Wilkinson FAIR Data Principles, Ten Years On

The Wilkinson FAIR data principles — Findable, Accessible, Interoperable, Reusable — were set out in a 2016 Scientific Data paper as guidance for machine-actionable data stewardship, not a certification standard. Ten years on, the four-letter framing has held up remarkably well; the fifteen sub-principles beneath it, and the infrastructure needed to satisfy them, have proved far harder to deliver than the paper’s authors could have anticipated.

The FAIR Guiding Principles for scientific data management and stewardship is the short, formal name for the 2016 paper by Mark D. Wilkinson and more than 50 co-authors, published in Nature’s Scientific Data on 15 March 2016 and now cited more than 22,000 times on Google Scholar. This piece assesses — with the benefit of a decade of implementation evidence — which parts of that original argument have proved durable, which have not, and what has changed around them.

• What did the 2016 paper actually propose?
• Which principles have held up best?
• Where has implementation fallen short?
• How has FAIR evolved since 2016?
• Common questions about the FAIR data principles
• What this means for research administrators
• The next decade for FAIR

What did the 2016 paper actually propose?

Wilkinson et al. did not propose four principles; they proposed fifteen. The paper groups them as F1–F4, A1–A1.2 and A2, I1–I3, and R1–R1.3, covering persistent identifiers, rich metadata, standardised access protocols, shared vocabularies, and licensed provenance. The four-letter acronym was always a mnemonic for a longer, more technical checklist aimed primarily at machines, not a plain-English summary for humans.

Crucially, the authors were explicit that FAIR is not synonymous with open. Data can satisfy every FAIR criterion while remaining access-restricted, provided the metadata describing it — and the conditions under which access can be granted — are themselves findable and accessible. That distinction, frequently blurred in institutional communications since, is one of the paper’s most durable and most widely misapplied contributions.

Which principles have held up best?

The Findable and Accessible criteria have aged the most gracefully. Persistent identifier infrastructure — DOIs for datasets via DataCite, ORCID for contributors, and standardised repository indexing — has matured into everyday practice at most research-intensive institutions, largely because it required infrastructure investment rather than cultural change.

Interoperable and Reusable have proved far more difficult, because they depend on domain communities agreeing shared vocabularies and licensing conventions — a social coordination problem the original paper flagged but could not solve by publication alone.

Where has implementation fallen short?

The gap between stated intent and practice is the paper’s clearest unresolved legacy. A 2017 critique by Dunning, de Smaele and Böhmer — “Are the FAIR Data Principles Fair?”, published in the International Journal of Digital Curation and now cited over 100 times — argued early that the “Interoperable” and “Reusable” criteria were too abstract to audit consistently across disciplines, a criticism that a decade of maturity-model attempts has only partly answered.

The principles are also silent on a question their authors never claimed to address: who controls data about people and communities. In 2019, the Global Indigenous Data Alliance introduced the CARE Principles — Collective Benefit, Authority to Control, Responsibility, Ethics — explicitly as a complement to FAIR, arguing that findability and reusability without governance and consent can entrench, rather than correct, historical power imbalances in Indigenous data.

• 2017 — GO FAIR launches as an international, self-governed implementation network for the principles.
• 2018–2020 — The Research Data Alliance’s FAIR Data Maturity Model Working Group builds indicators to operationalise assessment.
• 2019 — The CARE Principles for Indigenous Data Governance are published as a deliberate complement to FAIR.
• 2019–2022 — The EU Horizon 2020 FAIRsFAIR project builds CoreTrustSeal-aligned FAIR assessment tooling for repositories.

How has FAIR evolved since 2016?

None of this evolution happened inside the original paper — it happened around it, in exactly the pattern research-infrastructure standards tend to follow once their originating authors hand stewardship to a community. GO FAIR, the international support network established in 2017, now coordinates implementation guidance rather than the original authorial group. Funders have since operationalised the principles directly: UKRI’s research councils, including NERC’s Environmental Data Service, treat FAIR compliance as a condition of data management plans rather than an aspirational goal, tying the 2016 criteria to funding decisions a decade later.

This drift from originating paper to distributed stewardship community is a familiar shape in research-standards governance. CASRAI originated the CRediT contributor role taxonomy in 2014; the standard is now stewarded by NISO as ANSI/NISO Z39.104-2022, not by its original authors. FAIR’s authors similarly ceded day-to-day implementation authority to GO FAIR and to funder policy teams — a transition research administrators should expect, and plan governance around, whenever they adopt an originating paper as institutional policy.

Common questions about the FAIR data principles

What are the four pillars of the FAIR data principles?

The four pillars are Findable, Accessible, Interoperable, and Reusable. Wilkinson et al.’s 2016 paper expands these into fifteen numbered sub-principles covering persistent identifiers, standardised access protocols, shared vocabularies, and licensed provenance — a checklist of conditions, not a single pass/fail test.

What are the FAIR data principles of UKRI?

UKRI expects funded researchers to manage data so it is Findable, Accessible, Interoperable, and Reusable, using persistent identifiers and recognised repositories. Research councils such as NERC apply FAIR compliance as a condition within data management plans, directly linking funding decisions to the original 2016 criteria rather than treating them as optional guidance.

What this means for research administrators

Institutions that treated FAIR as a one-off compliance checkbox around 2016–2018 are now the ones struggling most with legacy data that is technically described but practically unusable. The principles that required infrastructure spend — identifiers, repositories, indexing — were achievable through procurement. The principles that required disciplinary consensus on vocabularies and licensing were not, and remain the primary bottleneck a decade later.

Research administrators managing data policy should treat “Interoperable” and “Reusable” as ongoing governance commitments requiring domain-specific standards work, not properties a dataset acquires once at deposit. Budgeting for provenance documentation and licence clarity at the point of data creation, rather than retrospectively, remains the single most effective lesson the last ten years of implementation evidence supports.

The next decade for FAIR

Ten years of implementation evidence supports a mixed but broadly positive verdict on the 2016 paper: its diagnosis was right, its four-letter framing proved durable and memorable, and its harder technical requirements around interoperability and reuse remain genuinely unsolved rather than merely under-adopted. The principles were never meant to be a finished specification, and the community built around them — GO FAIR, the Research Data Alliance, and funder policy teams — has treated them accordingly, iterating through maturity models and complementary frameworks like CARE rather than rewriting the original text.

That iterative pattern, more than any single technical fix, is probably Wilkinson et al.’s most consequential legacy: a paper written to be operationalised by a distributed community, not owned by its authors.