A curated global dataset of open science initiatives and policies

I. ARTICLE AND DATASET INFORMATION

• The 8 October 2026 Zenodo release contains 1,186 records and 71 variables spanning initiative identity, provenance, geography, status, activities, governance, funding, and identifiers.The dataset distinguishes standalone initiatives, services, deployments, umbrella organisations, and policy instruments.Source: §I · pp. 1–2
• Candidates came from directories, policy documents, official websites, and community suggestions, with identities checked, duplicates reconciled, and supporting evidence reviewed.Records may cover several open-science dimensions, and uncertainty is retained when sources do not settle a classification.Source: §I · pp. 1–2

Dataset metadata

• The archived dataset is a public Zenodo release, while the companion dashboard is updated separately.The release is version 1.0.0, published on 8 October 2026, under CC BY 4.0.Source: §I · p. 2, Dataset metadata · p. 2

1. Rationale and scope

• Inventories use different units and categories, so comparing infrastructures, services, networks, and policies requires a common structure and traceable sources.Infrastructure directories, monitoring inventories, and policy reports serve different purposes.Source: §1 · p. 2
• COSMI Hub was developed within OPENIT, grew from ASAPbio's prototype directory, and expanded beyond open-access initiatives to broader open-science activities.The article describes the CSV snapshot archived on 8 October 2026; the dashboard draws on an evolving working database and may show different totals.Source: §1 · pp. 2–3
• Because COSMI Hub is a purposive, non-probability collection, it cannot estimate how many open-science initiatives exist worldwide.Its fields support descriptive comparisons and independent recoding across infrastructures, governance, policy, and geography.Source: §1 · p. 3

2.1. Observation unit and inclusion criteria

• Each row represents a documented object, with record_level separating initiatives, services, deployments, and umbrellas from functional type and initiative_family links.Related services and deployments remain distinct records, so the collection is not a list of independent legal entities.Source: §2.1 · p. 3

2.2. Source selection and candidate identification

• Candidates came from a prototype directory, other inventories, official and documentary sources, community suggestions, and policy reports, but directory inclusion alone did not qualify a record.The working integration log recorded 171 distinguishable policy instruments across 27 EU Member States and eight comparator countries, as documented in 2026.Source: §2.2 · pp. 3–4

2.3. Harmonisation, screening and deduplication

• Names, websites, source identifiers, and initiative families were checked to reconcile duplicates; distinct services and deployments remained separate records.Unresolved matches and individual review decisions remain in working audit sheets rather than the archived CSV.Source: §2.3 · p. 4

2.4. Topical categorisation and organisational attributes

• Records can receive multiple evidence-supported open-science dimensions, while “possible” marks inconclusive evidence rather than forcing a single primary category.Organisational attributes are separately recorded from documentary sources, and the authors did not conduct systematic independent double-coding or formal adjudication.Source: §2.4 · pp. 4–5, Table 1 · p. 5

2.5. Identifier enrichment and entity matching

• External identifiers require entity-level checks because an identifier may refer to a host institution or umbrella organisation rather than a particular service.The enrichment exercise added optional ROR, Wikidata, and TSOSI identifiers, with proposed matches checked manually.Source: §2.5 · p. 6

2.6.1. File integrity and structural checks

• The archived CSV has 1,186 rows and 71 columns, 1,186 unique record IDs, and no repeated names after case-insensitive whitespace trimming.These checks establish structural properties, not the substantive accuracy of topical, governance, or operational-status codes.Source: §2.6.1 · p. 6

2.6.2. Coverage and missing data

• 724 records have unknown operational status, versus 416 ongoing, 38 discontinued, six pilot/pre-production, and two planned.The snapshot contains 987 standalone initiatives, 135 services, 32 umbrella entries, and 32 deployments; comparisons involving ongoing initiatives are sensitive to uneven status information.Source: §2.6.2 · p. 6
• The release has substantial missingness in some fields, including 1,066 missing or “NA” ROR identifiers, while latitude and longitude are absent for ten records.Working source registries and audit worksheets were not deposited, so readers cannot reconstruct every inclusion or coding decision; changing source pages also limit retrospective verification.Source: §2.6.2 · p. 6, §2.6.3 · pp. 6–7

3. Description of the dataset

• The Zenodo v1.0.0 snapshot is a UTF-8 CSV released on 8 October 2026 with 1,186 records, 71 variables and a CC BY 4.0 licence.Its archived MD5 checksum is ca754336562d01eb63e16814afd086fe.Source: §3 · p. 7
• The variables cover identity, identifiers, geography, provenance, record level, operational status, thematic activities, governance, funding and data quality.Table 1 lists ten thematic variables, while Table 2 explains selected fields; the tables are not a complete field dictionary.Source: §3 · p. 7
• The nonprofit and non_profit fields have differing definitions, value sets and provenance that require a verified mapping before reuse.Codes such as “unknown” and “possible” have field-specific meanings and should not automatically be recoded as “no.”Source: §3 · p. 8
• record_level describes whether a record is a standalone initiative, service, deployment or umbrella organisation, while object_type_openit describes functional type.initiative_family links related records without implying a formal legal parent–child relation.Source: §3 · p. 8

3.1. Field definitions and coding conventions

• The archived fields combine historical and harmonised coding: original_category retains an earlier single label, whereas ten dim_* variables provide multilabel topical coding.The deposit lacks a source registry, detailed coding-decision log and complete dictionary of all 71 variables.Source: §3.1 · p. 9
• The fixed Zenodo snapshot and evolving dashboard may differ in counts, fields and operational statuses; reported article figures refer to the archived CSV.The working database is maintained with controlled vocabularies, candidate lists and audit worksheets.Source: §3.1 · p. 9

4. Reuse potential and precautions

• Check record_level and initiative_family before counting rows as independent initiatives, because services and umbrella organisations may be recorded separately.Geographic fields indicate a documented base or representative location, not necessarily the territory served; cross-country comparisons also depend on source coverage, language and documentation.Source: §4 · p. 9
• Before converting dim_* fields to binary indicators, specify how “possible” entries are handled; Jaccard similarity measures overlap between two subsets of record IDs.The dashboard uses an evolving working dataset, so its maps, typologies and co-occurrence results may differ from analyses of the archived CSV.Source: §4 · p. 9
• For reproducibility, initially import the UTF-8 CSV as text, retain id and source_record_id, document recoding and filters, and cite the version-specific Zenodo DOI.Source: §4 · p. 9
• The dataset supports examining thematic overlap, gaps in external identifier coverage and relationships between umbrella organisations and services.Release comparisons can track collection changes only if newly identified records are distinguished from changes in the initiatives themselves.Source: §4 · p. 9

5. Data access, licensing and preservation

• The archived cosmihub_master_2026-10-08.csv contains 1,186 records and 71 variables and is licensed under CC BY 4.0.The live dashboard and supplementary project documentation are linked separately.Source: §5 · p. 10
• The version-specific DOI identifies the fixed 8 October 2026 dataset, and its MD5 checksum can check whether a download matches the archived copy.CC BY 4.0 permits redistribution and adaptation with attribution and an indication of changes.Source: §5 · p. 10
• Source inventories remain separate resources with potentially different licences, and the dataset and paper should be cited as separate research outputs.Source: §5 · p. 10

6. Software and documentation

• The R Shiny dashboard code is provided in the osinit GitHub repository as app.R.Source: §6 · p. 10
• The repository covers the dashboard application and its analyses.Source: §6 · p. 10
• The repository is not a complete record of manual data-curation decisions.Source: §6 · p. 10

7.1. Data protection and reuse rights

• COSMI Hub was compiled from publicly accessible directories, official websites, institutional documents and policy reports.Source: §7.1 · p. 10
• The dataset describes open-science initiatives and related entities, without collecting data from human research participants.Source: §7.1 · p. 10
• The CC BY 4.0 release does not override third-party rights or data-protection obligations for identifiable professional contacts.Source: §7.1 · p. 10

7.2. Declaration of generative-AI assistance

• ChatGPT assisted with drafting, restructuring and editing selected manuscript and metadata passages; the authors remain responsible for content, citations and accuracy.Source: §7.2 · p. 10

• Future releases could extend dataset coverage and improve documentation of variable definitions, provenance and version changes.Source: §8 · p. 10
• Where feasible, initiative representatives could review their records, with corrections documented separately from genuinely new initiatives.Source: §8 · p. 10
• Researchers, infrastructure providers and initiative representatives are invited to use the archived dataset, cite its version-specific DOI and suggest changes, which would be checked against existing documentary criteria.Source: §8 · pp. 10–11

• ID, MAS and AM share responsibility for conceptualization, investigation, methodology, supervision, validation and both stages of writing.Source: Author contributions (CRediT) · p. 12
• ID and AM handled data curation and project administration, while MAS and AM conducted formal analysis and acquired funding.Source: Author contributions (CRediT) · p. 12
• AM contributed resources, software and visualization; ID, MAS and AM also shared responsibility for investigation and validation.Source: Author contributions (CRediT) · p. 12

• The supplementary materials provide project background and methodology and validation notes.Source: Supplementary project documentation · p. 12
• A classification codebook documents the thematic categories.Source: Supplementary project documentation · p. 12
• The FAQ provides guidance for using the dashboard.Source: Supplementary project documentation · p. 12