible research data

Hi, I'm a research data engineer working with academic teams on data collection, cleaning, versioning, and reproducible analysis.

A lot of research data work breaks down before the model, chart, or paper ever happens. Sources change, scripts are undocumented, licenses are unclear, and six months later nobody can explain exactly how the dataset was created.

I'm especially interested in open science, dataset provenance, archival workflows, and practical ways to make research easier to rerun without slowing everyone down too much.

Curious how others here handle this: when you publish or share a data project, what do you document first... raw data, collection timestamps, cleaning scripts, licenses, environment files, or something else?