dmp

name	dmp
description	Produce the Data Management Plan a funder asks for (Horizon Europe, TÜBİTAK, an institutional policy) — derived from this project's dataset and run registers rather than filled into a downloaded template. Reports FAIR letter by letter with the record that evidences each, and writes no sentence for a dataset whose licence or citation DOI is not recorded. Use when a call requires a DMP, at a project's data-management review, or before depositing. Triggers on: 'veri yönetim planı', 'data management plan', 'DMP', 'FAIR plan', 'open data requirement'.
disable-model-invocation	true
argument-hint	[--write exports/dmp.md] [--check] [--sections <funder-sections.json>]

/dmp — the plan, generated from what the project already recorded

Most data management plans are written the same way: download the template, fill each box with a sentence that sounds like compliance, submit, never read it again. The reason is that the facts a DMP asks for — what data, which licence, which format, where it lives, who may have it — are scattered across a filesystem and a memory.

This project keeps them in two script-owned registers already. So the plan is derived, and what cannot be derived is left as an explicit blank rather than written over.

Gate

None. But the plan is only worth the registers behind it:

python scripts/dataset_db.py verify        # is the record complete, and do the bytes still match
python scripts/dmp_gen.py --check          # which FAIR letters are unevidenced

With an empty analysis/datasets.jsonl the result is UNCHECKED, not OK: a plan written before the data is registered describes data nobody can verify.

If analysis/datasets.jsonl is empty, the plan will say so rather than describe data that does not exist. Register the data first (/find-dataset, /collect-data).

Step 1: the gaps, before the document

python scripts/dmp_gen.py --check

{"result": "GAPS",
 "unevidenced_by_dataset": {"local-vib": ["F", "R"]},
 "no_sentence_written_for": [{"dataset": "local-vib",
                              "missing": ["citation DOI", "verified licence"]}]}

Fix these in the register, not in the prose:

python scripts/dataset_db.py set-field DS-0002 citation '{"doi": "10.5281/zenodo.123"}'
python scripts/dataset_db.py set-field DS-0002 license "CC-BY-4.0"
python scripts/dataset_db.py verify --id DS-0002

A licence typed into the DMP but not into the register is a licence that will disagree with the paper's availability statement six weeks later, because /submission-pack reads the register and this document would not.

Step 2: build

python scripts/dmp_gen.py --write exports/dmp.md

Section structure is the Science Europe core (a public, funder-neutral standard): data summary · FAIR ×4 · resources · security and storage · ethical aspects · other policies.

If your funder publishes its own section list, put it in a file and pass it:

python scripts/dmp_gen.py --sections proposal/dmp-sections.json --write exports/dmp.md

{"funder": "…", "source_url": "…", "read_on": "2026-08-15",
 "sections": ["data_summary", "fair_findable", "…"]}

Sections the registers cannot answer come back as "answer it yourself" rather than as a generated paragraph.

Step 3: fill the blanks that are genuinely yours

Everything marked TODO: in the rendered file is a commitment, and only you can make it:

• storage cost over the project's lifetime, and after it ends
• the person responsible for data management
• where working copies live, how they are backed up, who has write access
• the repository the final deposit goes to
• your institution's data policy and the funder's embargo terms
• the ethics approval reference — once your board has issued it

How FAIR is reported, and why not as a score

Each letter is either evidenced by a record or not, with the reason:

- cwru-bearing — evidenced: citation DOI 10.1234/abcd
- local-vib    — NOT evidenced: no citation DOI in the register

Three out of four is not "75% FAIR" — it is a dataset nobody can reuse. The document never contains the sentence "this project follows FAIR principles", because that sentence is available to anyone regardless of what they did.

What this does not do

• It does not know your funder's template. The generic core ships; your funder's section list is a file you fill.
• It does not invent a cost, a volume or a repository name. A plausible commitment nobody made is worse than a blank.
• It does not describe data the register does not hold. A dataset with no verified licence or no DOI gets no sentence — the gap is listed under its own heading, so a reviewer sees an honest absence rather than a confident lie.

Reference

• Script: python scripts/dmp_gen.py (--check, --write, --sections)
• Reads: analysis/datasets.jsonl, analysis/runs.jsonl, analysis/collection-protocol.md
• Writes: wherever --write points (exports/dmp.md by convention)
• Related: /find-dataset and /collect-data (which fill the register), /ethics-application (the board's version of the same facts), /replication-package (the deposit this plan promises), /proposal (the call that asked for the DMP)

---

A data management plan is a promise about the future written in the present tense. The only part worth writing is the part something on disk can still check next year.