You reopen an analysis from a few months ago and can't remember why three observations were excluded. The spreadsheet is there. The explanation isn't. Before another researcher can follow your work, your future self needs to be able to follow it.
The handbook emphasizes reproducibility and replication early in a research career. Begin with a transparent record of what you did, why you did it, and what another person would need to understand the result.
Define the task rather than relying on a label
Fields use the terms differently. Under the convention used in the National Academies' report, computational reproducibility concerns obtaining consistent results from the same data and computational procedures. Replicability concerns consistency across studies addressing the same question with their own data.
When describing a project, state whether you're checking an analysis, collecting new data, or testing a question in another context. That explanation is useful even when readers use different terminology.
Keep the path from source material to result
Record how information was collected or obtained, how it was prepared, and which steps produced the reported findings. Where appropriate, retain scripts, coding instructions, or other analysis documentation with clear versions.
A file called final-analysis doesn't explain the decisions that made it final.
Document decisions as you make them
Explain exclusions, changes, assumptions, and departures from the initial plan. Distinguish decisions made before examining results from later exploratory work. Another reader should be able to see the reasoning without guessing which version of a file you used.
Plan what can appropriately be shared
Transparency doesn't require exposing private or restricted information. Follow consent, access, institutional, and other applicable requirements. Explain limits and consider which permitted materials can still help others understand the work.
Make a replication question worthwhile
If you're conducting a new study, explain what it tests about earlier evidence and why the design is informative. Consistency, differences, and uncertainty can all matter. Don't treat agreement with an earlier finding as the only acceptable outcome.
Try the record on one result before the project grows
Choose a reported table, figure, or interpretation and identify the exact source files and steps behind it. Can you tell which version was used, what decisions changed the material, and where the reasoning is recorded? Follow the route rather than relying on what you remember doing.
If a colleague is appropriately authorized to review the work, ask them to identify where the explanation becomes unclear. Keep restricted information within approved access arrangements. A useful check doesn't require distributing material the team isn't permitted to share.
Fix the missing links while the work is fresh. That might mean a clearer file label, a note explaining a coding decision, or instructions for the order of analysis steps. The aim isn't a perfect archive with no possible questions. It's a record that makes questions answerable and helps you distinguish a documented decision from a detail everyone has been carrying only in memory.
Choose one result from your current project and trace it back through the files and decisions that produced it. Write down the missing steps you encounter. Fixing that path now is much easier than reconstructing it when a reviewer or collaborator asks later.
Adapted from The New Faculty Handbook by Zafer Unal, pp. 80-81. Terminology follows the stated convention in the National Academies' discussion of reproducibility and replicability; usage varies across disciplines.
