03 · What You Need to Know
The Right Standard Is Reconstruction, Not Volume
There is no universal number of pages, files, screenshots, log entries, or metadata fields that makes research sufficiently documented. A small observational study and a regulated multicenter clinical trial plainly do not require identical documentation systems.
The better question is functional: what must the record allow someone to establish?
Start With What You Would Need to Prove About the Research
For consequential parts of a study, documentation should usually be capable of establishing several basic facts where they are relevant:
- what procedure, analysis, or activity occurred;
- which evidence, participants, samples, files, or materials were involved;
- who performed or authorized an important action;
- when the activity or decision occurred;
- which protocol, instrument, dataset, software, code, or other version was used;
- what consequential changes, deviations, exclusions, corrections, or transformations occurred;
- how important outputs connect to their underlying evidence.
You may not need every element for every activity. The point is to identify the facts that would become difficult to establish if the documentation disappeared.
Documentation Should Be Proportional to Consequence
A formatting change to a graph label and a decision to exclude 15% of the sample should not normally receive the same documentary treatment.
As the potential effect of an action on the study increases, the case for stronger documentation usually increases with it. Decisions affecting eligibility, primary outcomes, data transformations, protocol deviations, analytical specifications, coding frameworks, corrections, or interpretations generally deserve more attention than routine operational choices.
| Type of Activity |
Typical Documentation Need |
Why |
| Routine action with no meaningful effect on evidence |
Little or no separate permanent documentation unless required |
Additional records may add little evidentiary value. |
| Standard procedure performed repeatedly |
Document the governing procedure plus sufficient records showing its application |
You need not rewrite the full procedure every time if its use can be established reliably. |
| Consequential analytical or methodological choice |
Document the decision, timing, rationale, and affected evidence where appropriate |
The choice may alter the findings or their interpretation. |
| Deviation from an approved or prespecified plan |
Document what changed, why, when, and any required authorization |
The difference between planned and actual practice needs to remain visible. |
| Correction of an important record |
Preserve the correction and enough history to understand what changed |
Silent replacement can destroy provenance. |
| High-risk or regulated activity |
Follow the applicable formal documentation requirements |
General academic practice cannot replace binding requirements. |
The Record Should Establish What Happened, Not Merely What Was Supposed to Happen
A protocol, preregistration, standard operating procedure, or analysis plan establishes intent. It does not by itself prove that the planned procedure was followed.
Defensible documentation therefore needs evidence of execution where execution matters. That evidence might be a laboratory entry, field record, instrument log, completed form, version-controlled script, processing log, qualitative memo, electronic system record, or another form appropriate to the research.
If the plan says one thing and the actual procedure changed, the record should preserve the difference rather than allowing the plan to masquerade as a description of what happened.
Document Enough Context to Make the Evidence Interpretable
NIH data-management guidance emphasizes that metadata and associated documentation allow users to understand how data were collected and how to interpret them. Examples include methodology, procedures, labels, variable definitions, and other information necessary to reproduce and understand the data.
This principle extends beyond datasets. A cryptic note, unexplained filename, unidentified specimen, undocumented variable code, or analysis script with unknown inputs may technically exist while providing little defensible evidence.
An adequate research record therefore needs enough context to remain meaningful after the immediate project context has faded.
Preserve Provenance Through Consequential Transformations
Research evidence often changes form. Raw observations become cleaned data. Variables are recoded. Images are processed. Qualitative material is coded. Samples receive identifiers. Analytical datasets are derived from source datasets.
NIH repository guidance treats provenance as the ability to record the origin, chain of custody, and modifications to datasets and metadata. This provides a useful principle for documentation more broadly: important outputs should not become disconnected from their origins.
You do not necessarily need to preserve every transient intermediate file. You do need enough information to understand consequential transformations and reproduce or verify them where appropriate.
Use Contemporaneous Records Instead of Future Memory
Documentation becomes less defensible when the only explanation is “I remember doing it that way.”
FDA good-documentation materials use the ALCOA framework in regulated settings: attributable, legible, contemporaneous, original, and accurate. FDA also describes additional characteristics including completeness, consistency, endurance, and availability. These are regulatory concepts and should not be converted into universal legal requirements for ordinary academic research. Still, they illustrate why timing, attribution, and preservation matter when records must establish what happened.
Record consequential activities when they occur or reasonably close to that time. If later reconstruction is necessary, identify it as reconstruction rather than creating the appearance of a contemporaneous record.
More Documentation Can Become Less Defensible If Nobody Can Identify the Authoritative Record
Keeping everything sounds safe. In practice, uncontrolled duplication can create its own problems.
Imagine finding seven datasets named “final,” five slightly different analysis scripts, screenshots with no dates, three conflicting codebooks, and an email saying “use the new version” without identifying which version that means. Nothing was deleted, yet the research history is still unclear.
Good documentation therefore also requires organization, naming, version control, relationships among records, and identification of authoritative files. Quantity cannot substitute for provenance.
Some Evidence Requires Stronger Documentation Because the Stakes Are Higher
The appropriate burden of documentation increases when errors or disputes could have substantial consequences, when procedures are difficult to reproduce, when records support regulatory or safety decisions, or when applicable policies explicitly prescribe documentation.
FDA requirements for regulated pharmaceutical activities, for example, demand substantially more formal documentation controls than many ordinary academic projects. Researchers should not imitate regulatory paperwork merely to appear rigorous, but neither should they use general academic norms when binding requirements apply.
A Defensible Record Does Not Mean a Record That Guarantees You Will Win an Argument
Documentation can establish what you did without establishing that what you did was correct.
A contemporaneous record may show that you used a particular exclusion rule, transformation, or statistical model. Another researcher may still conclude that the choice was methodologically inappropriate. That is a scientific disagreement, not necessarily a documentation failure.
Defensible documentation
Provides reliable evidence of what happened and the basis for consequential actions or decisions.
Defensible methodology
Concerns whether those actions and decisions were scientifically appropriate.
You need both. Meticulous documentation cannot rescue poor methodology, but poor documentation can make sound methodology surprisingly difficult to demonstrate.
The Research Record Becomes Especially Important When Work Is Questioned
Under U.S. Public Health Service research-misconduct regulations, destruction, absence, or failure to provide research records adequately documenting questioned research can constitute evidence of misconduct when specified evidentiary conditions are met. This does not mean that every missing file implies misconduct. The regulation expressly establishes conditions that must be demonstrated.
The broader lesson is narrower and useful: once research is disputed, documentation that once seemed administrative can become primary evidence of what actually happened.
Watch Out
Do not “improve” an incomplete historical record by silently recreating missing documentation. Preserve what exists, distinguish later reconstruction from contemporaneous evidence, and document the basis for any reconstruction. A polished record with a false chronology is less defensible, not more.
04 · A Practical Example
How Much Documentation Would Be Enough to Explain One Published Number?
Hypothetical Example
A published sample size of 742
A study recruited 800 participants, yet the main analysis reports 742. Three years later, a collaborator asks how the team arrived at that number.
Source evidence
The preserved source dataset contains the 800 collected records and stable participant identifiers.
Eligibility documentation
The protocol defines eligibility criteria, while a dated log documents eight participants later found to be ineligible.
Quality decision
A contemporaneous project record documents why 21 duplicate or invalid submissions were removed using predefined checks.
Missing-data rule
The analysis plan and code show why 29 additional records did not satisfy the requirements for the primary model.
Reproducible transformation
The cleaning and analysis scripts connect the 800 source records to the 742 observations used in the reported model.
Published result
Running the archived analysis on the documented analytical dataset produces the reported sample size and principal output.
The team does not need a narrative describing every click made during data cleaning. It needs sufficient evidence to account for the consequential transition from 800 collected records to 742 analyzed records.
That is the proportionality principle in practice.
07 · A Quick Checklist
Can Your Documentation Establish What You Actually Did?
For each consequential research step, check:
Can the records establish what actually happened rather than only what the protocol said should happen?
Can important actions and entries be attributed to the responsible person or system where relevant?
Were consequential activities and decisions documented close to the time they occurred?
Can source evidence be connected to cleaned, coded, transformed, or analytical versions where those transformations matter?
Can you identify which protocol, instrument, dataset, script, or other version was actually used?
Are consequential exclusions, deviations, corrections, and analytical decisions documented sufficiently to understand their effect?
Can important reported results be traced back through the relevant analysis to their underlying evidence?
Have unnecessary duplicates and ambiguous “final” versions been controlled so the authoritative record is identifiable?
Have you verified any documentation requirements imposed by your institution, funder, sponsor, ethics process, contract, regulation, or other governing authority?