Reproducibility: environments, containers, seeds

An analysis nobody else can re-run is just an opinion. You make it re-runnable by pinning the environment:

• a virtual environment (python -m venv .venv) gives the project its own isolated Python and packages, so it doesn't depend on whatever happens to be installed globally;
• a requirements.txt with pinned versions records exactly which libraries it needs, so a colleague installs the same ones.

Isolation plus pinned versions is the core of reproducibility: the same analysis runs identically on someone else's machine.

Two more layers complete the picture. A fixed random seed makes any randomised step — a train/test split, a shuffle, a sample — produce the same output every run, so results are repeatable. And a container (Docker) packages the operating system, Python, and libraries together, so "it works on my machine" becomes "it works on every machine."

Reproducibility is a discipline you practise rather than a single tool you install — but together, graph reasoning plus a pinned, isolated, seeded environment are exactly what a trustworthy project rests on.
algorithms” and leads into “The capstone and the full pipeline”.*