Reproducibility in Practice: Data, Code and What Another Lab Can Rerun
A result that cannot be regenerated from raw data and a script is a claim rather than a method. This stop covers deposition, versioning, and the difference between reanalysing the same data and repeating the experiment.
What a learner can do afterwards
- Package an analysis so a stranger can regenerate a stated figure from the raw data
- Distinguish reproducing an analysis from replicating an experiment, and say what each tests
- Choose an appropriate repository and metadata standard for a stated data type
1 · Read
A result is reproducible when a stranger regenerates your figure from raw data plus your script, asking you nothing. Deposit both under version control with tool versions pinned and the exact run command written down. Anything less is a claim, not a method.
Your plot came from raw.csv plus clean.py at version 1.3 under Python 3.11. The package lists all three plus the one run command. A stranger runs that command and gets your figure back, bit for bit.
Reproducing re-runs your analysis on your data, testing code and claims. Replicating repeats the experiment fresh, testing the phenomenon itself. One checks honesty, the other checks nature: never confuse them.
Deposit by data type: sequences to a sequence archive, tables and code to a general repository, each with standard metadata saying what, who, when, and how. Metadata is what keeps files findable and rerunnable years later.
Pin data, code, and command so a stranger rebuilds every figure alone.
2 · Watch
Take it off screen
Where it sits
Learn first
This opens up
Nothing builds on it yet.
Where this leads
8 questions wait behind this lesson, each with its answer explained. Every answer feeds the sky: stars light as they are learned, and dim when it is time to come back.