Reproducible Analysis Workflows

Analysis projects anyone can rerun: a standard layout, git, pinned environments, read-only raw data and one-command rebuilds, the routines that keep them honest, and replication packages for papers and audits.

Free · Mac (Apple silicon) and Windows · v1.0.5 · Other platforms

Then open the Marketplace in the app, search for Reproducible Analysis Workflows, and click Install. It goes into your own files, and works offline from then on.

What it installs.

1

Pillar

1

Area

50

Projects

201

Starter tasks

📊 Data & Analytics

Reproducible Analysis Workflows

Structuring notebooks and analysis code with version control, environments and documentation so results can be rerun and checked.

  1. Rerunning one of your own past results from scratch
  2. Standard folder layout for analysis projects
  3. Git repository for every analysis
  4. Pinned software environment with a lockfile
  5. Read-only raw data with checksums
  6. README that lets a stranger rerun the analysis
  7. Credentials kept out of analysis code
  8. Random seeds and session details recorded
  9. Single command that rebuilds every output
  10. Weekly commit and push routine
  11. Restart and run all before any notebook is shared
  12. Monthly rebuild on a clean machine
  13. Analysis decision log beside the code
  14. Automated checks on every pull request
  15. Peer review rota for analysis pull requests
  16. Quarterly dependency update with output comparison
  17. Monthly snapshot of the data extracts in use
  18. Close-out routine for finished analyses
  19. Annual rerun audit of past analyses
  20. Git skills for analysts beyond commit and push
  21. Literate analysis documents in Quarto or R Markdown
  22. Notebook hygiene with paired scripts and stripped outputs
  23. Unit tests for data cleaning and transformation functions
  24. Containers for analysis environments
  25. Caching expensive steps without stale results
  26. Parameterised reports for repeated analyses
  27. Numerical differences between runs and machines
  28. Structured course on reproducible research practice
  29. Choosing an environment manager for the team
  30. Refactoring a long notebook into scripts and functions
  31. Internal package for functions used across analyses
  32. Deciding where analysis data is stored and versioned
  33. Configuration file for paths, dates and parameters
  34. Figures generated entirely from code
  35. Separating exploratory work from reported analysis
  36. Replication package for a journal submission
  37. Handover before leave or a role change
  38. Independent rerun before a high-stakes report
  39. Freezing thesis analyses before submission
  40. Reproducibility workshop for colleagues
  41. Inheriting an analysis nobody can rerun
  42. Shared reproducibility standard for a research group
  43. Onboarding a new group member to the workflow
  44. Software engineer moving into research analysis
  45. Reproducible work on confidential data
  46. Analyses on a shared computing cluster
  47. Provenance record attached to every output
  48. Independent reimplementation of a key result
  49. Audit trail for analyses in regulated settings
  50. Long-term archive that still runs in ten years

Where it starts.

Each project arrives with its purpose, milestones and first tasks written out. Here is how the first three read.

Rerunning one of your own past results from scratch

Pick one figure or table you shared at least six months ago and try to regenerate it on your current machine using only what is saved in the project folder. Most people hit a missing file, a changed package or a forgotten manual step within the first hour. That list of failures is the honest starting point for everything else in this area.

First tasks

  • Choose one chart or table you shared more than six months ago
  • Copy the original output somewhere safe for comparison
  • Rerun the analysis on your current machine without asking anyone for help
  • Write down every step that failed or needed a guess
  • Compare the regenerated output with the original and note any differences

Standard folder layout for analysis projects

When every project puts data, code, notebooks and outputs in different places, nobody, including you in a year, can tell which script made which file. One layout used everywhere, with raw data, processed data, source code, notebooks, outputs and docs in fixed folders, makes any project readable in minutes and lets tools and templates assume where things live.

First tasks

  • Sketch the folders your last three projects actually needed
  • Write a one-line purpose for each top-level folder
  • Turn the layout into a template repository or project generator
  • Move your current project onto the layout and fix broken paths

Git repository for every analysis

Analyses kept as folders of files named final, final2 and final_really lose the history of why a number changed. Putting each analysis under git from day one, with an ignore file that keeps out data, outputs and secrets, gives you a timeline you can return to and a single place colleagues can clone.

First tasks

  • List every analysis folder on your machine that is not under git
  • Create a shared ignore file for data, caches, outputs and credential files
  • Initialise a repository for each active analysis and push it to a remote
  • Write each new commit message so it states the reason for the change

Make it yours.

Install Reproducible Analysis Workflows, then tell Aurum the date you're aiming for. It checks whether the plan fits the weeks you actually have.

Free · Mac (Apple silicon) and Windows · v1.0.5 · Other platforms

Blueprints are general information to help you organise your own work. They are not medical, financial, legal or other professional advice. See the Terms.