top of page

The DS2 Publication Report from EBDVF 2026

Laura Gavrilut
10 minutes ago
3 min read

The DS2 publication report from EBDVF 2026 is out. From Vision to Reality: Lessons and Priorities for Data Sharing, Interoperability and AI is now on Zenodo, free to read and reuse under a CC BY 4.0 licence.


It's a joint effort in between DS2 and its EU-funded sister projects: PLIADES, CyclOps, CEDAR and NOUS. The brief captures the session we ran together at the European Big Data Value Forum in Galway, Ireland, on 30 September to 1 October 2026. Its focus is what happens when European data spaces move from vision to real use.


Europe has spent years building the plumbing for data spaces: connectors, catalogues, identity systems and shared standards. Our session started from a harder question. How can independent data spaces exchange data without weakening control, trust or accountability?


All five projects kept arriving at the same answer. Technical connectivity alone doesn't create trust. A data provider needs to know their conditions still hold once data crosses an organisational, sectoral or national boundary. A data user needs to know what they're actually allowed to do with what they've received. Until both are true, data stays where it is.


Five projects, five angles on the same problem

Each project brought lessons from its own pilots, and together they cover the journey from raw data to trusted, AI-ready exchange.

  • DS2 tested trusted exchange between data spaces in city, environmental and agriculture use cases, and found that modules working on their own don't guarantee interoperability: it has to be proven end to end, from onboarding and discovery to negotiation, enforcement and transfer.

  • CyclOps, which automates the data and AI lifecycle and validated it in tourism, concluded that one connector for every data space is out of reach, and that policies and usage conditions must travel with the data.

  • NOUS links data spaces with cloud, edge, AI and HPC, and showed how open-source building blocks and a shared sandbox let others test, integrate and build on its work.

  • CEDAR, piloting robust AI for public governance in Italy, Slovenia and Ukraine, found that public bodies often withhold data because granting permission is legally uncertain, not because they're unwilling, and points to clear agreements and trusted intermediaries as the fix.

  • PLIADES, working on AI-enabled data lifecycles in mobility and industry, showed that data quality work helps AI only when it's tuned to the task: removing near-duplicate frames lifted accuracy from 79.4% to 81.0% mAP, while blurring faces and licence plates cut it from 79.1% to 71.2%.


The common threads

Five projects working in different sectors still converged on a few shared lessons.

  • Proof beats specification. Reference architectures don't validate interoperability by themselves. A specification becomes credible only when independent organisations deploy it, establish trust, exchange data and audit the result.

  • Rules have to travel with the data. Access is only half the story. Identities, usage policies and provenance need to be machine-readable and stay attached as data moves into downstream AI workflows.

  • The barriers are legal and organisational as much as technical. Public administrations and SMEs face heavy compliance and onboarding burdens. Simpler toolkits, standard agreements and trusted intermediaries lower the bar.

  • Data quality is a trade-off, not a free win. Cleaning and anonymising data can raise or lower AI performance. The work has to be tuned to the task.

  • Open and testable wins. Sandboxes, open components and real operational data let others build on the results instead of starting from scratch.


Who should read it

  • Policymakers will find concrete ideas for unblocking public-sector data sharing, from standard agreement frameworks to funded data steward roles.

  • Researchers get lessons and numbers from real pilots: what moved AI performance, what didn't, and where semantic standards still fall short.

  • Innovators and data space builders get a candid view of what it takes to deploy interoperability in practice, and where to find open components to start from.

Comments


bottom of page