What does it mean for a computational experiment to be reproducible?

Why

Package input identifiers and checksums, raw versus processed data, exact code commits, software versions, parameters, random seeds and execution instructions. Use containers or environment lockfiles where appropriate, but remember they do not automatically preserve external databases or nondeterministic hardware behavior. Define expected outputs, tolerances and test fixtures.

In a sequencing workflow, changing a reference genome build or annotation release can alter feature counts even with the same script. Capture those references and transformations. If data cannot be public, explain access constraints and supply synthetic tests where useful.

Under National Academies terminology, computational reproducibility uses the original data, whereas a fresh laboratory experiment is a replication. A complete archive still does not demonstrate that the scientific inference is correct. It demonstrates that the process can be inspected and rerun.