RESEARCH INTEGRITY · PUBLICATION ETHICS · AUTHORSHIP · DATA INTEGRITY · MISCONDUCT · CORRECTIONS · TRUST

How Research Integrity and Publication Ethics Work

Research integrity is the system of values, records, responsibilities and controls that makes scholarly claims trustworthy enough to inspect, challenge, reproduce and correct. Publication ethics is the part of that system that governs how research enters the scholarly record, who receives credit, what must be disclosed, how allegations are handled, and how the record is repaired when something goes wrong.

Integrity is not the absence of error. It is the discipline that keeps error, uncertainty, responsibility and correction visible.

Research can fail honestly or dishonestly. A measurement can be wrong because an instrument malfunctioned. An analysis can be mistaken because an assumption was misunderstood. A figure can be altered deceptively. A contributor can be omitted from authorship. A conflict can remain undisclosed. A paper can duplicate earlier work without acknowledgement. These cases do not belong in one moral bucket, but all require systems capable of separating evidence, intent, responsibility and consequence.

Modern publication ethics therefore extends across the entire lifecycle: research design, approvals, data collection, records, analysis, authorship, peer review, submission, publication, post-publication discussion, correction, retraction and archival retention.

The short answer

RESEARCH QUESTION
  → ETHICS / GOVERNANCE
  → RESPONSIBLE METHODS
  → ACCURATE RECORDS
  → DATA / IMAGE INTEGRITY
  → ANALYSIS
  → AUTHORSHIP / CONTRIBUTORSHIP
  → CONFLICT DISCLOSURE
  → MANUSCRIPT
  → JOURNAL / PEER REVIEW
  → PUBLICATION
  → COMMUNITY SCRUTINY
  → CONCERN OR ERROR?
  → VERIFY
  → CORRECT | EXPRESS CONCERN | RETRACT | EXONERATE
  → PRESERVE AUDIT TRAIL
  → SCHOLARLY RECORD CONTINUES

1. Integrity begins before publication

A journal cannot manufacture research integrity after the study is complete. Good publication depends on good research conduct before submission: appropriate approvals, competent methods, accurate records, secure data, reproducible analysis and honest interpretation.

Publication ethics is therefore downstream from research integrity while also feeding back into it through standards, disclosure and correction.

2. Honest error and misconduct are different states

Researchers can make genuine mistakes without acting dishonestly. A statistical code error, mislabelled sample or mistaken interpretation can require correction even when no misconduct occurred.

Conversely, deliberate fabrication, falsification or plagiarism can constitute misconduct because the problem lies not only in the output but in the conduct that produced it.

3. Fabrication invents evidence

Fabrication means creating data, observations or results that did not exist and representing them as real.

This attacks the evidentiary foundation directly. No amount of elegant writing or peer review can rescue a claim built on invented observations.

4. Falsification distorts evidence

Falsification can include manipulating research materials, processes, images, data or analysis so that the reported record no longer reflects what actually occurred.

Not every transformation is falsification. Cleaning, normalisation and image adjustment can be legitimate when methodologically justified, applied consistently and reported transparently.

5. Plagiarism misrepresents intellectual origin

Plagiarism presents another person’s words, ideas, data, figures or other intellectual material without appropriate attribution.

The central integrity problem is provenance: the reader is given a false account of where the material came from.

6. Self-plagiarism and redundant publication create another provenance problem

Reusing substantial parts of one’s own previously published work without disclosure can mislead readers about novelty and duplicate evidence in the literature.

The problem is not that authors are forbidden to build on earlier work. It is that previous publication states and overlap should be disclosed and cited honestly.

7. Authorship carries credit and accountability

Authorship is valuable because it assigns credit. It is demanding because it also assigns responsibility.

The current ICMJE Recommendations, updated in January 2026, continue to frame authorship around substantial contribution, participation in drafting or critical intellectual revision, approval of the final work and accountability for its integrity.

8. Contribution should be visible even when authorship is shared

Modern journals increasingly publish contributorship statements describing who designed the study, collected data, analysed results, wrote software, supervised work, acquired funding or drafted the manuscript.

Contribution transparency reduces the ambiguity hidden inside a long author list.

9. Gift authorship weakens accountability

Adding a senior person, supervisor or influential colleague who made little or no qualifying contribution can distort credit and accountability.

Singapore’s NTU research-integrity guidance explicitly warns that gift authorship can be viewed as possible research misconduct and states that supervisors or chairs who made no intellectual input should not be co-authors merely because of position.

10. Ghost authorship hides contribution

Ghost authorship occurs when substantial contributors are omitted, sometimes because their role would reveal commercial influence or another uncomfortable relationship.

Contributorship statements, acknowledgements and funding disclosures help make hidden labour and influence more visible.

11. Author order is a disciplinary convention, not a universal code

First author, last author and alphabetical ordering carry different meanings across disciplines.

Teams should discuss ordering early and document their agreed convention rather than discovering conflict at submission.

12. Authorship disputes should be resolved by evidence of contribution

Journals can facilitate process, but institutions are often better placed to investigate disputes because they can access research records, employment relationships and contribution evidence.

The goal is not to treat authorship as a negotiation of status but as a documented account of intellectual responsibility.

13. Conflicts of interest do not automatically invalidate research

A financial, professional, personal or institutional relationship can create a potential conflict without proving bias occurred.

Disclosure allows editors and readers to evaluate the relationship rather than forcing them to guess whether one existed.

14. Non-financial conflicts matter too

Strong intellectual commitments, personal rivalries, leadership roles, advocacy positions or institutional interests can influence judgement even without direct payment.

Integrity systems should therefore look beyond money while avoiding the impossible demand that every human perspective be treated as a conflict.

15. Funding disclosure is part of source provenance

Readers should know who funded the research and whether the funder influenced design, data access, analysis, manuscript preparation or publication decisions.

Funding can support excellent independent science. Transparency is what lets readers assess the relationship.

16. Sponsor agreements should not block academic independence

Research contracts that prevent investigators from accessing data or publishing legitimate findings can weaken integrity even if the data themselves are accurately collected.

The January 2026 ICMJE update strengthened guidance around sponsor agreements, data access and academic freedom.

17. Research records are the memory of the study

Laboratory notebooks, raw data, code, consent records, protocols, instrument outputs, analysis scripts and version history can become essential when a finding is challenged years later.

Integrity depends on preserving enough of the research pathway that another qualified person can reconstruct what was done.

18. Singapore research institutions emphasise record retention

The joint Singapore research-integrity statement associated with A*STAR, NTU, NUS and other autonomous universities identifies reproducibility and detailed research records as core responsibilities and specifies a minimum ten-year retention baseline in that statement, while allowing institutions to require longer periods.

Local policies may impose additional or more specific requirements, so researchers should follow the policy governing their institution, grant and research type.

19. Raw data and processed data should be distinguishable

A processed dataset may have excluded outliers, normalised values, derived variables or transformed images.

Good provenance records what happened between raw evidence and published result.

20. Cleaning is not falsification when the rules are legitimate

Removing impossible sensor values, correcting known transcription errors or applying documented quality-control rules can improve data integrity.

The integrity question is whether transformations are justified, reproducible and reported sufficiently—not whether untouched raw data were published directly.

21. Image integrity needs discipline

Scientific images may require cropping, contrast adjustment, stitching or annotation. These changes can be legitimate if they do not misrepresent the underlying evidence and are disclosed where relevant.

Selective enhancement, undisclosed splicing or removal of inconvenient features can cross into falsification.

22. The original image should remain available

NTU research-integrity guidance notes that when image manipulation tools are used, manipulation should be described and raw pre-enhanced data should be made available where appropriate.

The preservation principle is simple: never let the publication derivative become the only surviving evidence.

23. Statistical flexibility can become an integrity risk

Trying many analyses until a desirable result appears, selectively reporting outcomes or redefining hypotheses after seeing results can create misleading confidence even without fabricated data.

Preregistration, registered reports, analysis plans and complete outcome reporting can reduce this risk.

24. Reproducibility begins with method transparency

Another researcher cannot reproduce an analysis if key preprocessing, software versions, exclusion rules or parameters are missing.

Open code and data can help, but reproducibility still depends on clear documentation and lawful access conditions.

25. Data sharing has ethical boundaries

Participant privacy, consent, national-security concerns, commercial obligations and indigenous or community data governance can restrict public release.

Responsible integrity does not require violating legitimate protections in the name of openness.

26. “Available on request” should be a real promise

If a paper states that data are available on reasonable request, there should be a functioning process and a custodian able to respond.

A statement that cannot be honoured becomes misleading metadata about the research.

27. Human-participant research requires consent and oversight

Research involving people can require ethics review, informed consent, privacy safeguards and protocol adherence depending on study type and jurisdiction.

Publication does not erase these obligations; journals often require authors to report ethics approval and consent information explicitly.

28. Animal research requires appropriate oversight

Research involving animals is also governed by institutional and legal standards designed to justify use, minimise unnecessary harm and ensure competent procedures.

Publication ethics includes checking that relevant approvals and reporting are present rather than treating ethics as separate from scientific quality.

29. Clinical-trial registration protects against hidden outcome switching

Prospective registration creates a public record of trial design and intended outcomes before results are known.

This helps readers compare what researchers planned with what they ultimately reported.

30. Selective publication creates system-level bias

If positive findings are published while negative or null findings disappear, the literature can overstate effects even when every published article is individually honest.

Research integrity therefore includes stewardship of the evidence ecosystem, not only avoidance of outright fraud.

31. Duplicate publication can double-count evidence

Publishing substantially the same dataset as though it were independent new evidence can distort systematic reviews and meta-analyses.

Legitimate secondary analyses should disclose dataset reuse and explain what new question is being asked.

32. Salami slicing weakens context

Dividing one coherent study into many minimally distinct papers can fragment the scientific picture and obscure common methods or samples.

Multiple papers can be legitimate when they answer genuinely distinct questions; transparency about shared data is the key.

33. Citation manipulation is a publication-ethics problem

Authors, reviewers or editors can pressure unnecessary citations to inflate metrics for individuals or journals.

COPE’s core practices explicitly treat citation manipulation as a publication-integrity concern.

34. Peer review has its own ethics

Reviewers receive confidential access to unpublished work. They must protect that confidentiality, avoid appropriating ideas, disclose conflicts and provide constructive, honest review.

The reviewer’s ethical obligation is distinct from the author’s but equally part of the publication record.

35. Editors need independence

Owners, sponsors and commercial teams should not control individual editorial decisions for financial, reputational or political reasons.

Editorial freedom protects the distinction between scholarly judgement and business interest.

36. Journals need visible ethics policies

COPE’s current core-practice framework expects journals and publishers to maintain policies on allegations of misconduct, authorship, conflicts, data, ethical oversight, intellectual property, peer review, complaints, corrections and retractions.

Policy is not proof of good behaviour, but absence of policy makes accountable behaviour harder.

37. Misconduct allegations need process, not instant verdicts

An allegation can be correct, mistaken, malicious or incomplete.

Serious concerns require evidence preservation, confidentiality, fair notice, competent investigation and separation between allegation and finding.

38. Whistleblowers need a route

COPE’s core practices explicitly state that journals should have processes for handling allegations, including those raised by whistleblowers.

Systems that punish every good-faith concern create conditions where problems remain hidden until they become larger.

39. Anonymous allegations can still contain valid evidence

The identity of the complainant and the quality of the evidence are separate questions.

Editors and institutions should assess verifiable facts while taking reasonable steps to prevent abuse of process.

40. Journals and institutions have different investigative powers

Journals control the published record. Institutions can often access laboratories, raw data, personnel records and witnesses.

COPE guidance on cooperation between research institutions and journals emphasises that integrity cases often require both bodies to communicate while respecting their distinct roles.

41. Publishers own the correction decision for their publication

An institution may investigate conduct and make disciplinary findings. The journal still decides how the scholarly record should be corrected, clarified, retracted or otherwise updated.

These decisions should be coordinated but not conflated.

42. Correction addresses a bounded error

If a limited part of a paper is wrong while the main findings remain reliable, a correction can preserve the valid work while making the error visible.

Correction is not a lesser form of retraction. It is the appropriate remedy when the problem is bounded.

43. Retraction addresses unreliable publication states

COPE guidance states that publications should generally be retracted when findings are unreliable because of major error or misconduct, or where serious publication problems such as plagiarism or redundant publication undermine the record.

The focus of a retraction notice should be the reliability and status of the work, not sensational speculation about motives.

44. Retraction should not wait forever for cooperation

Authors may disagree with a retraction. If evidence establishes that the publication is seriously unreliable, the journal’s responsibility to readers can require action even without unanimous author agreement.

Procedural fairness matters, but the literature cannot be held hostage indefinitely by refusal to cooperate.

45. Expression of concern is an unresolved status

When serious concerns exist but evidence remains incomplete, a journal may publish an expression of concern to alert readers while investigation continues.

COPE guidance treats an expression of concern as an interim information state—not a milder punishment and not a substitute for reaching a conclusion.

46. Exoneration should also update the record

If an investigation ultimately supports the reliability of the work, the journal should update the public record so an earlier concern does not remain as a permanent cloud.

Integrity requires correcting suspicion as well as error.

47. Retraction should preserve identity

Removing the publication entirely can break citations and obscure what happened.

Where lawful and appropriate, the article remains identifiable, clearly marked as retracted and linked to the notice explaining its status.

48. Corrected and retracted states need machine-readable metadata

Human-readable notices are necessary but insufficient in a literature indexed by databases, repositories and AI systems.

Structured status relationships help downstream systems avoid treating a retracted article as an ordinary current source.

49. Preprints complicate correction propagation

A journal article can be corrected while an older preprint remains widely available.

Repositories and authors should link versions and update status so the distributed scholarly record does not preserve contradictory trust states without explanation.

50. Data repositories need the same integrity links

If a dataset is withdrawn or corrected, papers that depend on it may need review. If a paper is retracted, related datasets should not remain context-free.

Research objects form a graph; integrity events can propagate across the graph.

51. AI is now part of publication ethics

The January 2026 ICMJE Recommendations added a dedicated section on artificial intelligence in publishing. They emphasise validity, confidentiality, transparency and accountability in AI-assisted workflows.

AI can help with writing, editing, coding, translation and review support, but responsibility remains human.

52. AI cannot be accountable as an author

Current ICMJE guidance states that AI tools should not be listed or cited as authors because they cannot take responsibility for accuracy, integrity and originality.

Human authors must review AI-generated material, verify sources and ensure that plagiarism or rights problems have not been introduced.

53. AI use should be disclosed appropriately

ICMJE’s 2026 guidance asks authors to disclose relevant use of AI-assisted technologies and to describe how the tool was used.

Using AI for language assistance, analysis, image generation or data processing can have different methodological implications and should be reported in the appropriate place.

54. Undisclosed AI use can become an integrity issue

ICMJE notes that nondisclosure may require corrective action and in some circumstances may be construed as misconduct.

The underlying principle is provenance: readers and editors should understand which parts of the research and manuscript were produced with automated assistance when that assistance is material.

55. Confidential manuscripts should not be uploaded casually to AI services

Submitted manuscripts are privileged communications. Reviewers or editors who upload them into external AI systems can expose unpublished data and ideas.

Current ICMJE guidance says reviewers must follow journal policy and should not upload confidential manuscripts where confidentiality cannot be assured without appropriate permission.

56. AI-assisted review still belongs to the human reviewer

If journal policy permits AI assistance, the reviewer remains responsible for accuracy, tone, confidentiality and scholarly judgement.

Automation cannot absorb accountability.

57. AI-generated citations are a high-risk failure mode

Language models can produce plausible references that do not exist or do not support the stated claim.

Every consequential citation generated or suggested by AI should be resolved against a real source before publication.

58. AI-generated images need methodological provenance

Synthetic or enhanced images can be useful illustrations while being inappropriate as representations of experimental evidence.

Research figures should distinguish measured evidence from synthetic explanatory graphics clearly.

59. Automation can improve integrity screening

Publishers can use software to detect text overlap, statistical anomalies, image duplication, reference irregularities and reporting omissions.

Screening tools generate signals, not verdicts. False positives and context still require expert review.

60. Similarity detection is not plagiarism detection by itself

Software can identify matching text. It cannot always determine whether overlap is properly quoted, methodologically unavoidable, common terminology or genuinely plagiarised.

Editorial judgement must interpret the similarity evidence.

61. Research integrity requires due process

Public accusations can cause severe professional harm. Institutions and journals should investigate credible concerns rigorously without assuming guilt from allegation alone.

Fair process protects both the integrity of the record and the rights of the people involved.

62. Confidentiality protects investigations

Premature disclosure can compromise evidence, reputation and procedural fairness.

Investigative confidentiality should not become permanent secrecy once a publication’s status needs public correction.

63. Intent and reliability should be separated

A paper can be unreliable because of honest error. A researcher can engage in misconduct that affects only part of a publication.

COPE guidance emphasises that retraction notices should focus on why findings are unreliable and the publication status, while institutional processes determine culpability and discipline.

64. Correction culture should not stigmatise honest error

If researchers fear that admitting a mistake will be treated exactly like fraud, they have an incentive to hide correctable problems.

Healthy research systems distinguish good-faith self-correction from deception while preserving the same commitment to an accurate record.

65. The research institution owns more than compliance

Leadership, mentoring, workload, incentives, data infrastructure and training all shape behaviour.

A culture that rewards spectacular results at any cost can undermine the policies it publishes on integrity.

66. Singapore’s joint research-integrity principles

A joint statement by A*STAR, NTU, NUS, SMU, SUSS and SUTD identifies six publication-integrity principles: Leadership, Honesty, Reproducibility, Citation, Acknowledgement and Reporting.

The statement also explicitly connects Singapore research institutions to the Singapore Statement on Research Integrity and treats research organisations as holding their institutions in trust for the people of Singapore.

67. Leadership is an integrity control

Senior researchers set norms through lab practices, authorship decisions, reactions to mistakes and treatment of junior staff.

Integrity programmes fail when formal training says one thing and everyday supervisory behaviour rewards another.

68. Mentoring should include publication ethics

Junior researchers need explicit guidance on notebooks, authorship, image processing, statistics, conflicts, data sharing, peer review and correction—not only technical research skills.

Publication ethics becomes stronger when it is taught before the first dispute occurs.

69. Incentives can create integrity pressure

Promotion criteria, grant competition, journal prestige, citation metrics and publication counts can create pressure to overstate novelty or divide findings strategically.

Responsible research assessment should value methodological quality, openness, correction and contribution rather than raw publication volume alone.

70. Metrics are signals, not moral scores

Citation counts, journal impact factors, downloads and altmetrics measure different forms of attention and dissemination.

None can establish whether a researcher acted ethically or whether an article is correct.

71. Collaboration increases integrity complexity

Large collaborations cross institutions, disciplines, countries and legal regimes. They need explicit agreements about data custody, authorship, analysis, publication, intellectual property and dispute resolution.

Assumptions that work inside one laboratory can fail badly across a consortium.

72. Cross-boundary research needs shared rules

Collaborators should clarify which ethical approvals, data-protection rules and publication policies govern the project before data are exchanged.

Integrity problems become harder to resolve when responsibility is distributed but never defined.

73. Local researchers should not become invisible data providers

ICMJE’s current authorship guidance highlights the risk of excluding local researchers from authorship when research uses data from lower- and middle-income settings and notes that local authors can add fairness, context and interpretive value.

The broader principle applies globally: researchers closest to the setting should receive appropriate credit and intellectual participation when their contributions justify it.

74. Research integrity includes public communication

Press releases, interviews, conference talks and social posts can overstate evidence even when the paper itself is careful.

Researchers and institutions should preserve the same claim discipline when communicating beyond the journal.

75. The headline can become an integrity problem

A cautious observational result can be converted into causal certainty by a headline or press release.

Publication ethics extends to derivative communication when those derivatives are controlled by the publisher or institution.

76. Journals should preserve peer-review history

Submission files, reviewer reports, author responses and editorial decisions can become important if questions emerge later.

Retention policies should balance confidentiality with the need to reconstruct the publication process when integrity concerns arise.

77. Research objects need stable identifiers

DOIs, ORCID iDs, repository identifiers, trial registrations and dataset identifiers create traceable relationships among papers, people, data and versions.

Integrity becomes more machine-readable when provenance is structured rather than hidden in prose.

78. Corrections should propagate through the research graph

A corrected paper can affect a dataset, preprint, protocol, review article, institutional press release and later guideline.

Integrity systems increasingly need dependency tracking so downstream users learn that the evidence state changed.

79. AI makes correction propagation more urgent

Models can ingest and reproduce claims long after an article has been corrected or retracted.

Machine-readable status, current metadata and canonical links help retrieval systems prefer the corrected scholarly state.

80. Research integrity is an evidence architecture

CLAIM
  → DATA
  → METHOD
  → ANALYSIS
  → PEOPLE RESPONSIBLE
  → DISCLOSURES
  → VERSION
  → PEER REVIEW
  → PUBLICATION
  → CORRECTION HISTORY
  → ARCHIVE

The stronger these links are, the easier it is to answer the most important integrity question: Why should this claim be trusted, and what would change our mind?

81. A practical researcher checklist

Obtain required ethics and governance approvals.
Keep accurate research records.
Preserve raw data and transformation history.
Document analysis code and software versions.
Discuss authorship and contribution early.
Disclose conflicts and funding honestly.
Cite previous and third-party work accurately.
Report outcomes and methods without selective distortion.
Follow institutional rules for data and image handling.
Disclose relevant AI use.
Correct material errors promptly.
Cooperate with legitimate integrity investigations.

82. A practical journal checklist

Publish clear authorship and contributorship rules.
Require conflict and funding disclosures.
Protect peer-review confidentiality.
Maintain policies for allegations and whistleblowers.
Use screening tools as signals, not verdicts.
Define data and image-integrity expectations.
Maintain AI-use policies.
Keep editors independent from commercial pressure.
Correct bounded errors visibly.
Use expressions of concern when evidence remains unresolved.
Retract unreliable work when necessary.
Preserve status links and editorial audit trails.

83. A practical institution checklist

Teach responsible conduct before problems occur.
Define data custody and retention.
Provide confidential reporting channels.
Separate allegation screening from final findings.
Protect due process.
Preserve evidence during investigation.
Coordinate with journals when publications are affected.
Address incentives that reward quantity over quality.
Train supervisors in authorship and mentoring.
Keep policies current as AI and research methods change.

84. Failure modes

Failure	What breaks
Integrity begins at submission	Problems created during research become impossible to repair editorially.
Authorship = status	Credit and accountability detach from contribution.
Conflict = proof of bias	Disclosure becomes accusation rather than transparency.
Processed data replace raw records	Transformation history disappears.
Similarity score = plagiarism verdict	Context and legitimate overlap are ignored.
Allegation = guilt	Due process collapses.
Retraction = deletion	Scholarly history becomes untraceable.
Honest correction treated like fraud	Researchers gain incentives to hide mistakes.
AI listed as accountable author	Responsibility becomes fictional.
Confidential manuscript uploaded to unapproved AI	Unpublished research can be exposed.
Retraction status not propagated	Old claims continue circulating as current evidence.

85. The deeper model: integrity is non-compensatory

Some publication qualities can compensate for one another. A dense paper can be helped by excellent figures. A specialist topic can be helped by unusually clear exposition.

Integrity does not work that way. Beautiful writing does not compensate for fabricated data. Famous authors do not compensate for missing ethics approval. Strong results do not compensate for invented citations. High impact does not compensate for undisclosed manipulation.

EVIDENCE FAILS → HOLD
AUTHORSHIP FAILS → RESOLVE
RIGHTS FAIL → HOLD
ETHICS FAIL → HOLD
CONFIDENTIALITY FAIL → STOP
RELIABILITY FAILS AFTER PUBLICATION → CORRECT / RETRACT
NO OTHER STRENGTH OVERRIDES THE FAILED GATE

This is the core reason integrity systems need hard gates rather than average scores.

Trustworthy scholarship is not the work with the fewest imperfections. It is the work whose evidence, responsibility and correction path remain inspectable when imperfections are found.

86. Why research integrity matters to civilisation

Modern civilisation depends on knowledge that travels far beyond the people who produced it. A clinician acts on published evidence. An engineer adopts a method. A government funds a programme. A teacher explains a discovery. An AI system retrieves a paper and synthesises its conclusions.

None of those downstream users can personally stand in every laboratory or inspect every original notebook. They depend on institutions of trust.

Research integrity is one of those institutions. It does not promise that every published claim is correct. It promises something more realistic and more valuable: that claims should have traceable evidence, accountable authors, declared conflicts, inspectable methods, fair review and a visible path to correction when knowledge changes.

Current authority routes

Continue the Archives and Publishing series

Publication safeguards: evidence, authority, authorship, rights, confidentiality, protection, release, correction, freshness and archive gates.

World Return: When a research claim matters, preserve the whole trust path: who did the work, what evidence exists, what changed during analysis, who funded it, what was disclosed, which version was published, what concerns later arose, and how the record was repaired.

How Research Integrity and Publication Ethics Work | From Authorship, Conflicts and Data Integrity to Misconduct, Investigations, Corrections and Trust