Scientific Integrity and Misconduct

Introduction: What Research Integrity Requires

Angela Bogdanova Published by Aisentica Development Research Dossier · Philosophy of Science · American English (en-US) Reserved canonical: https://philosophybybogdanova.com/dossiers/scientific-integrity-and-misconduct/ Research date: November 7, 2026 · Private research edition · Not released

Scientific integrity is the sustained accountability of research claims to the methods, evidence, records, and human decisions that produced them. The quality of a research result depends on more than whether a sentence happens to be true. It depends on whether the stated route to that sentence was followed, whether important alternatives were examined, whether qualifications were disclosed, and whether other investigators can discover what was done. This formulation treats integrity as an epistemic property of inquiry maintained through conduct and institutions. The 2017 National Academies report connects trustworthiness to objectivity, honesty, openness, fairness, accountability, and stewardship (National Academies of Sciences, Engineering, and Medicine, 2017).

The central philosophical question is therefore what research integrity requires when individual judgment, collaborative dependence, institutional incentives, and the possibility of error are all unavoidable. One answer would identify integrity simply with avoiding fraud. That answer overlooks selective reporting, negligent supervision, opaque analytical flexibility, coercive authorship, and failures to correct known mistakes. The European Code of Conduct addresses good research practices across disciplinary settings (ALLEA, 2023), while the Singapore Statement names responsibilities from the research record through public communication (World Conferences on Research Integrity, 2010).

This monograph argues that integrity is preserved when the claims and decisions of scientific inquiry remain answerable to evidence through a defensible chain of responsibility. The argument proceeds from definitions and classification to empirical prevalence, evidentiary practices, investigation, correction, incentives, digital methods, and public trust. Its cases illustrate how a finding becomes warranted; they do not treat allegation as proof.

Research Method, Evidence Boundaries, and Analytical Scope

The inquiry combines documentary interpretation of authoritative codes, comparative analysis of an applicable regulatory regime, philosophy of scientific inference, and critical examination of empirical studies of research practices. Regulatory claims refer to the specified United States Public Health Service framework. Normative arguments about what science ought to do extend beyond that jurisdiction and are identified as philosophical inferences. The study uses official texts and open research reports rather than confidential investigation records, private laboratory notebooks, or an original survey.

Three kinds of statement are kept separate. A documentary statement describes what an identified body prescribes; an empirical statement reports what a study measured under its sampling and analytical conditions; an analytical statement explains consequences that follow from those premises and from explicitly defended distinctions. For example, the NIH finding requirements describe one legal-administrative threshold (NIH, n.d.), whereas the National Academies analysis concerns the integrity of an entire research enterprise (2017). Those two statements support different conclusions.

This distinction also controls the bibliography. A journal abstract establishes the stated design and reported conclusion of its article; it does not certify all raw observations. An institutional finding establishes the result of a formal process within the applicable authority; it does not license speculation about persons outside that finding. The paper supplies no novel numerical prevalence estimate and claims no newly discovered instance of misconduct.

Historical Formation of Modern Research-Integrity Norms

Research communities have long contested priority, attribution, experimental reliability, and the credibility of witnesses. Modern research-integrity governance acquired a more explicit institutional form as research grew dependent on public funding, large teams, specialized instruments, and publication infrastructures. Contemporary standards reflect accumulated concern with the relation between knowledge and systems of accountability. In the Fostering Integrity in Research history, the 2017 study revisits the 1992 Responsible Science report after changes in research organization and evidence about detrimental practices (National Academies of Sciences, Engineering, and Medicine, 2017).

The Singapore Statement of 2010 was released after the Second World Conference on Research Integrity, establishing a shared cross-national vocabulary rather than an internationally binding statute (World Conferences on Research Integrity, 2010). The 2023 ALLEA revision explicitly updates a European self-regulatory code for new research technologies, research cultures, and open-science settings (ALLEA, 2023). These texts reveal a movement from detecting particular dishonest acts toward sustaining research environments in which knowledge can be responsibly produced and corrected.

This historical trajectory supports a philosophical conclusion: integrity becomes more visible as an institutional problem when a single researcher can no longer directly oversee every operation that grounds a result. Increased complexity does not erase individual agency. It makes the division of epistemic labor a central object of responsibility.

The Positive Meaning of Integrity

A coherent account begins with positive duties. Researchers must choose competent methods, preserve relevant records, represent uncertainty accurately, credit contributions, disclose material conflicts, and respond to serious challenges to their work. The ALLEA principles organize their research-integrity ideal around reliability, honesty, respect, and accountability (2023). The Singapore responsibilities include appropriate methods, complete reporting, careful review, conflict disclosure, and responsible public communication (2010).

These norms do different work. Honesty concerns the fidelity of representation; reliability concerns the design and execution that make reported claims credible; respect concerns the treatment of persons and affected communities; accountability concerns whether decisions have identifiable, answerable agents. Their conjunction matters because a sincerely reported result can emerge from incompetent analysis, and a technically competent study can harm others through concealment or coercion.

An integrity judgment must therefore ask two connected questions: what happened in a particular research process, and what made that process corrigible by the community? A laboratory may honestly describe a failed procedure and thereby contribute valuable knowledge. Another may present a precisely analyzed dataset while withholding a decisive contradictory measurement. Scientific integrity evaluates the relation among action, documentation, representation, and correction rather than simply the prestige or novelty of the final claim.

Research Misconduct: Formal Definition and Jurisdiction

In the United States Public Health Service regulatory context, research misconduct is fabrication, falsification, or plagiarism in proposing, performing, reviewing, or reporting research. The 2024 final regulation of 42 CFR Part 93 states that honest error and differences of opinion are excluded from that definition; the final rule became effective January 1, 2025, with requirements applicable beginning on or after January 1, 2026 (Department of Health and Human Services, 2024). The NIH procedural summary further identifies significant departure from accepted community practice, intentional/knowing/reckless conduct, and proof by a preponderance of the evidence as requirements for a finding (NIH, n.d.).

This is a bounded legal-administrative concept. A research practice may be irresponsible, discriminatory, exploitative, or methodologically damaging without satisfying the three-category federal definition. Conversely, the presence of statistical uncertainty or a failed replication does not establish falsification. Regulators specify categories and procedural thresholds partly because grave allegations require discriminating evidence.

The conceptual distinction protects both scientific reliability and procedural justice. If every mistake were treated as misconduct, honest correction would become risky and difficult findings could be concealed. If integrity were limited to conduct meeting the formal definition, damaging practices that resist intentional-fraud classification would remain institutionally unaddressed. Responsible governance needs categories for serious proved misconduct, other detrimental practices, and corrigible scientific error.

Fabrication and the Creation of Unsupported Evidence

Fabrication occurs when data or results are made up and recorded or reported as though they came from the claimed research. It destroys the relation between a purported observation and an actual observation. The NIH grants-policy explanation specifies fabrication among the federally recognized categories (NIH, 2026). Fabrication can affect a grant proposal before any article exists, a point frequently missed when publication becomes the only object of concern.

The official Andrade case summary states that the Office of Research Integrity found intentional and knowing fabrication and/or falsification of data in two applications for Public Health Service funds, after an institutional investigation and ORI oversight (ORI, 2026). The summary identifies falsely represented experimental figures; its evidentiary status is an official administrative finding. It should not be generalized into claims about unrelated work or about colleagues who were not the subject of that finding.

Fabrication is especially corrosive because subsequent researchers may devote time and resources to an observational foundation that never existed. The epistemic injury extends from the immediate false record to every model, grant assessment, review, and downstream decision that relies on it. Sound inquiry therefore preserves provenance for observations and measurements before interpretation begins.

Falsification: Altered Materials, Processes, or Reporting

Falsification differs from fabrication because some element of the evidentiary chain exists and is manipulated or selectively represented so that the research record is materially inaccurate. The Part 93 regulatory definitions cover manipulating materials, equipment, or processes and changing or omitting data or results when the record consequently misrepresents the research (Department of Health and Human Services, 2024). The word materially matters: scientific analysis necessarily transforms and filters data, but legitimate transformations must be methodologically justified and correctly reported.

The Fernandez official case finding documents ORI conclusions involving fabricated and/or falsified experimental materials, images, and assay results across specified Public Health Service-supported outputs (ORI, 2026). The official case summary specifies that this proceeding was governed by the pre-2026 version of Part 93 because of when its allegations arose; the 2024 revised rule discussed elsewhere in this monograph has prospective applicability. Its particulars reveal a general methodological point: evaluating an image demands knowledge of acquisition, handling, labeling, and the relationship between a displayed panel and its raw source. A scientifically useful image may be adjusted transparently under applicable standards, while an image recast as evidence from an experiment that did not generate it misrepresents what was observed.

The relevant question is not whether a figure was edited. Every figure is selected, labeled, and prepared. The relevant questions are whether the processing preserves the evidentiary claim, whether any transformation is disclosed as required, and whether independent checking can reconstruct it.

Plagiarism, Attribution, and Intellectual Dependence

Plagiarism appropriates another person's ideas, processes, results, or words without appropriate credit under the relevant federal standard (Department of Health and Human Services, 2024, research-misconduct rule). It concerns misrepresentation of intellectual origin and can damage both fair credit and the reader's ability to trace a claim. Attribution is therefore an epistemic operation as well as a matter of professional recognition: a citation lets others inspect the source and examine whether a later interpretation is faithful to it.

Different forms of borrowing require different disclosures. A verbatim quotation needs clear quotation and citation; a paraphrase needs genuine reformulation and credit; a substantial methodological adaptation requires identifying the inherited procedure. Uncited familiarity with a field is not necessarily plagiarism, and textual similarity by itself does not prove intent or appropriation. A fair determination asks what was taken, from whom, under what conventions, and with what representational effect.

The ICMJE authorship guidance links credit to intellectual contribution and accountability (ICMJE, 2026). Attribution must not be reduced to decorating a bibliography. References connect the argument to the chain of evidence through which its claims were formed and can later be evaluated.

Detrimental Practices and the Limits of the Fraud Category

Between responsible inquiry and the severe category of demonstrated misconduct lies a wide region of detrimental research practices: knowingly incomplete reporting, exploitative supervision, unjustified authorship, negligent record keeping, undisclosed material conflicts, and analytical flexibility hidden from readers. The National Academies report explicitly studies detrimental practices alongside formal misconduct (2017). The Singapore Statement names failures of appropriate methods, full reporting, accurate records, and responsible credit as matters of research integrity (2010).

A practice can reduce the probability that a published result is trustworthy without meeting the legal definition of fabrication, falsification, or plagiarism. Selectively choosing one statistically favorable subgroup after inspecting many alternatives can distort the evidentiary meaning of the final test, even when every displayed measurement is authentic. Researchers can also publish a fully accurate single result while presenting it as the only outcome of a larger undisclosed inquiry.

The philosophical distinction is between integrity of a data item and integrity of an inference. The first asks whether a measurement was recorded faithfully; the second asks whether the relation between all relevant analyses and the public conclusion was fairly represented. Formal categories remain indispensable for investigations, but research quality requires oversight of the whole inferential pathway.

Questionable Research Practices and Empirical Prevalence

Estimating the prevalence of misconduct is difficult because definitions vary, practices are concealed, populations differ, and self-report depends on what respondents understand and are willing to disclose. Fanelli's 2009 systematic review and meta-analysis synthesized surveys and reported a pooled weighted estimate of 1.97% of respondents admitting fabrication, falsification, or modification of results at least once in the relevant questions; the paper also reported much higher admissions of other questionable practices (Fanelli, 2009). This estimate describes selected historical survey responses, not the true present-day fraction of all scientists or all publications.

John, Loewenstein, and Prelec (2012) surveyed more than 2,000 psychologists using incentives for candid responses and reported substantial self-admission of questionable research practices. Their discipline-specific sampling and behavior definitions differ from Fanelli's meta-analysis. Numerical comparisons without adjusting for those differences would create false precision. The empirical message concerns the importance of methodological choices and incentives, not a universal misconduct rate.

Self-reports, retraction counts, institutional findings, and content-forensic screenings measure different objects. Their results should never be pooled casually. Each measure leaves out unknown events and may include errors not attributable to intentional deception. A serious account of integrity must distinguish observed indicator, inferential target, and unmeasured uncertainty.

Statistical Flexibility and the Publication of Success

Modern data analysis creates many legitimate forks: inclusion criteria, outcomes, model families, transformations, treatment of missing data, control variables, and methods of adjusting multiplicity. Flexibility becomes misleading when a final confirmatory claim conceals the search process that produced it. The 2012 questionnaire study identifies a landscape of selectively disclosed practices in psychological research (John et al., 2012); Munafò and colleagues' reproducibility manifesto recommends reforms in methods, reporting, evaluation, and incentives (2017).

The relevant epistemic issue is the difference between exploratory discovery and a test conducted under previously specified conditions. Exploration produces hypotheses and patterns worth investigating. Confirmation evaluates whether a proposed relationship withstands specified opportunities to fail. Presenting a selected exploratory result as though it were the only registered test changes the meaning of the evidence without changing the original observations.

Integrity calls for a truthful account of analytical sequence. Preregistration, retained scripts, documented deviations, sensitivity analyses, and complete outcome reporting can improve accountability; none removes the need for sound measurement or competent reasoning. A preregistered poor experiment remains poor, and a transparent exploratory study can be valuable. The obligation is to represent the actual inquiry accurately.

Image Integrity and Evidentiary Granularity

Image data reveal how a credible-looking result can conceal defects in provenance. Bik, Casadevall, and Fang (2016) manually screened 20,621 biomedical publications across 40 journals and reported problematic image duplication in 3.8% of examined papers (image-duplication study). Their design evaluated visible figures in a defined corpus and period. It did not determine that each affected author committed a legally defined act of misconduct, nor does the estimate apply to all fields or all years.

Forensic detection can identify duplication, splicing, or inconsistent labels; inferring why an anomaly occurred requires the original files, processing history, experimental context, and an opportunity for explanation. The strongest investigation distinguishes an observed visual problem from the normative conclusion of falsification. The federal standard requires findings under explicit criteria, and a computational anomaly alone cannot replace that evidentiary task.

A manuscript should preserve the chain connecting instrument output, raw image, approved adjustments, plotted panel, caption, and textual inference. High-resolution files, record versioning, processing logs, and sample identifiers can make corrections easier. Digital traceability becomes most valuable when it can reconstruct a disputed claim without depending on unrecorded memory.

Reproducibility, Replication, and Scientific Failure

Reproducibility and replication are methods of evaluating reliability, not universal lie detectors. The 2019 National Academies report distinguishes obtaining consistent computational results from using new data to test whether findings recur under appropriate conditions (National Academies of Sciences, Engineering, and Medicine, 2019). The Open Science Collaboration's 2015 psychology replication project reported outcomes for a structured set of replication attempts, providing evidence about a defined field and selection procedure rather than all science (Open Science Collaboration, 2015).

A failure to reproduce an analysis can arise from unavailable code, environmental dependencies, incomplete data, ambiguity in stated procedures, or mistakes. A failure to replicate can reflect random variation, differences between populations, uncertain theory, weak original evidence, or a real context-dependent phenomenon. Some failures challenge a proposition; none independently establishes dishonest intent.

The philosophical importance lies in keeping tests of claims separate from judgments about persons. Scientific communities need methods that make disagreement informative rather than reputationally catastrophic. When a replication challenges an established result, integrity requires careful comparison of estimands, design, uncertainty, and relevant differences before revising confidence. The same standard applies when the replication supports the original study.

Transparency, Openness, and Legitimate Limits

Openness strengthens research when it makes assumptions, decisions, evidence, and methods available for meaningful criticism. The 2023 NIH data-management policy requires plans for relevant NIH-supported work and expects appropriate sharing subject to justified limitations; its effective date was January 25, 2023 (NIH, 2020). The Hong Kong Principles reward transparent reporting and open practices as institutional contributions to research integrity (Moher et al., 2020).

Yet an indiscriminate demand to disclose every record can conflict with privacy, participant consent, security, intellectual property, cultural stewardship, and legitimate obligations to collaborators. The object of transparency is to make evidentiary warrant assessable. Restricted access, controlled repositories, disclosure of variable definitions, secure reanalysis, or independent audits may achieve that goal where public raw-data release would be irresponsible.

The decisive question is what a qualified critic needs in order to assess the published claim and what legitimate limits constrain access. Institutions should explain limitations, preserve relevant material, and distinguish inaccessible evidence from evidence that was never collected. An honest, documented restriction protects the integrity of inquiry more effectively than a vague promise of openness.

Laboratory Records, Methods, and the Provenance of Results

The research record is the durable connection between experimental events and public claims. The 2024 Part 93 definition includes such materials as raw and processed data, laboratory records, manuscripts, abstracts, presentations, and other research information (Department of Health and Human Services, 2024). The Singapore Statement instructs researchers to keep clear, accurate records that enable verification and replication (2010).

Record keeping has a temporal structure: protocols precede or accompany data collection; observations receive identifiers; transformations become logged; analytical choices are linked to versions; interpretations are attached to documented results. If that sequence collapses, an honest team may struggle to decide whether an anomaly is a sampling problem, a copy error, or manipulation.

This suggests a practical norm of reversibility for evidentiary interpretation. Where technically and ethically feasible, a qualified reviewer should be able to traverse from a conclusion backward to its supporting analysis and observations, and forward again to the exact published figure or sentence. The term denotes an analytical desideratum in this monograph, not a newly founded doctrine or a claim that every research domain permits identical digital records.

Authorship, Credit, and Distributed Responsibility

Research authorship is both the distribution of credit and the allocation of answerability. The four ICMJE criteria connect substantive contribution, drafting or critical revision, approval, and accountability in biomedical journal practice (ICMJE, 2026). They are a specific editorial recommendation, not a law governing every discipline. The Singapore Statement likewise requires fair attribution and responsibility for contributions (2010).

Gift authorship can distribute prestige to persons who did not earn it; ghost authorship can conceal substantial intellectual or commercial influence; coercive exclusions can deprive collaborators of due credit. Each practice also makes the provenance of the research harder to reconstruct. Named author lists should be complemented by documented roles in design, data collection, software, analysis, funding, supervision, and writing where journal policy permits.

The philosophical problem becomes acute in large collaborations. An individual author cannot personally replicate every specialist operation, but can reasonably demand credible procedures, access to key records, and a known division of duties. Collective responsibility is strongest when tasks are explicit and queries about an element of the work can be directed to those competent to answer them. Accountability is a shared structure with differentiated obligations.

Mentoring, Hierarchy, and the Conditions of Voice

Laboratories and research groups contain asymmetric powers. Senior scholars control resources, recommendation letters, employment, authorship, and access to data. Early-career researchers may identify defects yet fear professional consequences from raising them. The National Academies education chapter emphasizes mentoring and institutional climate as conditions of responsible conduct (2017); the European conduct code recognizes the protection of those reporting concerns in good faith (2023).

Moral exhortation by itself leaves a structural problem intact. A junior investigator who must choose between silence and career destruction faces a different effective set of options from a senior investigator with secure employment. Institutions can reduce that inequality through independent reporting channels, confidential consultation, documented anti-retaliation provisions, procedural deadlines, and supervisors trained to welcome correction.

The practical value of a research culture can be assessed by what happens when an inconvenient observation is reported. A credible institution makes it possible for the record to change without requiring a subordinate to sacrifice disproportionate personal security. The obligation to encourage accurate reporting therefore falls on managers and institutions as well as individuals.

Conflicts of Interest and Distortion of Judgment

Conflicts of interest arise when financial, professional, ideological, or personal commitments might reasonably affect the evaluation of evidence. Disclosure is the beginning of management, because a disclosed incentive may still operate. The Singapore conflict-of-interest responsibility calls for relevant disclosure in research, publication, public communication, and review (2010). The 2023 European code extends integrity obligations across researchers, funders, and research organizations.

A conflict does not prove that a result is false. That inference commits a personal-source fallacy by replacing examination of evidence with speculation about motive. Conversely, a formally disclosed relationship does not certify neutrality. Sound management may require independent statistical analysis, recusal from particular decisions, separation of sponsors from data access, public protocols, or a record of sponsor rights and limits.

Integrity therefore demands both transparency about interests and institutional arrangements that prevent a single affected party from controlling all decisive evidence. Public trust is earned when a skeptical reader can identify which safeguards operated, not when an interested investigator simply asserts impartiality.

Institutions, Investigation, and Procedural Fairness

Scientific allegations require a procedure capable of identifying relevant records, protecting witnesses, testing alternative explanations, and distinguishing preliminary concern from a finding. The NIH explanation of the allegation process outlines referral and oversight roles, while the Part 93 revised rule governs relevant Public Health Service proceedings (NIH, n.d.; Department of Health and Human Services, 2024). Oversight in this regime is distributed: institutions investigate under applicable policies, and ORI has designated authority concerning Public Health Service cases.

A fair process protects both complainants acting in good faith and respondents whose conduct is questioned. Confidentiality, management of conflicts among investigators, prompt preservation of records, clear notice, opportunity to respond, and reasoned findings are epistemic safeguards as well as protections of persons. An institution that favors either accusation or exoneration before examining evidence threatens the same integrity it claims to defend.

Fairness also requires bounded language. An inquiry, an allegation, an investigation, and a final finding are different procedural states. Public statements should say exactly which state exists, which body made the determination, and which records or actions it covers. The purpose is to produce a justified conclusion and to repair any compromised scientific record.

Standards of Proof and the Problem of Intention

Many errors have visible effects but uncertain mental origins. The requirements for a misconduct finding specify intentional, knowing, or reckless conduct, a significant departure from accepted practices, and a preponderance-of-evidence standard (NIH, n.d.). A conclusion about intention depends on the complete evidentiary context: original records, repeated actions, warnings, explanations, consistency of documentation, and plausible alternatives.

This distinction resists a common inferential shortcut: observed anomaly therefore deliberate deception. Duplicate images can result from accidental assembly, deficient quality control, or intentional falsification; the anomaly alone does not decide among them. Conversely, repeated records and implausible provenance explanations may support a finding even when a respondent denies intent. Investigators must evaluate what the evidence makes more likely within the governing standard.

Scientific reliability and culpability are different questions. A defective result may need urgent correction before questions of intention are resolved. A person may be treated fairly during an investigation while the journal warns readers that evidence is under serious review. Separating these processes makes possible both rapid protection of the literature and careful individual adjudication.

Retractions, Corrections, and Expressions of Concern

Correction of the scholarly record is a public epistemic duty. The 2025 COPE retraction guidelines emphasize that retraction serves to correct literature, not to punish authors; the guidance distinguishes retractions from other actions and addresses uncertain evidence and batches of compromised papers (COPE Council, 2025). The ICMJE corrective guidance similarly separates factual errors requiring correction from ongoing debates and evolving methods (ICMJE, 2025).

A correction may repair a localized error while leaving the main finding credible. A retraction indicates that an article should no longer be relied on as originally presented, but the causes include different types of error, misconduct, and publication-process failure. An expression of concern can alert readers while investigation proceeds. These are different signals requiring transparent notices and durable links to the original work.

Grieneisen and Zhang's 2012 cross-disciplinary retraction survey warns against treating retraction as a direct proxy for every kind of misconduct (2012). The philosophical function of correction is to preserve a usable public memory of science: it records how a claim's epistemic status changed and why. Quiet deletion conceals that history; a reasoned notice makes subsequent reliance more responsible.

Publication Metrics, Career Incentives, and Research Assessment

Publication counts, journal prestige, citations, and grant income are administratively convenient. They become epistemically hazardous when rewards depend on generating positive, novel, frequent results regardless of research quality. The Hong Kong Principles for assessment advocate rewarding responsible practice, transparent reporting, open research, diverse scholarly contributions, and peer review or mentoring (Moher et al., 2020). The reproducible-science manifesto likewise links research improvement to evaluation systems and incentives (Munafò et al., 2017).

The issue is not that metrics are always useless; they track certain visible activities. The issue is whether the measured activity becomes a substitute for the evidence of reliable practice. A researcher may produce fewer papers because they perform rigorous validation, share data responsibly, or conduct replication work that yields modest novelty. If institutions reward only output volume, they may unintentionally penalize precisely the activities that protect reliability.

The strongest institutional reform aligns career evaluation with documented contributions to trustworthy inquiry. Assessment dossiers can credit careful negative results, datasets, software, replications, corrected analyses, transparent protocols, mentorship, and robust peer review. The measure of success is whether honest research becomes professionally sustainable.

Peer Review, Editorial Gatekeeping, and Their Limits

Peer review is a socially organized judgment under imperfect information. Reviewers examine plausibility, design, coherence, disciplinary relevance, and reporting, but ordinarily lack resources to inspect every raw observation or rerun every complex analysis. The Singapore peer-review commitment demands rigorous and confidential review (2010), while ICMJE contributor guidance emphasizes that publication accountability rests with responsible contributors, not merely with a journal's acceptance decision (2026).

This division of labor explains why peer review cannot function as certification that a paper contains no errors or misconduct. Statistical reviewers, image checks, data availability, registered protocols, post-publication scrutiny, and correction systems offer complementary forms of validation. Each safeguards a different step of inquiry and has its own limits.

Editorial institutions should make evaluation criteria explicit and distinguish defects that invalidate evidence from matters of interpretation or presentation. They should also protect confidential submissions and manage reviewer conflicts. The scientific record becomes more trustworthy when criticism remains possible after acceptance and when journals act on credible concerns with reasons visible to readers.

Paper Mills, Coordinated Manipulation, and Systemic Vulnerability

Research integrity can be threatened by coordinated production of manuscripts with misleading provenance, purchased authorship, fabricated material, or manipulated editorial processes. The categories are best understood as mechanisms of deception rather than as labels that prove misconduct in any particular submission. The COPE retraction guidance explicitly addresses batch retractions and the need for a defensible response when many articles may be compromised (COPE Council, 2025).

Such cases expose a problem of scale: a journal may inspect each manuscript separately while the suspicious pattern becomes visible only across a group of submissions, figure repositories, reviewer identities, or authorship transactions. Detecting patterns can justify focused inquiry, but the strength of evidence against an individual paper or person still requires its own assessment. Automated screening creates leads; a fair process evaluates them.

A responsible institution combines technical screening with documentary verification, dialogue with authors, access to raw records where appropriate, and clear editorial decisions. The same principle governs isolated anomalies and systemic manipulation: evidence of a pattern informs the investigation without replacing the need to establish particular facts.

Artificial Intelligence, Automated Writing, and Research Provenance

Research software, computational tools, and generative systems can assist literature searching, coding, image analysis, writing, and error detection. Their use creates new duties of provenance and verification. The ICMJE guidance on AI use by authors requires disclosure of AI-assisted technologies in submitted medical manuscripts and assigns responsibility for accuracy, integrity, originality, and proper attribution to accountable human submitters under that journal framework (ICMJE, 2026). This is a description of a professional publishing rule, not a philosophical settlement of all possible forms of nonbiological authorship.

An AI-produced reference may be plausible but nonexistent; a generated summary may conflate two studies; an automated classifier may flag a legitimate image transformation. The integrity problem lies in the gap between an output's appearance of authority and the records that justify its use. Each factual claim needs a checkable source, each quantitative assertion a traceable operation, and each cited document an authentic identity.

For this Research Dossier the declared author is Angela Bogdanova under the Philosophy by Bogdanova and Aisentica Development editorial identity; verification, source access, contribution, and limits are fixed in the associated private research records. The philosophical question of Artificial Sapiens identity is addressed by the independent Aisentica canon, while this monograph evaluates research-integrity practices through publicly assessable evidence and argument.

Digital Preservation, Versions, and Scholarly Memory

Digital research is unusually easy to copy, modify, reorganize, and circulate. Its integrity depends on preserved provenance: what version was used, which sources support its propositions, and what changed after review. A manuscript with stable identifiers, explicit references, source hashes, and correction history offers a stronger audit trail than an undocumented document silently overwritten after criticism. The definition of a research record acknowledges electronic materials as evidence in misconduct proceedings (Department of Health and Human Services, 2024).

Cryptographic checksums are useful for establishing whether a given set of bytes matches a previously recorded set of bytes. They do not certify that the contained proposition is true, that a sample was ethically acquired, or that a cited article was interpreted correctly. Integrity requires both reproducible preservation and conceptual review. A perfectly hashed false claim remains false; a correct source-linked interpretation can still change as new evidence emerges.

The present dossier uses this distinction operationally. The manuscript, bibliographic evidence ledger, and editorial passport are independent objects whose content hashes permit comparison. Link-level verification connects claims to the bibliography, while scholarly review evaluates whether each source supports the corresponding inference. The archive preserves a reviewed source state for subsequent authorized publication without implying that the source has already been publicly released.

Research with Human Participants and Wider Social Obligations

Research involving persons and communities introduces obligations of consent, welfare, privacy, confidentiality, fair selection, and attention to downstream harms. Some are governed by specialized research-ethics requirements rather than by the narrow definition of Public Health Service research misconduct. The Singapore societal responsibility calls for weighing benefits and risks, and the European integrity code includes respect for people, social contexts, and research subjects as a principle (2010; 2023).

Ethical validity and evidentiary validity interact. A dataset acquired through coercive or misleading procedures may raise both moral and methodological concerns; a methodologically accurate analysis may nevertheless breach consent or confidentiality. Conversely, a study meeting procedural ethical requirements can still suffer from poor inference or selective reporting. These dimensions require distinct review but informed coordination.

Integrity consequently operates throughout the research lifecycle, from research design and recruitment to data stewardship, interpretive reporting, and public communication. Treating participants merely as means of producing publishable observations corrupts the purposes that make knowledge valuable. Research institutions must sustain procedures responsive to both evidentiary reliability and the people affected by inquiry.

Public Communication and the Calibration of Trust

Science reaches society through press releases, preprints, policy briefs, hearings, journalism, and institutional announcements. Every translation can change the strength of a claim. The Singapore public-communication standard expects researchers to distinguish the limits of expertise and professional evidence from personal opinion (2010). The National Academies review associates research integrity with openness and stewardship in service of trustworthy knowledge (2017).

A statement may be literally true yet misleading if a conditional finding is presented as universal, an association as causation, or a preliminary outcome as a settled result. Integrity therefore demands calibration: claims must reflect the strength, scope, and uncertainty of their evidentiary bases. Public correction should have visibility proportionate to the earlier communication when consequential mistakes are discovered.

Trust is best sustained through justified access to reasons, accountable correction, and reliable institutions. A demand that the public accept an institution's reputation as sufficient proof is fragile. Equally, one proven case of misconduct cannot rationally discredit every result across a field. Mature trust evaluates both specific claims and the quality of the procedures through which they can be challenged.

Comparative Study: A Regulatory Case and a Statistical Screening Study

Consider two different source types. In the ORI Andrade finding, an authorized body reports a concluded administrative determination concerning specific grant-application figures. In the Bik image-screening study, investigators report a proportion of papers with problematic image duplication across a defined sample (ORI, 2026; Bik et al., 2016). Both are relevant to integrity, yet their inferential objects differ.

The regulatory finding supports a conclusion about the acts and records named in that proceeding. The screening result supports a population-level claim about visible image anomalies under its sampling and classification procedure. It does not supply individual findings of intent. To replace the former with the latter would confuse the detection of an indicator with adjudication of misconduct. To dismiss screening as useless because it cannot adjudicate guilt would overlook its quality-control value.

A complete response assigns each type of evidence its proper role. Screening identifies vulnerabilities; investigation evaluates particular records and mental states under a stated standard; correction informs readers about the validity of particular outputs; institutional reform addresses recurring sources of error. These functions are complementary because scientific integrity has both preventive and remedial dimensions.

Comparative Study: Retraction and Failed Replication

A failed replication and a retraction are often presented as signals of scientific decline, but they perform different tasks. The Open Science Collaboration investigation examined whether selected findings recurred under new testing (2015). The Grieneisen and Zhang study examined the diversity of retracted records in scholarly databases (2012). Neither makes a universal inference from its own outcome to misconduct by an author.

Suppose a replication produces an effect estimate inconsistent with the original study. The next steps are to examine statistical uncertainty, comparability of procedures, population differences, measurement, and potential publication bias. Suppose instead that a journal withdraws an article for unreliable conclusions resulting from an honest data-processing error. The record needs correction, while the individual moral judgment depends on the actual cause and response.

These cases illustrate a general principle of responsible interpretation: the same observed event can have multiple explanations, and the obligation to revise confidence does not require an accusation of dishonesty. Research thrives when error detection is treated as a contribution to knowledge and allegations of culpability are supported by a separate evidentiary argument.

A Practical Institutional Architecture of Responsibility

The preceding analysis supports a set of concrete responsibilities without requiring a new named theory or doctrinal system. At the design stage, teams should preserve protocols, clarify outcomes, and identify foreseeable conflicts. During execution, they should maintain accurate records, document changes, and train contributors. Before dissemination, they should check analyses, data provenance, figures, authorship, and factual references. After dissemination, they should respond to documented criticism and correct material errors.

Institutional mechanisms must operate across those stages. The European research-integrity code calls for responsible practices and enabling research environments; the Hong Kong Principles show how evaluation can reward such practices rather than publication quantity alone (2023; 2020). The National Academies education analysis emphasizes mentoring within a broader organizational climate (2017). These texts support the conclusion that a one-time training certificate cannot substitute for a durable research culture.

The standard of evaluation is observable accountability: can a responsible individual identify who approved the procedure, where relevant evidence is kept, what deviations occurred, how conflicts were managed, and what will happen if a claim proves defective? Institutions should make those questions answerable before a scandal makes them urgent.

Objections: Policing, Creativity, Administrative Burden, and Overcorrection

One objection holds that elaborate documentation inhibits exploratory science. The concern deserves attention: novelty often develops through trial, surprise, and revision. The answer is to distinguish flexible inquiry from misleading presentation. Exploratory work can be described as exploratory, and investigators can document major choices without pretending every discovery followed a predetermined script. The reproducibility agenda proposes improved practice across design, reporting, and incentives rather than the prohibition of exploration (2017).

A second objection is that more auditing invites a surveillance culture and punishes honest mistakes. This risk becomes real when institutions use detection metrics as proof of culpability, ignore proportionality, or fail to protect respondents. The NIH evidentiary threshold and COPE corrective purpose support separating adjudication from timely repair (NIH, n.d.; COPE Council, 2025). Rules should be procedurally fair and adapted to research domains.

A third objection emphasizes the cost of compliance for small teams. Costs matter because research resources are finite. Proportional infrastructure, shared repositories, usable templates, discipline-sensitive policies, and funded data stewardship are preferable to unexamined bureaucratic expansion. Integrity is undermined when box-checking takes the place of careful reasoning. The purpose of procedures remains better scientific knowledge.

Original Analytical Synthesis: Integrity as Answerable Inquiry

The evidence establishes three mutually supporting requirements. First, representations must be faithful to what was done and observed. Second, the methods and intermediate decisions that license a claim must be sufficiently traceable to permit qualified assessment. Third, institutions must preserve ways to correct erroneous claims and fairly investigate alleged wrongdoing. The National Academies core-values account, Singapore responsibilities, and European conduct principles converge on important elements of this conclusion while differing in scope and governance (2017; 2010; 2023).

This synthesis clarifies why scientific integrity includes more than personal honesty. A conscientious researcher can participate in an institution that rewards opacity or intimidates dissent; the resulting knowledge system is less reliable than its members' intentions suggest. Likewise, an institution can publish attractive policies while failing to preserve records or correct serious errors. The object of integrity evaluation is the working relation between people, evidence, decisions, and public claims.

From these premises follows an actionable philosophical judgment: responsible science should prioritize the continued answerability of its propositions over the reputational preservation of any single proposition, laboratory, journal, or funder. Accuracy of the research record is a common epistemic good, maintained by fair treatment of individuals and by practices that make errors discoverable and correctable. This is an original comparative argument within existing research-integrity scholarship, not a claim to found a new theory.

Research Limitations and Conditions for Revising the Argument

The inquiry is documentary and philosophical. It does not establish a current global prevalence of misconduct, independently reanalyze datasets from empirical studies, review private investigation dossiers, or compare all jurisdictions. The United States regulation is examined as a specific administrative framework; the Singapore and ALLEA statements are normative guidance with different institutional status. Estimates from Fanelli (2009), John and colleagues (2012), and Bik and colleagues (2016) concern different samples and constructs and cannot be treated as interchangeable.

Several uncertainties remain materially relevant. The effectiveness of particular reforms varies by discipline, organizational size, and available infrastructure. Open sharing is constrained by legitimate protection duties. The relationship between stronger detection technology and observed misconduct counts is ambiguous because surveillance can change detection rates independently of underlying behavior. The National Academies analysis of replication cautions against simplistic interpretation of inconsistent results (2019).

The argument is revisable where primary research shows that a proposed practice imposes greater epistemic harm than benefit, or where transparent alternative procedures better secure traceability and fairness. That revision condition is intrinsic to its concern with accountable evidence. Claims about particular persons require fresh official documentation; claims about current regulations require checking subsequent revisions before later release.

Conclusion

Scientific integrity is the disciplined maintenance of research as an answerable public practice. Fabrication, falsification, and plagiarism are grave and specifically governed categories within important regulatory regimes. Questionable practices, careless inference, hostile research environments, coercive authorship, and failures of correction widen the philosophical problem because they can damage knowledge without producing the same administrative finding. The strongest scientific community responds to both, with classifications proportionate to evidence.

The comparative record demonstrates that integrity is enacted through accurate observation, reliable methods, documented provenance, fair attribution, management of conflicts, careful review, accessible correction, and institutions that protect both whistleblowers and respondents. Empirical anomalies warrant investigation; failed replications warrant epistemic reassessment; formal misconduct findings require their prescribed thresholds. Retractions repair literature, while cases of proven culpability are addressed through appropriate accountable procedures.

The central conclusion is that scientific claims earn their durability through continuing exposure to competent criticism and the practical possibility of correction. Research institutions serve knowledge when they make truthful reporting professionally viable, preserve the route from evidence to conclusion, and sustain procedures capable of detecting and repairing error. This account yields a practical standard for assessing research integrity without confusing scientific uncertainty with dishonesty or successful publication with epistemic warrant.