The only question is when, and how much of it you get back. Eight months into a project you need to regenerate Figure 3 with one more data point. There are four scripts with promising names, two of them called something like analysis_final, and neither produces the numbers in your draft.
Paperback wrap: back cover, spine, and front cover.
Nothing was deleted. Nothing was corrupted. You simply have no record of what you did, and that turns out to be the same thing as not having done it.
This is the practical subset of reproducible research, written for a student with a project and a deadline. Not Docker images and continuous integration. Six habits, none taking more than an afternoon, that between them close every way a project falls apart.
What You Will Learn
Why raw data is read-only, and how to enforce it at the filesystem level
Git in seven commands, without the branching mythology
One command that rebuilds every figure and number from raw data
The environment file everybody skips, and why silent version changes hurt most
Assertions and seeds: the tiny subset of testing that pays for itself
A project template, a Git card, and a checklist built to be photocopied
Inside the Book
The downloadable editions include the complete chapters and toolkit listed below.
Why Work Disappears
The Week You Lost
What Reproducible Actually Means
The Five Artefacts
Files and Records
Naming Things
A Project Layout That Survives
Raw Data Is Read-Only
The Digital Lab Notebook
Backups, and the 3-2-1 Rule
Version Control
Git Without the Mythology
The Seven Commands You Need
Branches, and Whether You Need One
What Never Goes in a Repository
Code That Runs Again
Scripts, Not Sessions
One Command Rebuilds Everything
The Environment Is Part of the Result
Seeds and Determinism
Notebooks: Where They Help and Where They Hurt
Analysis and Outputs
Every Figure Made by Code
Numbers That Trace Back
Checking Analysis Code
Documenting for Your Future Self
Sharing and Handing Over
The README That Makes It Usable
Licences, DOIs, and Archives
The Handover
Toolkit
The Project Template
Git Command Card
The Reproducibility Checklist
README and Statement Templates
Environment Recipes by Language
Ten Failures, and What Would Have Prevented Them
Two PDF editions are included
Both layouts are free to download from this page. Use the 6 × 9 edition when you want the paperback reading experience or a compact printout. Use the A4 edition for a larger screen page and easier home or office printing.
A4 screen edition
PDF · 134 pages · 471 KB
A larger page for reading on screen, and the easier one to print at home or in an office.
I Build Tools That Pay for Themselves And Give Away the Ones That Can’t
Everything here is quick to set up, light on your pages and built to do one real job well. I only charge for the tools that can make you money. The rest are free.