Go to the Indico Home Page Oldest event Older event Up to category Printable version
  • Compact style
  • Indico style
  • Indico style - inline minutes
  • Indico style - numbered
  • Indico style - numbered + minutes
  • Indico Weeks View
    • Switch to the management area of this event
    Login

    9th Open Science Practitioners Forum

    → Europe/Zurich
    9th Open Science Practitioners Forum
    Zoom Meeting ID
    62568646845
    Host
    Merten Dahlkemper
    Useful links
    Join via phone
    Zoom URL
    Hide

    Date: 28 October 2026, 15:00–16:55 (CET), Zoom

    Chair: Clemens Lange (CMS) 

    Minutes: Merten Dahlkemper (CERN Open Science Office) 

    Topic: Open science infrastructure provided by CERN

    1. News from the Open Science Office

    Anne Gentil-Beccot

    Team. Paulina Birtwistle joined the Open Science Office on 1 September 2026 as Open Access Manager, moving over from the library section of the Scientific Information Service (SIS). The office now consists of Anne Gentil-Beccot (Open Science Coordinator), Antonia Winkler (Data Steward), Merten Dahlkemper (Community Engagement Manager), Micha Moskovic (INSPIRE Community Manager), Pia Kretschmar (Open Access Junior Manager, SCOAP3 operations), and Paulina Birtwistle (Open Access Manager).

    New websites. The former SIS website has been split into three: openscience.cern (support on all open science practices), library.cern, and archives.cern. The three services remain within one group.

    Data management.

    • A ServiceNow support line for data management is being implemented and will go live within days, reachable via research-data-management@cern.ch. It provides guidance and advice and does not replace the support offered by the CERN Open Data Portal team, to which portal-related questions will be redirected.
    • Discussions have started with IT-CA on research data management guidelines for the CERN community, so that advice on when to use HEPData, CDS, Zenodo, or the CERN Open Data Portal is aligned.
    • HEPData training sessions have been run for small experiments (NA64, SHiP, MoEDAL, SND@LHC), with more to come. Training material on other open science practices is planned.
    • The first Open Science Story, on ATLAS Open Data, was published on the Open Science website last week and will appear in the CERN Bulletin next week. More stories will follow in the coming months.

    STARDAST. CERN (Antonia Winkler, Merten Dahlkemper, Micha Moskovic, Anne Gentil-Beccot) participates in STARDAST (STewardship And Recognition for DAta Science Talent), a Horizon Europe project to professionalise data stewardship through curricula, training, and a European network of data stewards. The project involves 19 organisations, is led by the European Molecular Biology Laboratory (EMBL), and started on 1 September 2026. CERN contributes to all work packages and will lead Work Package 4 (knowledge dissemination), which starts in about two years.

    Open access.

    • Thanks to the Swiss consortium of libraries, the Springer Nature agreements have been renewed and run until 2029. The agreements with IEEE and the American Physical Society (APS) are about to be renewed for 2027–2029.
    • SCOAP3 (Sponsoring Consortium for Open Access Publishing in Particle Physics) is in Phase 4, which ends in late 2029. Following a survey of publishing practices in the high-energy physics (HEP) community this year, the partners are developing a scenario for Phase 5 (from 2030).
    • J-QUANT, a new quantum physics journal published by SISSA, is fully open access and funded by sponsorship, with CERN as one of the sponsors. It is free for authors and readers.

    Open Research Europe (ORE). CERN now hosts and operates ORE, the multidisciplinary open access publishing platform with transparent post-publication peer review, created in 2021 and funded by the European Commission together with other European funders. A dedicated team sits in the Open Science section, with editorial services outsourced. The platform is already live; the official launch announcement follows next week. As CERN is a funder, CERN authors and collaborations are eligible to publish there. A webinar takes place on 1 December (registration: https://cern.ch/ore-launch).

    OSPO community event. The CERN Open Source Program Office (OSPO) holds a community event on 13 November 2026 at 14:00 in the Council Chamber (Indico). The call for lightning talks is still open.

    Open Science Steering Board (OSSB). Reporting on behalf of Kamran Naim, Anne summarised the OSSB meeting of 20 October 2026, chaired by the CIO. The main item was the CERN Open Science Strategy 2027–2030. A draft is being finalised and will go through internal and external consultation, with the aim of discussion at Council in March 2027. In parallel, work will start on the revisions of existing policies (open science, open data, open access) that the strategy requires.

    Contact: os-office@cern.ch (general), open-access-questions@cern.ch, research-data-management@cern.ch

    No questions.

    2. INSPIRE

    Micha Moskovic

    INSPIRE is the main information platform for HEP and related fields (nuclear physics, and more recently quantum information), with about 1.9 million literature records, about 100,000 author profiles curated by humans, and about 25,000 visits per day from almost every country. Its records go back to the early 1970s and its predecessor SPIRES, which was the first website in the US and the first database on the web. INSPIRE is run by a collaboration of HEP laboratories and a library (DESY, Fermilab, IHEP, IN2P3, and TIB).

    Its main strength compared with generalist platforms such as Google Scholar is its focus and data quality. arXiv and published versions are merged into a single record, so citations are not double-counted, and reference lists are extracted from the full text. INSPIRE also offers HEP-specific metadata (collaborations, experiments), a powerful search syntax, and services such as jobs, conferences, and seminars. A data collection is in beta, currently with HEPData as its only source.

    Curation under pressure. Records pass through harvesting (arXiv, publishers, user submissions), pre-processing (matching, merging, content selection), and curation. Curation is the most time-consuming task, and the effort per record has fallen steadily: from manual data entry before the web, to verification, to further automation on the current platform. The reduced involvement of SLAC (2018) and DESY (2025) has only partly been compensated by new partners (STFC, TIB, with INFN expected to join). INSPIRE has stopped some systematic checks, increased automation (for example for affiliations) and crowd-sourcing (for example for references), and added quantum information to attract support, which also brought a very high volume of papers. Since 2025, the goal of manually curating every core record has not been met and a backlog is building up. The proposed way forward is to move from systematic curation to flagging and checking problematic records, based on missing metadata or suspicious content.

    Sustainability and user priorities. The European Strategy Group's deliberation document for the European Strategy for Particle Physics Update states that INSPIRE's long-term sustainability requires renewed international commitment and adequate funding from the European particle physics community and beyond. To decide where to focus limited resources, INSPIRE will launch a community survey before the end of 2026.

    Modernisation. Since Q3 2026, the back office (dating from 2016, based on Python 2 and a bespoke workflow engine) runs on a new platform based on Apache Airflow (presented at the IT Technical Forum). The record editor is being reimplemented, and the user interface has received major updates behind the scenes.

    AI. Bot traffic is still manageable, but low-quality AI-generated submissions are increasing, and the content policy is being revised accordingly. INSPIRE is exploring integration in both directions: a Model Context Protocol (MCP) server, so that INSPIRE can be queried from AI assistants, and natural language search and summarisation within INSPIRE. With limited in-house expertise, INSPIRE is seeking collaborations, including a PhD project.

    Plans for 2027: focus manual effort where it has the most impact; simplify the institution collection by removing departments, so that affiliations point to the top-level institution; add data sources to the data collection (CERN Open Data, Zenodo, and others); and extend it to research software.

    Discussion. No questions. The chair stressed how much researchers rely on INSPIRE, for example to keep publication lists up to date as members of large collaborations.

    3. Updates to the CERN Open Data Portal

    Diana Rand, on behalf of the CERN Open Data team

    The CERN Open Data Portal hosts more than 5 PB of data in over 80,000 entries and more than 3 million files, from nine experiments (the four large LHC experiments, fixed-target and legacy experiments, and non-CERN experiments).

    Recent updates.

    • JADE has published its first data on the portal. The first OPAL release and theoretical simulated data from MCnet will follow in the coming months.
    • New releases from ATLAS and LHCb, including ntuples from the LHCb Ntupling Service.
    • Record IDs now carry an experiment prefix (for example atlas-160006) instead of being purely numerical.
    • Cold storage, in production since last year, has been further improved (file transfers, user experience), as has monitoring.

    New release process (in production since June 2026). Previously, experiment curators had to test metadata on a local instance of the portal and contact the Open Data service managers several times: for identifiers such as DOIs, for deployment to the QA instance after a GitHub pull request, and for deployment to production. Curators can now do all of this in the web interface of the production portal:

    • Log in with CERN SSO. Access to release management is granted per experiment.
    • Create a draft release by uploading a JSON metadata file or loading it from a URL, then add, edit, preview, or bulk-edit records and documents (including image upload).
    • Automatic validations check schema, identifiers, metadata consistency, files (no duplicates, no missing files), and links and images in documents. Several failures can be fixed automatically, and experiment-specific or optional validations can be added.
    • Staging makes the release read-only and visible to curators with the relevant permissions, with a disclaimer. Staged records do not appear in search results but can be shared by link for review. DOIs can be generated at this point, and the release can be rolled back to draft.
    • Publishing makes the records public. It is one click and cannot be rolled back.

    ATLAS and CMS have already used the new process for several small releases.

    Next steps: test the process with larger releases (many records and files) and tune it based on curator feedback; integrate with experiment data management systems (for example retrieving datasets from Rucio); and integrate automatically with the LHCb Ntupling Service.

    Discussion.

    • Matt Bellis asked whether existing record IDs would be reassigned, and whether tools such as the CERN Open Data client are affected. Diana Rand: existing record IDs continue to work and also resolve with the prefix; new records always carry the prefix. Tibor Šimko confirmed that the client treats record IDs as strings and is not affected.
    • Clemens Lange asked whether there is documentation of the new process for the experiments' data curation teams. Diana Rand: so far no, but the team will consider it. Clemens noted that central documentation would help, as experiments such as CMS currently maintain their own and the number of experiments on the portal is growing.

    4. REANA

    Tibor Šimko

    REANA is a containerised platform for running, organising, and sharing declarative computational workflows written in popular workflow languages (CWL, Dask, Serial, Snakemake, Yadage) on Kubernetes, HTCondor, or Slurm.

    REANA 0.95.0-alpha.6 (released 2 October 2026) comprises about 156 pull requests across 15 repositories since the previous alpha in June, developed with extensive use of AI-assisted coding. Highlights:

    • A faceted web interface for the workflow list (own and shared workflows, filters by status), plus shareable links that point to individual log lines or workspace files.
    • A new login and token flow based on OpenID Connect (OIDC) and JSON Web Tokens (JWT): reana-client login opens the browser, and the client can store and switch between several REANA servers, similar to Kubernetes or Docker contexts. Using the same identity provider and token flow as other services should make single sign-on across services (notebooks, storage) possible.
    • Server-side workflow validation in a sandboxed pod, which makes the client thin and free of heavy workflow tooling.
    • As a result, the first complete release of the Go client, with feature parity with the Python client, standalone executables for Linux and macOS (ARM64 and AMD64), and 5–10 times faster performance.
    • An upgrade to Snakemake 9.27, with container images validated before launch so that errors are reported early instead of leaving workflows hanging.
    • HTCondor: an upgrade to the HTCondor 2 Python API, submission from ARM64 clusters, and forwarding of Kerberos credentials to jobs (container and CVMFS/Apptainer modes) for access to restricted data. This has already been used for CMS reconstruction tests.
    • Slurm: a personal Apptainer image cache per user, reused across workflows and jobs, which avoids repeated conversions and clashes between jobs converting the same image.

    Community.

    • At CHEP 2026, REANA presented its Dask support. Discussions there covered workflow languages popular in the community, including CWL (used internally by the LHCb DiracX developers) and law (originating in CMS and now also used in ATLAS). Whether REANA should support law is an open question.
    • REANA lessons were part of the HEP Software Foundation (HSF) / IRIS-HEP training on analysis reproducibility.
    • In June 2026, a joint hackathon with the Leibniz Institute for Astrophysics Potsdam (AIP) explored using REANA as a backend to validate LLM-authored workflows and reduce hallucinations. This work is at an early stage.
    • Sully Billingsley (ATLAS) is extending the REANA web interface for the ATLAS H→4ℓ analysis (interactive plots, systematics, correlation matrices). Celebi, an analysis management toolkit for HEP used in part of LHCb, also has a React-based visualiser. Making the REANA interface more easily pluggable could be a topic for next year.

    EOSC. Release 0.95.0-alpha.6 is so far deployed only on the dedicated REANA EOSC instance. The team is stabilising it towards a production 0.95 release for wider deployment at CERN, to be accompanied by blog posts. REANA is a partner in the European Commission-funded EOSC Mesh project (September 2026, three years), which aims to build the CERN Node of the European Open Science Cloud (EOSC) Federation and support cross-node federated workflows.

    Discussion.

    • Clemens Lange asked what determines the release cadence, given that fixes (for example the container issues a few months ago) cannot reach production quickly. Tibor Šimko: the current series changes core functionality (login, client) and is not backwards-compatible, so the features must be stabilised before a release can be cut. Preparing a release across all components takes about half a day, and the team is changing substantially, with a largely new team next year. Automated release-note generation has been adopted, and releases should become faster once the new generation has landed.
    • Clemens Lange asked about contributions from outside the CERN team. Tibor Šimko: contributions are rising compared with a year ago, including pull requests from CMS and ATLAS members and from students, helped by AI coding tools. EOSC will bring new users from non-physics communities (for example life sciences); how many will contribute to development remains to be seen. The community is growing slowly.

    5. Zenodo

    Nicola Tarocco

    Zenodo is a catch-all repository for research content from any discipline, free and open to any researcher worldwide. It is not a preprint server, applies no peer review, and does not replace institutional repositories (at CERN, CDS is the designated repository for CERN research). Curation can be delegated to community owners, who review submissions to their communities. Zenodo holds about 8 million uploads (about 2 PB) from about 1.3 million users. The largest community by number of records is biodiversity, which deposits images and data extracted from papers so that they can be cited individually. Zenodo runs on InvenioRDM, developed with CDS, the CERN Open Data Portal, and external partners, so new features also benefit CDS and other repositories.

    Projects 2026–2027. Zenodo is mainly funded through external grants.

    • HORIZON-ZEN+ (European Commission, 2025–2027) supports the EU Open Research Repository on Zenodo. In 2026 the focus is on self-service features to reduce support requests: deleting a record shortly after publication, modifying files, and requesting more quota. In 2027 the focus moves to new citation sources, geolocation metadata (with map display and search), and highlighting machine learning-ready and AI-ready datasets.
    • AIRDEC (The Navigation Fund, 2025–2027), with the CDS team in SIS, extracts metadata from uploaded PDFs to pre-fill the submission form. Users always review and confirm the suggestions; the AI never changes metadata on its own.
    • EVERSE (European Commission, 2024–2027): Zenodo will integrate with gitlab.com for software archiving, as it does with GitHub today. CDS already offers this for CERN's GitLab instance.

    New user policy (2027). Zenodo has so far trusted its users to upload in-scope research content. Generative AI has made it easy to produce fake research at scale, and Zenodo now receives a constant stream of out-of-scope content. The current mechanisms (automated spam detection, manual verification of email domains, spotting by the team while on support duty, and user reports) do not scale, and there is no dedicated funding for content moderation. At the same time, AI bots harvesting content (total file bandwidth of about 1 GiB/s) have caused several outages and stress CERN's infrastructure, Zenodo probably being CERN's largest web platform. Infrastructure improvements remain a cat-and-mouse game. Zenodo receives 50–70 new support tickets per day, handled by one engineer per week on rotation.

    From 2027, accounts will therefore be split in two:

    • Verified accounts log in through an institution on a validated list of research organisations and reconfirm their affiliation periodically. They can do everything they can do today.
    • Unverified accounts (for example personal Gmail logins without a linked institutional account) can access content and create unpublished, non-findable drafts, but cannot publish until they link an institutional account.

    Funding. The team is excellent but not sized for this scale. Grants fund projects, but not maintenance, operations, or support: "Projects build it. Who keeps it running?"

    Discussion. Clemens Lange remarked that Zenodo is to some extent a victim of its own success and radical openness, and that others will learn from the changes. Nicola Tarocco said that solutions are being discussed in various formats, and that the similar issues raised by INSPIRE suggest teaming up.

    6. The CERN Node in the EOSC Federation

    Giovanni Guerrieri, on behalf of the CERN EOSC Task Force

    The EOSC Federation, described as "a WLCG for open science" (after the Worldwide LHC Computing Grid), aims to be a "system of systems" that helps researchers in all disciplines store, share, process, analyse, and reuse FAIR (findable, accessible, interoperable, and reusable) research outputs. It has been in production since 8 October 2026, with 14 nodes (thematic or national) that comply with a common set of federating capabilities: common authentication and authorisation, a shared catalogue of resources and services, helpdesk, service monitoring, and a service management system (EOSC Federation handbook). The CERN task force has about 30 members across IT and OSI [TODO: spell out OSI on first use], and is seeking standing representation from the physics communities.

    Vision and principles. In the research lifecycle (plan, collect, analyse, preserve and reuse), data collection is largely handled by the collaborations. Analysis, however, often resembles the "Wild West", and preservation for future reuse tends to be neglected. The CERN Node aims to give researchers a common space to access data, resources, and repositories without connecting the technical dots themselves. It offers "lazy" solutions with the shortest path from infrastructure to physics, is modular rather than a monolithic analysis facility, and focuses on interoperability: authenticate once, then use every service one is cleared for. Three principles apply:

    • No researcher, no node: the infrastructure follows community needs, with arXiv:2605.12229 as one baseline and first contacts with experiments on use cases (CMS reconstruction, LHCb data processing, ATLAS theory).
    • Usability first: HEP users are "ruthless" and rarely return to a service that does not work, so effort goes into the first five minutes.
    • Community engagement: use cases from the communities take priority.

    Architecture and stages. Users from any institution log in through the EOSC Federation (MyAccessID, including CERN SSO), and access to individual services is granted per group after moderation. Interactive services (SWAN as virtual research environment, REANA) run in a network-isolated segment at CERN. lxplus, batch, the WLCG grid, and experiment frameworks are not accessible. Existing services (CERN Open Data, CERNBox, Rucio, Zenodo, INSPIRE, HEPData) are linked. The roll-out has three stages:

    1. Run what we published (available): curated open data, tutorials, and ready-made analyses in the browser for education and outreach.
    2. Run what you wrote (in progress): own code and reproducible workflows on open data, for example for theorists and phenomenologists, with extra capacity decided case by case.
    3. CERN users with enhanced capacity (planned, not scheduled).

    Ready-made analyses from the ATLAS, CMS, and LHCb open data groups can be run directly in SWAN or REANA, with software from CVMFS and data read in place from EOS. A 10,000-event file runs through Rivet in about 4 minutes on one core, and a 100,000-event study in about 20 minutes on two cores. Researchers can bring their own code, make it reproducible with REANA and Rucio, and publish via Zenodo (DOI) linked to INSPIRE and HEPData. As a non-HEP example, the FORGE project (a small team, about six months, EOSC Gravity cascading grant) used Rucio and REANA to transfer and analyse genomics data across nodes. This worked, but is still manual, uses simulated data, and offers no path for sensitive data yet.

    Community engagement is the current priority. The task force asks communities for "one workflow you cannot run easily today" (for example ML training that needs GPUs or high-performance computing, analysis facilities, phenomenology with open data, or teaching), to be collected in a portfolio shared across the Federation. In a year or so, a Federation "jamboree" should invite users to stress-test the services.

    Open issues: service levels are best effort, with no availability target yet; the Federation has not settled a policy on assurance levels and multi-factor authentication (MFA), and the CERN Node is pushing for one; there is no final storage retention policy, so users should keep their own copies of results; there are no GPUs on the node; and cross-node jobs (a job at one node on data at another, with one login) are targeted for March 2027.

    Try it: eosc.cern. Alpha testers with a CERN account can request access to the interactive services via eduVPN at https://eosc-cern-node-dev.app.cern.ch/services (access is moderated and may take some time).

    Discussion.

    • Merten Dahlkemper asked whether there is a forum or mailing list for the planned community engagement. Giovanni Guerrieri: a newsletter is being prepared, and the task force would like to use the OSPF to disseminate it once ready.
    • Clemens Lange asked which computing resources are available and whether users can scale out to the batch system. Giovanni Guerrieri: there is no access to CERN batch; users get a few cores and some memory within the isolated segment. Once cross-node communication is in place, jobs could be sent to compute resources at other nodes (for example SURF in the Netherlands), similar to how WLCG works today.

    7. Any other business

    • Merten Dahlkemper asked the experiment representatives whether a meeting like the OSPF, but focused on specific infrastructures and technical issues, where experiments and infrastructure owners discuss concrete problems, would be useful. The idea was raised in the Open Data working group for the CERN Open Data Portal but could apply to any infrastructure.
    • Matt Bellis said the discussion had been very useful. For CMS, the main current issue is the upcoming disk limitations for the CERN Open Data Portal, on which CMS is working with all parties involved.
    • Clemens Lange suggested reaching out to specific users to find out whether they feel sufficiently supported.
    • Zach Marshall agreed and noted that service providers may know better than experiment coordinators which members of an experiment use their services: in a 6,000-person collaboration, heavy users are not always visible. He offered to help bring these users together where possible ("but it is cat herding") and thanked the organisers for a very useful meeting.
    • Clemens Lange reminded participants of the OSPO community event on 13 November 2026 and its open call for lightning talks.

    Follow-ups

    • CERN Open Data team: consider central documentation of the new release process for experiment curators.
    • CERN EOSC Node task force: share the newsletter once available; the OSPF will help disseminate it. Communities are invited to submit use cases ("one workflow you cannot run easily today") and to sign up as alpha testers.
    • INSPIRE: community survey on user priorities to be launched before the end of 2026.
    • OSPF organisers: follow up on the proposal for a more technical, infrastructure-focused meeting, including reaching out to users identified by the service providers; publish the 2027 OSPF schedule with topics.

    Next meeting

    Probably January 2027. A schedule for 2027 with planned topics will be published soon.

    There are minutes attached to this event. Show them.