Interoperability between the OpenAIRE Graph and national CRIS systems is not a metadata-quality side project. It is how Europe's public research output becomes findable across borders without every institution rebuilding the same pipe. If you run a CRIS, or if your university reports into one, the decisions made at that boundary decide whether your researchers' work survives aggregation, deduplication, and citation linking.
The OpenAIRE Graph is a scholarly knowledge graph that aggregates research results, data, software, projects, funders, and organisations from repositories, journals, data archives, and CRIS systems. A CRIS, or Current Research Information System, is the national or institutional database that records publications, projects, people, equipment, and other research activity. When a national CRIS interoperates with the Graph, the country's research outputs stop being a local spreadsheet export and start participating in the European open science record.
But interoperable can mean four different things depending on which CRIS you're talking to.
The problem is not aggregation; it's alignment
OpenAIRE does not scrape national CRIS databases. It fetches metadata through standard endpoints, chief among them OAI-PMH, the Open Archives Initiative Protocol for Metadata Harvesting, and CERIF-XML. CERIF, the Common European Research Information Format, is the data model maintained by euroCRIS. It describes research entities and their relationships: people to projects, projects to outputs, outputs to funders. That relational model is the hard part. Publication metadata can be mapped easily. Relationship metadata cannot.
A CRIS record that says Smith, J., University of X without an ORCID iD is a ghost. OpenAIRE's entity linking and deduplication algorithms depend on persistent identifiers such as DOIs, ORCID iDs, grant IDs, and organisation identifiers. When a national CRIS sends only a local author identifier, the software must guess. Guessing is fine for browsing and fatal for reporting.
How the pipeline actually runs
Here is the flow in most national deployments. You start with an institutional publication record. The local research information system assigns an internal identifier. The national CRIS aggregates that record and maps it to CERIF or another export format. OpenAIRE then harvests the endpoint, normalises the fields, clusters equivalent records, and enriches the entity with identifiers from Crossref, DataCite, ORCID, and funder systems.
Every step is a crosswalk. A crosswalk is the conversion of one metadata schema into another. The term matters because crosswalks are lossy. If your CRIS enters Journal of Applied Physics as J Appl Phys, the crosswalk might not match the Crossref title. If it enters a grant number as ERC-2020-ADG-101020368 without a funder identifier, the Graph cannot attach the project to the European Research Council.
OpenAIRE publishes specific guidelines for CRIS managers. The rules are not decorative: provide a stable identifier, a title, a date, a type from a controlled vocabulary, and links to funders and projects. What trips up CRIS managers is the vocabulary mapping. A national classification of journal article may map to six OpenAIRE publication types.
This is where the surge in repository deposits has made the pipeline louder. More deposits without discipline in metadata only raises the error rate downstream.
Where national CRIS systems diverge
Norway's CRIStin and Finland's VIRTA publication information service operate as national aggregators, but they were built for national reporting first and open science second. CRIStin, for example, requires local institutions to validate publications for performance-based funding; the export to OpenAIRE is a secondary use of that data. Finland's VIRTA pulls metadata from universities and exports to multiple services, including OpenAIRE. The export pipelines work, but each has a different set of local identifiers and controlled vocabularies.
Other countries have taken different paths. Portugal's CIENCIAVITAE links individual researcher profiles to a national ecosystem; Croatia's CroRIS serves national reporting; and the Czech RIV collects results for evaluation. Each of these systems uses CERIF or a close relative in different ways. Some expose OAI-PMH, others provide custom APIs. That variety is not the failure. The failure is when a system sends free-text fields where the Graph expects identifiers.
OpenAIRE's CRIS guidelines spell out the expected fields, but reading the guideline is not the same as enforcing it. A field can be mandatory in the schema and still be empty in practice. That is what the Graph sees every month.
Why this matters for research assessment
A poorly mapped CRIS feed produces missing counts in the Graph. Research managers then see a lower publication count than the national reporting system shows. The difference is not caused by OpenAIRE losing data; it is caused by the CRIS sending ambiguous records. When output is ambiguous, the Graph may reject the record or assign it to the wrong entity. Neither outcome is neutral for a university being assessed on research output.
Institutional CRIS administrators can fix this. They need three things: a unique internal identifier that never changes, persistent identifiers for people and funders, and a controlled publication-type vocabulary. None of these are expensive. They are a discipline problem, not a technology problem.
There is also a governance problem. National CRIS systems are usually built for national evaluation, which asks a very specific set of questions. The OpenAIRE Graph asks broader questions: Which projects produced this? Which datasets back it? Which organisations collaborated? The two agendas are compatible, but only if the export mapping is treated as part of the CRIS's core function, not as a once-a-year dump.
Some systems do this well. CRIStin in Norway and VIRTA in Finland have maintained export endpoints long enough that their metadata quality improves over time. The improvements are not glamorous: cleaned affiliation strings, more complete ORCID coverage, fewer duplicate variants of the same journal name. But that is exactly what makes a national CRIS legible to the Graph.
What you should do with your CRIS export this week
Here is the practical list.
- Export one month of records from your CRIS and check whether every publication record has a DOI or another persistent identifier.
- Check how many people records include ORCID iDs. If the number is below 70 percent, start there before adding new fields.
- Look at your funder and project metadata. Ask whether the grant identifier follows a standard form the Graph can match.
- Run the same query in the OpenAIRE Graph and compare counts. Do not expect exact equality, but a persistent gap over 10 percent is a mapping problem.
- Read the OpenAIRE Guidelines for CRIS managers. Give the mapping table to whoever configures the export, not to a committee.
The last point is the one universities ignore. Interoperability is decided at configuration time. A working group can agree on principles for six months and never fix the actual XPath transform. The person who knows the system should be the person reading the guideline.
What changes next
OpenAIRE continues to tighten its entity-linking logic. The more the Graph can match, the less manual correction is needed downstream. This has a direct effect on national CRIS managers: records that previously passed through as duplicates or orphan records now need identifier hygiene from the source. That pressure will reach institutions through national mandating agencies, open science monitoring, and the EU's research assessment reform.
The previous article on Crossref and DataCite grant metadata makes the connected point: funders and publishers are linking grants at the source. A CRIS that does not align with those source systems creates a second, conflicting version of the same fact. The fix is not to argue with the Graph; it is to send it cleaner identifiers.
For research offices, the operational lesson is simple. Your researchers do not care which national CRIS you run. They care that their portfolio appears correctly when someone searches for their work. The OpenAIRE Graph is one of the places that search happens. Clean, identifier-rich CRIS feeds are how you make sure the record is right.
Photo by Gabriel Cox on Unsplash