Every research project — a PhD, a postdoc study, a lab experiment — follows the same path through the research data lifecycle. This page is that path as one workflow, not a stack of separate rules. Start at Plan and move along; each step tells you what to do and links to the detailed how-to and the tool you need.
Three pillars travel through every step
- Data — your files, stored and backed up (Nextcloud, HPC)
- Documentation — what the data mean and how they were made (ELN + metadata + README)
- Analysis — the code that turns data into results (Git + JupyterHub/HPC)
Keep the three linked — the same IDs everywhere, outputs made by code, a README that maps them — and your results stay reproducible. That linkage, resting on good scientific practice, is the whole point.
The workflow
The steps
Ordered along the lifecycle. Start at the top; each step links to the detail and the tool.
1 · Plan — before you collect anything
Write your Data Management Plan first — 20 minutes that shapes the whole project. Then create the standard folder structure and agree a naming convention. → DMP Template
project/
├── README.md <- the map to everything
├── data/raw/ <- originals, READ-ONLY, never edited
├── data/primary/ <- raw converted to open formats (by code)
├── data/processed/ <- derived data (by code)
├── data/metadata/ <- metadata tables + data dictionary
├── src/ notebooks/ <- the code that makes everything downstream
├── results/ docs/ <- figures, tables, protocols, DMP
Naming: YYYYMMDD_project_sample_condition_type_v01.ext — no spaces, dates first, one consistent pattern.
2 · Collect — capture and document
Put raw data in data/raw/ and make it read-only (chmod -R a-w data/raw/). Record every run in your electronic lab notebook within 48 h, using the same ID you’ll use everywhere.
→ LabFolder
3 · Describe — metadata
Apply the metadata template at capture time; use your field’s controlled vocabulary, not free text. Never leave a blank to mean “not done”. → Metadata Templates
4 · Process & analyse — by code only
Convert raw → primary → processed with scripts (no manual edits); version code in Git; run on JupyterHub/HPC. The rule: if you deleted everything downstream of data/raw/, your code in src/ should regenerate it.
→ JupyterHub · Make your project FAIR
5 · Preserve — store and back up
Keep shared data in a Nextcloud group folder; back up raw 3-2-1 (3 copies, 2 media, 1 off-site); archive for the long term. → Nextcloud
6 · Share — Gate G1: leaving your private space
Before data goes to anyone else: run the FAIR self-check, clear the sharing checklist, and if the data is person-derived, follow the GDPR rules (pseudonymise; patient-identifying data stays at UKW). → FAIR Self-Check · Sharing & Publishing Checklist · GDPR & Sensitive Data
7 · Publish — Gate G2: the real publication
Deposit to a repository with a DOI and a licence — not a Nextcloud link. Use the two-track approach for sensitive data (open outputs open; individual-level data controlled-access). → Sharing & Publishing Checklist
8 · Reuse — the cycle restarts
Your published, FAIR dataset becomes the Plan-phase input for the next project — yours or a colleague’s. → RDM Lifecycle
The rules of thumb (carry these through every step)
- Raw data is sacred — never edit it; work on copies.
- Name things so a stranger understands — no spaces, dates as
YYYYMMDD. - One ID everywhere — filenames, metadata, lab notebook.
- Code, don’t click — every result is re-runnable from
src/. - Write it down as you go — ELN and README now, not later.
- Back up 3-2-1.
- FAIR when you share — metadata, licence, access level before anything leaves your hands.
More background and external guidance: External RDM Resources and The Turing Way.