Open science makes research findings, methods and appropriate scientific outputs easier to find, evaluate, reuse and build upon. It involves publications, datasets, software, protocols, teaching resources and participation—but it does not remove duties to protect confidential, sensitive or restricted information. Its strongest achievement is evidence that another person can meaningfully inspect and challenge.
This guide follows 25 chapters and 21 worked decisions through FAIR principles, data sharing, code, preregistration, reproducibility, licensing, ethics and research infrastructure. It connects the relevant eduKate Gold Standard roots without replacing their specialist treatment.
Choose a reading route
Understanding openness, rights and FAIR principles · Chapters 1–7 · Begin here →
Data, methods, code and publications · Chapters 8–14 · Begin here →
Communities, infrastructure and protection · Chapters 15–21 · Begin here →
Governance, evaluation and next steps · Chapters 22–25 · Begin here →
Full chapter index · Practice investigations · Stewardship problems · Connected specialists · Primary sources
Complete chapter contents
Understanding openness, rights and FAIR principles
- 1. What open science makes possible
- 2. Open access, open data and open research are not synonyms
- 3. Open science is not permission to publish everything
- 4. The FAIR principles: four distinct tests
- 5. Persistent identifiers and discoverability
- 6. Metadata is how evidence survives its creators
- 7. Open methods and preregistered plans
Data, methods, code and publications
- 8. Raw observations, cleaned datasets and the transformation record
- 9. Open source code and computational environments
- 10. Reproducibility and replication test different things
- 11. Licences, reuse permissions and attribution
- 12. Data repositories and long-term stewardship
- 13. Open peer review and editorial transparency
- 14. Preprints and the boundary between dissemination and validation
Communities, infrastructure and protection
- 15. Negative results, corrections and a recoverable record
- 16. FAIR machine actionability and interoperable vocabularies
- 17. Open educational resources and accessible explanation
- 18. Citizen science and community participation
- 19. Indigenous and local knowledge and fair collaboration
- 20. Equity, access fees and reward systems
- 21. Security, dual use and proportional restrictions
Governance, evaluation and next steps
- 22. Open science metrics that measure useful practice
- 23. Institutional policies and funder requirements
- 24. Design an open workflow from the first week
- 25. Questions readers frequently ask
An open object needs more than a download link
| Research object | What makes it useful | What may need protection |
|---|---|---|
| Publication | Readable claim, cited methods and version | Third-party rights and correction status |
| Dataset | Identifier, codebook, provenance, units and access path | Participants, location and sensitive details |
| Code | Version, dependencies, instructions and tests | Secrets, restricted inputs and unsafe exposure |
| Protocol | Original plan, changes and limits | Confidential or sensitive operational details |
| Teaching resource | Accessible explanations and permitted adaptation | Copyright and misleading examples |
1. What open science makes possible
Open science seeks to make research knowledge, methods and useful outputs more accessible, verifiable, reusable and inclusive. Its aim is not simply free PDFs, a public spreadsheet or a repository icon beside a paper. A strong open-science practice enables another qualified person to understand the question, examine the evidence, evaluate the method, reproduce an appropriate analysis and build responsibly on the result.
The UNESCO Recommendation on Open Science, adopted in 2021, addresses open scientific knowledge, infrastructure, engagement and dialogue across knowledge systems. Its broad vision still requires specific decisions about intellectual property, privacy, safety and equitable access. The first practical question is which research object can safely be made accessible, to whom, with what documentation and under what conditions.
2. Open access, open data and open research are not synonyms
Open access usually concerns the ability to read scholarly publications without a subscription barrier. Open data concerns availability and reuse of appropriate datasets with usable documentation. Open research practices extend to protocols, analysis decisions, code, peer review, educational materials, hardware and community engagement. Each can improve transparency in a different way.
A free-to-read article with inaccessible data may communicate findings while limiting verification. A downloadable dataset without variable definitions may technically be open but unusable. An uploaded code file without dependencies may be impossible to run. Open science asks what another person actually needs to check or extend the work.
3. Open science is not permission to publish everything
Human participants may have consented to a defined use of their information, not unlimited public sharing. Field research can involve endangered species locations, sensitive cultural knowledge or security-relevant infrastructure. Privacy, ethics, agreements and law can constrain which objects may be released and how. Openness must be bounded by responsibilities to affected people and communities.
Choose between public data, controlled access, aggregate summaries, synthetic teaching examples or restricted archival storage based on the actual risk. Explain withheld information through appropriate metadata when safe. The aim is maximum legitimate transparency and reuse, not a reckless all-or-nothing rule.
4. The FAIR principles: four distinct tests
FAIR stands for Findable, Accessible, Interoperable and Reusable. Findability relies on identifiers, metadata and indexing. Accessibility concerns retrieval through well-defined arrangements, which can include authorisation. Interoperability requires compatible formats, structures and vocabularies. Reusability requires information about provenance, licences, conditions and the meaning of variables.
FAIR is not the same as open to everyone. The GO FAIR documentation explicitly permits authentication and authorisation where necessary. A sensitive dataset can support responsible FAIR practice through discoverable metadata and clear controlled-access procedures. Treating FAIR as a licence to bypass confidentiality misunderstands the principle.
5. Persistent identifiers and discoverability
A stable identifier lets an article, dataset, software release or other research object be cited and retrieved consistently. DOIs and repository identifiers are commonly used, but an identifier alone does not guarantee the underlying object is complete, current or appropriate for reuse. A persistent URL to a blank file is not a meaningful research contribution.
Provide a human-readable title, authors or stewards, date, version, subject, relevant methods and a link to related publications. Check that the identifier resolves to a landing page describing the object. If the data later become restricted or withdrawn, appropriate metadata can preserve a truthful record without exposing protected content.
6. Metadata is how evidence survives its creators
Data columns need definitions, units, instruments, sample context, codebooks and clear descriptions of missing or excluded observations. Without them, a value of 25 may be millimetres, degrees or a questionnaire code, and a blank cell may mean not applicable rather than a failed measurement. The files can be perfectly preserved while their meaning is lost.
Write metadata while research is being collected. Record coordinate reference systems, category definitions, dates, calibration and provenance where appropriate. Documentation built after the original team disperses may require unsupported guesses. Good metadata is a scientific asset, not clerical decoration.
7. Open methods and preregistered plans
Pre-registration records planned questions, outcomes or analysis decisions before relevant data are examined, helping distinguish planned confirmation from later exploration. In some fields a registered report can receive methodological review before results are known. These practices can reduce selective reporting when they fit the research design.
Pre-registration does not guarantee that the plan is scientifically sound, ethically permissible or executed correctly. Research often yields valuable unforeseen observations. Label deviations and exploratory analyses honestly rather than hiding them. The standard is traceability of the decision path, not suppression of creativity.
8. Raw observations, cleaned datasets and the transformation record
Research frequently converts instrument files, survey responses or archival materials into analysis-ready tables. Cleaning may correct documented errors, standardise units, merge records or exclude unusable measurements. Every such operation can alter the evidence and therefore deserves a reproducible account.
Retain raw data securely when permitted, keep transformation scripts and record reasons for exclusions. Make clear which processed file a published figure used. Simply distributing the final cleaned spreadsheet may be insufficient if readers cannot determine what was changed between collection and result.
9. Open source code and computational environments
Analysis code helps other people verify calculations, but a code repository alone may not reproduce a result. Software versions, packages, external services, random seeds, hardware behaviour and file paths can affect execution. A successful workflow documents an environment and supplies a clear route from input objects to reported outputs.
Where appropriate, provide a readme, installation instructions, test data, versioned release and computational checks. Do not upload private credentials, participant identifiers or licensed proprietary inputs merely to make the repository look complete. Open code should be both usable and safe.
10. Reproducibility and replication test different things
Computational reproducibility asks whether a finding can be obtained using the original data and methods under documented conditions. Replication tests an inference with new data or an independently repeated study. The National Academies distinguish these concepts because matching an old calculation and observing a similar outcome in a fresh sample are different kinds of evidence.
A released script can help computational reproducibility without proving the original study’s causal claim. Conversely, a well-designed independent replication may provide valuable evidence when confidential original data cannot be published. State which test was performed rather than claiming that a repository makes the science ‘replicated’.
11. Licences, reuse permissions and attribution
An open file does not automatically carry a clear legal permission for every use. Creative work, scientific data, code and institutional records can have different copyright, licensing and contractual conditions. Reusers need to know whether copying, adaptation, distribution or commercial use are allowed, and how attribution should be given.
Choose and record licences that match the nature of the object and rights actually held. Do not attach a permissive licence to third-party content for which the author lacks permission. If material is restricted, describe the permitted route and conditions. Responsible openness requires legal clarity as well as technical access.
12. Data repositories and long-term stewardship
Repositories provide infrastructure for storing and describing research objects and, where supported, assigning identifiers, preserving versions and managing access. A public file-sharing link may work for a week but is not a durable preservation plan. Repository suitability varies across disciplines, data sensitivity, formats and institutional requirements.
Evaluate who operates the repository, permitted file types, metadata support, access controls, preservation commitments and withdrawal procedures. Link data to publications and code versions. For sensitive human research, controlled access may be necessary even when the surrounding scientific description is openly discoverable.
13. Open peer review and editorial transparency
Peer review can use different openness arrangements, including publishing reviewer reports or reviewer identities under suitable policies. Transparency may help readers see how a claim was challenged and revised, but it can also create risks for early-career reviewers, participants or confidential intellectual work if applied indiscriminately.
Disclose the journal’s actual review model instead of using ‘open peer review’ as an unspecified quality badge. Reviewer identities, reports, editorial independence and post-publication discussion are separate decisions. Open commentary does not ensure that a paper is correct, just as closed review does not necessarily make it flawed.
14. Preprints and the boundary between dissemination and validation
Preprints can share results before conventional journal review, accelerating scrutiny and collaboration. Their status must be explicit, particularly when findings could influence health, safety or public policy. A preprint, accepted manuscript, version of record and corrected or retracted publication may differ materially.
Check version numbers, date, review status and linked corrections before citing. A working DOI or widespread social-media discussion is not independent validation. Readers should know whether the claim has been peer reviewed, replicated or remains a provisional research report.
15. Negative results, corrections and a recoverable record
Unsuccessful experiments and null results can prevent repetition of ineffective work and improve understanding of model limits. Sharing such findings requires adequate methods and uncertainty, not the declaration that a treatment ‘does nothing’ because one small test was inconclusive. Similarly, correction notices can preserve trust when errors are found.
Record primary outcomes, deviations and limitations. Make updated code, data and textual interpretations version-aware. A scientific record becomes more credible when important changes remain traceable instead of being silently overwritten. The Research Ethics root covers the people and integrity responsibilities that continue throughout publication.
16. FAIR machine actionability and interoperable vocabularies
Computational systems increasingly need to find and combine research objects without manually asking each author what a field means. Shared identifiers, machine-readable metadata and domain-standard vocabularies can make scientific resources interoperable. However, machines still need governance, semantic clarity and quality checks.
Two datasets both using a field called ‘age’ may refer to a person’s age, the age of a specimen or time since installation. Agreement in file format does not make their meanings equivalent. Document definitions, transformations and controlled vocabularies where appropriate. The goal is trustworthy reuse, not merely the successful import of rows.
17. Open educational resources and accessible explanation
Teaching materials, diagrams, problem sets and explanatory code can support learning beyond the original university or laboratory. Reuse becomes easier when sources specify permitted adaptation, prerequisites, accessible alternatives and any discipline-specific limits. An open textbook is not automatically accessible to every learner if its file format or language makes use difficult.
Provide captions, clear reading routes and explanations in usable forms where feasible. Make the scientific limitations of simulations or teaching datasets visible. Separate illustrative activity results from measured research findings. The educational value of openness is realised through comprehension and responsible use.
18. Citizen science and community participation
Citizen science projects can invite public contributions to observations, classifications or data interpretation. Participants may supply local knowledge and time that would otherwise be unavailable. Ethical project design explains what contributors are asked to do, how data quality is checked and who can access the results.
Give clear feedback, credit where appropriate and a route to correct erroneous records. Avoid implying that volunteers’ local observations are automatically representative of a whole region. Participation should be safe, meaningful and transparent about how contributions are used.
19. Indigenous and local knowledge and fair collaboration
Research can involve knowledge held within communities under collective or culturally specific norms. A broad data-release principle should not override legitimate responsibilities concerning origin, consent, representation, attribution or benefit sharing. The UNESCO framework recognises dialogue with diverse knowledge systems as part of open science.
Discuss governance with relevant holders before collection and reuse. A public researcher should not claim authority to release knowledge simply because it was recorded once. Respecting boundaries can enable stronger, more enduring collaboration than an extractive approach marketed as openness.
20. Equity, access fees and reward systems
Removing a subscription barrier from a scholarly article does not eliminate every cost. Publication charges, language requirements, computing capacity, internet access and institutional infrastructure can still exclude researchers and readers. Open science systems need to consider who can publish, maintain infrastructure and benefit from shared resources.
A research evaluation that rewards only prestigious article counts may discourage sharing code or carefully curated datasets. Recognise diverse contributions and the work of data stewards, educators and community partners. Equitable openness is an ongoing governance question, not a solved technical setting.
21. Security, dual use and proportional restrictions
Some research materials could expose vulnerable infrastructure, sensitive personal information or hazardous capabilities if distributed without safeguards. A responsible openness plan evaluates these risks rather than assuming that public access is always harmless. Restrictions should be justified and proportionate, with sufficient non-sensitive explanation for scholarly accountability.
Describe which information is withheld and why at an appropriate level. Consider safe summaries, controlled verification or independent access processes. The choice should reflect actual risks and obligations, not a convenient excuse to conceal inconvenient findings.
22. Open science metrics that measure useful practice
Counting repository links or open-access publications is an incomplete measure of openness. A dataset may be discoverable but unusable; a code archive may be available yet fail every reproduction attempt. Conversely, a sensitive repository with controlled access may offer excellent metadata and legitimate verification routes.
Evaluate documentation completeness, link persistence, accessible formats, successful independent reuse, correction responsiveness and inclusion of affected communities. Avoid combining all of these into a single opaque ‘openness score’. Show readers what works, what remains blocked and why.
23. Institutional policies and funder requirements
Funders and journals may specify data-management plans, deposit timing, availability statements, transparency standards or justified limitations. The NIH Data Management and Sharing Policy applies to research within its defined scope and expects approved plans to be implemented. The Transparency and Openness Promotion framework offers research and verification practices that journals may adopt.
Do not treat one organisation’s requirements as universal global law. Check the current grant, institutional, disciplinary and jurisdictional conditions. A project may also involve contractual or privacy restrictions that require documented exceptions. Early planning prevents an impossible promise at publication time.
24. Design an open workflow from the first week
At project initiation, name the research question, anticipated outputs, source data, ethical limits, metadata fields, code dependencies, licensing and publication route. Agree who owns each task and when information will be released or restricted. This is easier than attempting to reconstruct a lost workflow after a team disperses.
During analysis, keep versioned scripts and a traceable link from data to figures. At reporting, give readers the approved access path to each object and record limitations and changes. At archiving, plan what happens if a link expires or a participant’s information must be withdrawn. Openness becomes a designed feature of research, not a final upload ceremony.
25. Questions readers frequently ask
Is FAIR the same as public access? No; well-described controlled access can still meet relevant FAIR principles. Does making a dataset downloadable prove reproducibility? No; documentation, methods and software may also be needed. Does a preprint count as peer-reviewed science? Not necessarily. Does a licence remove consent and privacy obligations? No.
Do negative findings need to be shared? They can be important when responsibly reported. Is open code automatically safe to publish? No; inspect credentials, sensitive data and rights. Can a small classroom project practise open science? Yes, through original synthetic examples, clear methods, reproducible calculations and honest source credit.
← Previous · Practice investigations
Twenty-one decisions in responsible open research
Each case below is an illustrative problem for learning, not an allegation about a named researcher, journal or organisation.
1. The discoverable dataset behind an access gate
Invented clinical registry: raw participant data cannot be released publicly, but the steward provides an indexed description with purpose, variables, collection period, access conditions and a contact route. A qualified reviewer can request approval for a justified analysis.
The resources are not indiscriminately open, yet useful metadata and governed retrieval can make reuse possible while respecting privacy. The proper question is whether the access process is transparent and responsible, not whether anyone can click ‘download’.
2. A FAIR dataset without defined units
A public CSV contains values under fields labelled ‘time’, ‘rate’ and ‘area’. It has an identifier and a licence but provides no information on units or sample selection. A researcher can download it but cannot confidently combine it with their own data.
The repair is to provide a codebook, unit descriptions, measurement protocol, date interpretation and missing-data rules. Findability is only one part of useful interoperability and reuse.
3. A repository that contains a result but not its calculation
An article reports a mean change of 12 units, and the related repository contains a final results table. The raw observations and exclusion decisions are undocumented. An independent reader cannot tell how the published figure was produced.
Where ethical and lawful, preserve source data, versioned transformation steps and output checks. If underlying data must remain confidential, describe an approved route for independent verification or clarify the exact limit. A final table is not automatically a complete reproducibility package.
4. Preregistered outcome and exploratory discovery
A study preregisters a main comparison, but a surprising subgroup difference appears after analysis. The team wants to present that subgroup as the original research question because it sounds more interesting.
The ethical solution is to report the preregistered outcome honestly and label the new finding exploratory, then propose independent confirmation. Open science protects discovery by preserving the distinction between planning and hindsight, not by forbidding new questions.
5. A software release with an undocumented environment
An analyst publishes a script that works on their laptop but fails elsewhere because a library changed behaviour. A second researcher cannot recover the figure although the code text is available.
Record a versioned environment, dependencies, sample input and expected output. A test fixture or container can help where appropriate. Avoid packaging secrets or sensitive input files simply to make execution convenient.
6. A private dataset accidentally exposed through a log file
A team removes personal data from the public CSV yet includes debugging logs with record identifiers and raw free-text responses. The package looks de-identified only if one ignores the extra files.
Audit the entire release bundle, including notebooks, temporary outputs, plots, history and metadata. Restrict or remove sensitive artifacts while preserving an honest explanation of what was done. Data security is part of research reproducibility.
7. An old DOI that resolves to a superseded version
A reader downloads a dataset with a stable identifier, not noticing a newer release corrected calibration units. Their calculation disagrees with the associated article even though the files are valid versions.
Compare publication date, version, provenance and correction notes. Cite the exact version used. Stable identification is useful when it preserves revision history rather than concealing the fact that a result has changed.
8. A broken repository link after a staff departure
A research team’s website hosted analysis files in a former employee’s personal folder. Years later the link fails and no one knows which version produced a cited graph.
Use a suitable long-lived repository or institutional stewardship process. Store the project inventory, licence and version identifiers with the publication. A research object that cannot be found or interpreted is less reusable regardless of its former publicity.
9. Two categories that only look interoperable
Two surveys both use a field called ‘satisfied’, but one asks about a service this week and the other asks about overall experience over a year. Merging the percentages silently creates a misleading result.
Review questions, reference periods, response scales and populations. Interoperability depends on compatible meaning, not merely matching column labels. Record transformations or refrain from combining values when equivalence is unsupported.
10. An open figure with an invisible image transformation
A microscopy figure has been contrast-adjusted and cropped, but the caption does not state the processing. Another laboratory sees different pixel distributions from the raw image and cannot interpret the apparent discrepancy.
Document original files and image-processing operations, preserve provenance and distinguish permitted display enhancement from changes that would misrepresent observations. Open files without a truthful transformation record do not guarantee reliable visual science.
11. A citizen-science dataset with duplicated submissions
Volunteer observers report an unusual environmental event. Several upload the same photo or re-enter one record, causing an exaggerated apparent event count.
Keep appropriate deduplication and provenance controls, report uncertainty and explain corrections. The eduKate duplicate-observation guide helps readers see why sharing data is not the same as validating counts.
12. A retracted paper in an automated literature search
A search workflow retrieves a paper with a working DOI but misses an associated retraction notice. The summary includes a conclusion that the corrected literature no longer supports.
Record publication status and linked notices during review, not only the DOI. Re-evaluate the evidence if the source is withdrawn or materially corrected. The public record is strongest when corrections remain machine- and human-discoverable.
13. Software licensed openly but dependent on restricted input
A research package includes a permissively licensed script but requires an inaccessible commercial or confidential database. Other users can read the code but cannot reproduce the reported output.
Disclose input dependencies and rights honestly. Where possible, include safe synthetic test data for exercising the workflow and an approved verification route for the real analysis. Do not pretend synthetic tests confirm the real-world finding.
14. The classroom project with a fully reproducible synthetic dataset
Students simulate arrivals at a help desk using explicitly invented random rates. They publish a readme, model assumptions, code, random seed, generated data and calculations. Others can reproduce the simulation without exposing anyone’s private records.
The project demonstrates authentic open-science practice while remaining a teaching exercise. The caption must not describe the simulated arrivals as measurements from an actual school office.
15. A costly open-access route and an excluded author
A multi-institutional team can read open publications, but a contributor lacks funds for publication charges. A publication pathway that shifts all costs from readers to authors may leave some collaborators disadvantaged.
Discuss viable journal, repository and rights-retention options before choosing a venue. A fair collaboration takes account of publication access and the labour behind datasets, methods and community engagement.
16. Open code containing a secret credential
A notebook is ready to publish except that its history includes an API credential. Removing the visible current cell does not necessarily remove the earlier secret from version history.
Rotate or revoke exposed credentials, inspect full history and build a safe release without secrets. Document how users provide their own authorised configuration. Transparency should not create unauthorised access.
17. A confidential interview with a reusable analysis protocol
Researchers cannot disclose full interview recordings because participants could be identified. They can still explain sampling, interview questions where appropriate, coding decisions, limitations and aggregate analytic steps.
Readers should know which details cannot be independently checked and why. Where suitable and approved, independent auditors can inspect restricted material. Responsible openness is richer than a binary open-or-closed label.
18. A changed-context test in an Indigenous knowledge project
A scientist prepares to upload maps created through collaboration with a community. A normal repository checklist asks for public identifiers and open licensing, but parts of the map reflect culturally sensitive knowledge.
Stop and revisit governance commitments, permissions and knowledge holders’ expectations. Publish safe descriptions or controlled outputs as appropriate. The same commitment to accountable research remains, but the access conditions must follow context.
19. Reproducing an estimate is not replicating a finding
A second analyst runs the original code with the original data and obtains precisely the published estimate. That is evidence for computational reproducibility. Another team gathers new measurements and obtains a broadly compatible result under a documented new study.
These are different checks. The first confirms a calculation path; the second tests how a conclusion travels beyond its original sample. Keep the terms distinct in the public explanation.
20. An open-data licence that does not settle participant consent
A dataset is released under a permissive reuse licence, but source participants were promised that interview quotations would remain confidential. Intellectual-property permission and confidentiality duties address different questions.
Re-examine the original agreement and disclosure risks. Remove or restrict sensitive content where required and correct the access record. A licence cannot grant ethical permission that the data steward did not possess.
21. Measuring whether an open science effort worked
A laboratory reports uploading ten datasets. But seven have broken links, two lack units and one contains sensitive identifiers. Counting uploads gives a positive metric that conceals severe usability and governance problems.
Evaluate persistence, metadata, access safeguards, successful reuse, independent verification and correction. A smaller set of carefully stewarded research objects may serve science better than an impressive but unusable file count.
Eight advanced open-science stewardship problems
The final set of invented scenarios tests whether a reader can apply openness, preservation, rights and research protection together under changed conditions.
1. A metadata record after responsible withdrawal
Suppose a deposited dataset is withdrawn after its steward discovers that the combination of columns could identify participants. Erasing every trace of the deposit would leave readers wondering why published analyses no longer have supporting files. Keeping the raw files available would preserve the hazard.
A responsible repository can maintain an appropriate landing page stating that the dataset has been withdrawn, with identifier, version, notice and a safe reason where disclosure permits. This illustrates why stable metadata can remain accessible without making sensitive contents publicly downloadable.
2. Open access does not settle copyright in a figure
A research author publishes a freely readable article containing a photograph made by a third party. Another educator assumes that because the article is open access, the photograph may be modified and republished in any context without checking rights.
The correct decision depends on the applicable licence, credited creator, exclusions and intended use. Separate access to the article from permission to reuse embedded third-party material. The safest open educational resource tells reusers which components they may adapt and under what conditions.
3. Two code archives that produce different random outcomes
An algorithm uses stochastic sampling, but its released code provides no seed and only vague dependencies. A second analyst gets a different final number from the same apparent input. That difference may reflect an expected random realisation, implementation details or a genuine bug.
Publish the relevant software versions, random seed and statistical tolerance where appropriate. Reproducibility need not demand an identical random sequence in every legitimate stochastic workflow; it does require clarity about which numerical or statistical properties should agree.
4. A data dictionary with ambiguous missing values
An education dataset uses zero, blank and −99 in the same field. One analyst interprets all three as absence, while another treats zero as a valid outcome and −99 as not observed. Their summary statistics diverge.
A useful data dictionary distinguishes valid zero, unknown, inapplicable and withheld. Record the treatment of each before combining datasets or sharing an analysis-ready table. Open files without this semantic information can encourage confidently incorrect reuse.
5. An early preprint that changes after review
A preprint reports a promising effect. Later peer review reveals a statistical assumption that needs correction, leading to a revised paper with a more cautious conclusion. A summary system continues citing the first version without a visible date or warning.
Link versions and report the current status. Preserve the provenance of corrections. The point of early sharing is to invite scrutiny, not to confer permanent authority on the first publicly available claim.
6. A reproduction package that requires impossible hardware
A complex analysis depends on specialised computing resources that are not available to every independent researcher. It might still be possible to provide methods, lighter validation tests, intermediate outputs and clearly documented constraints without pretending anyone can rerun the full workload on a laptop.
Describe hardware, computation cost and acceptable alternative checks. Honest accessibility includes the practical resources required, not merely the fact that source code appears in a repository.
7. A volunteer observation corrected after public release
Two students discover that an uploaded citizen-science observation used the wrong species label. Other researchers have already downloaded the dataset and may use the mistaken value in a published map.
Preserve correction history, update the identified record and communicate a versioned change through appropriate channels. Where possible, provide downstream users with a way to identify affected exports. Transparent correction protects the scientific record better than silently rewriting the original upload.
8. Open methods when raw personal data cannot travel
A university team has a lawful, secure workspace for confidential survey records. Collaborators elsewhere cannot download those records, but they can review analysis code and a synthetic test file that uses the same schema.
Combine governed access, disclosure-reviewed summaries and reproducible code where feasible. State what synthetic tests do and do not verify. The changed-context lesson is that strong openness adapts to participant protection instead of sacrificing it.
Connected eduKate reading library
Scientific foundations and integrity
- Research Ethics
- Scientific Method
- Research
- Scientific Computing
- Science Communication
- School Research Data Integrity
Education, community and open infrastructure
- Citizen Science
- School Citizen Science Data Contributions
- Global Connectivity | Academic Publishing
- Teaching Library and Information Science
- Scientific Literacy
Critical cases
- Hidden Identifiers in Open Research Data
- Retracted Research Cited as Current
- School Citizen Science Duplicate Control
Primary research, governance and technical standards
- UNESCO — Open Science and the 2021 Recommendation
- UNESCO — Recommendation on Open Science (official text)
- GO FAIR — The FAIR Guiding Principles
- GO FAIR — How to Implement FAIR
- Center for Open Science — TOP Guidelines
- NIH — 2023 Data Management and Sharing Policy
- National Academies — Reproducibility and Replicability in Science
- COPE — Retraction Guidelines (2025)
A practical research sharing checklist
- Define which object is being shared and what readers need to verify.
- Identify rights, participant expectations, access restrictions and any relevant security risks.
- Provide stable identifiers, documentation, variables, units, versions and dependencies.
- Describe transformations, excluded results and changes without hiding inconvenient findings.
- Test the release as an independent user and establish a correction or withdrawal route.
Return to The Gold Standard, Education and Learning Transfers into Life, or the How X Works master hub.