What Research Records Should You Keep, and Why Does Research Integrity Depend on Them?

What Evidence of Your Research Should Still Exist After the Paper Is Written?

A published paper is a compressed account of a much larger research process. It may report the final sample, methods, analyses, and conclusions, but it rarely preserves every observation, protocol version, coding decision, instrument setting, exclusion, correction, or analytical step that produced those conclusions.

That creates a practical question: what should you keep?

The answer matters because research integrity depends partly on whether there is reliable evidence of what actually happened. If a result is questioned months or years later, the manuscript alone may not be enough to determine how the study was conducted or how a particular number was produced.

Keep Enough of the Research Record to Reconstruct and Support the Work

A Research Record Is Broader Than the Dataset in Your Final Analysis

Researchers sometimes treat “research records” as another name for “research data.” That is too narrow. Data may be central to the record, but the record can also include materials that establish where the data came from, what happened to them, which procedures were followed, and how decisions were made.

Under the U.S. federal research misconduct framework, for example, the research record includes the record of data or results embodying the facts resulting from scientific inquiry and may include research proposals, laboratory records in physical or electronic form, progress reports, abstracts, theses, oral presentations, internal reports, and journal articles. This is a useful illustration of how broad the concept can be, although the precise legal or institutional definition applicable to your own work may differ.

The distinction becomes important when deciding what to preserve. A clean dataset may show the values you analyzed, but it may not explain why certain observations disappeared, which version of an instrument produced them, or how raw responses became analytical variables. The difference between research data and research records therefore affects what a defensible archive needs to contain.

Keep Records of What You Planned

Depending on the study, relevant planning records may include research proposals, protocols, preregistrations, analysis plans, sampling plans, codebooks, instrument versions, interview guides, laboratory procedures, ethics approvals, amendments, data-management plans, and other documents that define how the research was intended to proceed.

These materials provide a baseline against which later changes can be understood. A protocol that changed during the project is not necessarily evidence of a problem. Research often requires legitimate adaptation. The important issue is whether the history of those changes remains understandable.

Keep Evidence of What You Actually Did

A protocol tells you what was supposed to happen. Records created during the research help establish what did happen.

Depending on the methodology, these could include laboratory notebooks, field notes, recruitment and screening logs, instrument records, survey administration documentation, interview or observation records, calibration information, experimental logs, case report forms, electronic system records, or other contemporaneous documentation.

The appropriate record varies considerably across disciplines. A historian working with archives, an education researcher conducting classroom observations, a computational scientist running simulations, and a laboratory researcher performing experiments will not produce identical documentation. The relevant question is whether the records are appropriate to the methods and sufficient to establish how the work was carried out.

Preserve the Connection Between Raw Evidence and Reported Results

Keeping the final spreadsheet is useful, but often insufficient. A trustworthy record should preserve the provenance of important research outputs: where information originated and what happened to it before it appeared in a table, figure, model, quotation, or conclusion.

That may require retaining original or source data where permitted, cleaned datasets, codebooks, transformation documentation, analysis scripts, software settings, coding frameworks, calculation files, version histories, or other materials needed to understand the analytical path.

This does not mean that every intermediate file must necessarily be preserved forever. Rather, the records you retain should make consequential transformations understandable. The more difficult it would be to explain a result without a particular record, the stronger the case for preserving that record.

Document Decisions That Affect the Evidentiary Chain

Some of the most important research records contain decisions rather than measurements. Why was an observation excluded? Why was a coding category revised? Why did recruitment stop? Why was one analytical model reported instead of another? Why did the team depart from an original procedure?

Not every research decision belongs in the published article, but that does not make the underlying documentation disposable. Records of decisions that never appear in the final paper may later be essential for understanding how the published work was produced.

Preserve Corrections Rather Than Making the History Disappear

Research records can contain mistakes. A transcription error may be discovered, a sample identifier may need correction, or a field entry may turn out to be wrong. Correcting such problems can be entirely appropriate.

The integrity issue is whether the correction obscures the history of the record. Where feasible and appropriate to the system being used, corrections should preserve what was changed, when it was changed, and enough information to understand the reason for the change. The specific procedures should follow applicable institutional, regulatory, laboratory, or system requirements. Detailed questions about documenting corrections to research records deserve particular care because poorly controlled editing can make legitimate corrections difficult to distinguish from retrospective alteration.

Keep Metadata and Context That Make Records Interpretable

A file can technically survive while becoming scientifically useless. A folder containing “finaldata2_revised_REALFINAL.csv” may be evidence that academia remains undefeated in file naming, but the filename alone does not explain what its variables mean or how it relates to the study.

Useful records therefore need context. Depending on the project, this can include dates, identifiers, variable definitions, units, version information, software or code dependencies, instrument details, file relationships, provenance information, and explanations of processing steps.

This is part of what makes an adequate research record more than a collection of surviving files.

Preserve Records of Oversight and Compliance When They Matter

Research may also generate records associated with ethical, regulatory, contractual, or administrative requirements. Examples can include ethics approvals and amendments, consent documentation where retention is required and permitted, permissions, safety documentation, data-use agreements, monitoring records, or correspondence documenting consequential approvals and decisions.

These materials should be handled according to the applicable requirements, particularly when they contain identifiable, confidential, proprietary, or otherwise sensitive information. “Keep the record” does not mean “make the record broadly accessible.” Preservation and access are separate questions.

Research Integrity Depends on Traceability, Not Just Trust

Good research records allow claims about the research process to be checked against contemporaneous evidence. This matters during ordinary collaboration, manuscript preparation, replication attempts, audits, monitoring, corrections, data reuse, and investigations of possible problems.

Research misconduct frameworks make the connection especially visible. The U.S. Office of Research Integrity defines falsification to include changing or omitting data or results so that research is not accurately represented in the research record. The point is broader than misconduct proceedings, however. Records help distinguish documented research practice from reconstruction based largely on memory.

That distinction can protect researchers as well as expose problems. If a disputed result was produced legitimately, contemporaneous records may demonstrate exactly what happened. If an honest error occurred, a coherent record may help identify where it entered the workflow and what findings it affected.

Good Records Also Make Research More Reproducible and Reusable

NIH describes data management as involving the validation, organization, protection, maintenance, and processing of scientific data to support accessibility, reliability, and quality. It also notes that careful documentation helps researchers and collaborators use data consistently and accurately and can support future use.

Record keeping and data sharing are nevertheless not identical. NIH's Data Management and Sharing Policy, for example, defines “scientific data” for that policy in a way that specifically excludes materials such as laboratory notebooks, preliminary analyses, completed case report forms, paper drafts, peer reviews, and communications with colleagues. Some materials can therefore be important research records without being scientific data that must be shared under a particular data-sharing policy.

There Is No Universal Retention Period for Every Research Record

How long records must be kept depends on the rules governing the project. Requirements can come from institutions, funders, sponsors, laws, regulations, ethics bodies, contracts, journals, repositories, or disciplinary standards, and different categories of records may be governed by different periods.

For example, NIH states that recipient institutions generally must retain records pertinent to an award for three years from the date of submission of the final expenditure report, subject to exceptions and other applicable requirements. NSF likewise specifies a general three-year retention period for records pertinent to an award from financial closeout, subject to the federal rules it cites. These examples should not be converted into a universal “three-year rule” for research records.

Before destroying anything consequential, determine which requirements apply to the particular record and whether circumstances such as an audit, dispute, investigation, patent matter, participant-protection requirement, or institutional hold require longer preservation.

What a Defensible Record Looks Like When a Result Is Questioned

No single file proves the integrity of the entire study. Together, however, these records establish a traceable path from the original evidence through the exclusion decision to the published result. The team does not have to reconstruct the story from memory two years later.

That is the practical value of documentation: not paperwork for its own sake, but evidence that allows the research process to be examined.

Research Records Are Often Lost Through Ordinary Assumptions

Decide What to Keep by Asking What the Record Must Be Able to Show

A useful record-keeping strategy begins with reconstruction rather than storage capacity. Imagine that a knowledgeable person later asks how one important finding in your paper was produced. What evidence would they need to follow the chain from the research plan to the original observations, through processing and analysis, to that finding?

The answer will vary by methodology, but this question exposes documentation gaps remarkably quickly.

Set these practices at the beginning of a project where possible. Retrospective documentation is inherently more difficult because files disappear, software changes, team members leave, and memories become less reliable. Record keeping works best as part of the research workflow rather than as an archival exercise performed after manuscript acceptance.

Finally, establish responsibility. Team members should know where authoritative records are stored, which versions are controlled, who can access them, how corrections are made, and what happens to records when personnel leave. A research record that exists only on one researcher's personal device is a continuity problem waiting for excellent timing.

Before You Consider Your Research Record Complete

Common Questions About Research Records

Preserve the Evidence Behind the Research, Not Just the Paper

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.