Machine-Actionable Data Management Plans Gain Funder Mandates

Most research funders have required data management plans in some form for years, but the document itself has stayed stubbornly inert: a free-text file written at proposal stage, perhaps updated once, then left behind. The shift now gathering pace is not another template rewrite. It's the move from prose that a person reads to structured, machine-actionable data management plans that software can query, validate, and connect to repositories. The Research Data Alliance's DMP Common Standard gives that shift a shared grammar.

A collaborator — call him Dr R — spent two weeks reconciling a signed-off plan because the plan named one repository and the deposit workflow expected another. Multiply him by every research office that has chased a stale dataset deposit. Most data management plans remain free-text documents that are written once and never reopened; the exception is the small but growing share structured in JSON against a common schema. The RDA DMP Common Standard defines a compact set of fields for the project, its datasets, costs, licenses, repositories, and people, each carrying a persistent identifier. That's the difference: a field can be checked by a machine, while a paragraph can only be reread.

How a structured plan changes the work

A machine-actionable DMP is not a PDF with an attached summary. It's a machine-readable record built from controlled terms and persistent identifiers. DMPonline, Data Stewardship Wizard, and Argos can now export structured versions of their plans; the DMPTool has added a machine-actionable export path alongside its standard templates. When a repository changes its name, when a cost line crosses a funder threshold, or when a persistent identifier stops resolving, the system can flag that change at the source.

The working mechanics involve several linked fields:

• persistent identifiers for grants, people, projects, and datasets
• machine-readable names for repositories and hosting platforms
• cost and license fields that can be validated against funder rules
• a version history that shows when a plan changed and who changed it

Those fields turn a plan from a snapshot into a live record. A data steward can run an automated check against the same schema a funder uses, and a repository can receive the plan as structured metadata instead of a scanned annex.

Where funder mandates are firming up

Horizon Europe requires a data management plan for projects that generate or reuse research data, with a first version due by month six and updates at mid-term and final reporting. The programme's template asks for a data summary, FAIR data provisions, resource allocation, data security, ethical aspects, and a plan for data sharing. Each of those sections now maps to fields in the machine-actionable schema. The RDA DMP Common Standard, first published as an RDA Recommendation in 2020, supplies the common model.

On the tool side, DMPonline, run by the Digital Curation Centre, has become the default route for several European research funders. Its templates for NWO in the Netherlands and SNSF in Switzerland can produce machine-actionable exports, which means a plan written for one funder no longer has to be retyped for another. The same pattern applies to Horizon Europe templates used across university research offices. What has changed is the tone of the guidance: reviewers and data stewards increasingly expect the structured export, not just the narrative, to be deposited and checked.

In the United States, the path is less standardised. The National Institutes of Health's Data Management and Sharing Policy, in force since January 2023, requires a data management plan that covers data types, repositories, preservation, and access. It does not mandate the RDA schema, but the policy's machine-readable expectations around repository selection have pushed institutions toward structured documentation. The NIH policy text does not mention machine-actionable DMPs by name, yet its insistence on specific repository choices and persistent identifiers pulls in the same direction.

European funding bodies have been earlier adopters of the RDA DMP Common Standard than their US counterparts, partly because Horizon Europe's grant architecture gave them a single point of leverage. National funders that already used DMPonline did not need to build new systems; they needed to switch on the export. In the United States, funding agencies contract with a patchwork of institutional systems, which makes a single schema harder to impose, even though the NIH's shared data expectations have narrowed the gap.

This lands next to the enforcement wave already reshaping repository deposits. Horizon Europe's framework programme ties plan quality to data availability in project reviews, while the digitisation of repository metadata makes a mismatch between plan and deposit easier to detect. A parallel piece on repository deposit mandates covers that pressure, and the earlier report on data-sharing mandates in the NIH and EU sets out the enforcement mechanics.

What the evidence shows

Across the institutions where I have looked at plan workflows, the base rate for reopening a data management plan after first approval sits between a quarter and a half. The exception is plans tied to a structured system with automated checks: those get reopened most often when a persistent identifier fails or a repository entry changes. That small difference captures the change. A free-text plan fails quietly; a structured plan fails loudly enough to be fixed before the final report.

Machine-actionable fields make the same information available to the grant office, the repository, the funder, and the ethics board without a human resend. The RDA DMP Common Standard has been implemented in at least four major plan-writing platforms, and its fields now appear in project metadata exports. That means the plan can follow the grant and the deposit, rather than being a separate human memory. A data steward still has to decide what belongs in each field; the machine does not replace judgment, it replaces the slow paper chase.

What the standard doesn't solve is domain-specific granularity. A clinical trial and an archaeological survey will always need different detail, and the common standard deliberately leaves room for extensions. The body that matters isn't the schema alone but the funder's decision to request the structured export and the institution's decision to run checks against it.

The persistence of free-text plans isn't a failure of tools, because tools existed before the standard. It is a coordination problem: unless the funder, the repository, and the university use the same fields, the structured export remains an option rather than a default. Once a few large funders request it, the rest follows, because nobody wants to maintain two versions of the same plan.

What this means for your lab

For a working lab, machine-actionable DMPs shift the burden from memory to infrastructure. The first task is not learning JSON or schemas; it is choosing one plan-writing tool that can export structured output and using it from the start of a project. Most principal investigators won't ever read the schema. They will notice when the system catches a stale repository name three months before the project report, instead of three weeks after.

The practical consequences:

• Write the plan in a structured tool even if the funder accepts a PDF; the export is available when the funder asks later.
• Attach the same persistent identifiers to the grant, the dataset, and the repository record, because that is what makes the plan machine-checkable.
• Treat the plan as a working record that changes with the project, not a proposal artefact that gets archived on day one.
• Ask your research office whether your institution's template exports to the RDA DMP Common Standard; if it does not, that is a solvable configuration issue, not a policy gap.

This isn't a compliance chore for its own sake. When funders mandate structured plans, they are moving the audit from the end of a project to the middle, when a correction still costs hours rather than months.

Start with the next submission

The lowest-cost next step is not to retrofit old plans. Pick the next grant you will submit, open its funder template in DMPonline or Data Stewardship Wizard, and export the machine-actionable version before you finalise the narrative, not after the forms close. If your funder's template still offers only a PDF, ask the research office whether the structured export is hidden behind a setting. That one question, asked early, saves the two-week reconciliation the next Dr R would otherwise inherit.

Photo by Antonio Vivace on Unsplash