Explain the route by which knowledge was produced
A methods section enables a knowledgeable reader to evaluate selection, implementation and analysis. It is neither a diary nor a list of technique names. Information matters when a decision shapes the object studied, evidence quality or interpretation.
Start from the research question and design. Explain why the approach fits and identify the unit examined. Separate planned procedure from what occurred. A reasoned deviation is more informative than a polished fictional plan.
Repetition differs across fields: rerunning a calculation, applying a coding system, reconstructing source selection or repeating an experiment. State the intended form instead of claiming universal reproducibility.
Bound materials and selection clearly
Describe the population or search space, selection route, period, inclusion and exclusion, material sources, versions and relevant access constraints. “Suitable texts were selected” is not inspectable; criteria and boundary decisions are.
For participants, appropriately describe recruitment, consent, withdrawal and achieved sample. For archives, record collection, reference and edition; for software, data format and input version. Limit personal detail to what can lawfully and ethically be reported.
Distinguish unavailable, deliberately excluded and technically lost material. These states can affect inference differently and must not disappear under “cleaned”.
Show the actual process in inspectable stages
Order stages chronologically or by a clear process logic. For each, identify input, action, output and decision point. A diagram can support this account but cannot explain why one branch was taken.
A standard procedure may be cited to an accessible specification. Describe every adaptation, translation or alternative threshold. A generic manual reference is insufficient where several valid implementations remain possible.
Date pilots and changes. If a procedure changed after results were viewed, make that visible. Exploration may be legitimate, but it must not be presented as an unchanged advance decision.
Prioritise settings by their effect
Consequential settings include thresholds, units, filters, stopping rules, random seeds, model or software version and outlier treatment. A long list of defaults without explanation produces only apparent precision.
Qualitative work may need segmentation, coding, multiple assignment and disagreement rules. Experiments may need conditions, instrument range and calibration. Literature work commonly needs databases, dates, search fields and screening rules.
Use units consistently and explain selection within a range. If a service hides a version or setting, record that it was not displayed and describe the observable function. Never invent a parameter to make the report look complete.
Report deviations and limits honestly
A transparent account also states what cannot be repeated. Proprietary services, protected data, unique events or missing versions need a justified substitute trail such as synthetic material, controlled access, export or precise functional description.
A successful rerun does not by itself prove good design; a non-identical rerun does not automatically refute a study. Evaluate documentation, appropriateness and stated scope through the criteria announced for the project.
Finish with a bounded handover test
Give a knowledgeable reader only the methods section and permitted appendices. Ask them to identify materials, central stages and three result-sensitive parameters. Record missing information as exact questions. This is a documentation test, not a full replication or effectiveness claim.
Compare the finished section with results and appendices. Sample size, exclusions, versions and labels must agree. Update methods whenever the analysis changes, and retain a concise revision reason.
The criteria log records selection; the coding guide defines qualitative decisions. Browse the research methods hub. The single foundation route is how verifiable checking works.
Finally inspect references to instruments, protocols and software. A name should resolve to the version or description actually used. Store supporting materials under stable, permitted access and state what a future reader can reasonably obtain.
Review the connections between stages as well. Every output should have an identifiable consumer in the next stage. Where material is filtered, reconcile counts before and after filtering and explain the treatment of excluded units. Stable case identifiers and a decision log help with manual judgements without duplicating protected content.
Record empty and negative outcomes. A search returning no item, a test finding no difference or a code not occurring should be distinguishable from a step that was never run. Give status, time and expected output. For later changes, record old setting, new setting, reason and affected results, then repeat the relevant steps rather than presenting the process as falsely linear.