CHEMOLYTICA
InsightsINS 001

Rerun Your Own Analysis Before Anyone Else Does

Why the most useful habit for a computational chemistry student is deleting every output and running the notebook again.

Summary

A result you cannot regenerate is a result you cannot defend. This note argues for one simple habit: before you trust a number, clear every output and rerun your analysis from the raw data, top to bottom.

Every thesis notebook I have seen, my own included, accumulates history. Cells are run out of order, a variable is redefined halfway down, a figure is regenerated after the data underneath it has changed. The notebook still looks finished. The numbers in it may no longer come from the code in it.

[A paragraph describing a concrete moment when this went wrong, or nearly did.]

1What reproducible means here

In this note, an analysis is reproducible if someone else, or you in six months, can start from the raw data and the code, run everything in order, and get the same figures and the same numbers. Figure 1 shows the shape of it: one direction, no hidden steps.

Raw data Notebook Figures Result rerun from scratch, top to bottom
Figure 1

A reproducible analysis is a straight line from raw data to result that can be rerun, unchanged, at any time.

2The habit

Before you report a number, restart the kernel, clear every output, and run the notebook from the first cell to the last. If it fails, or if a number changes, you have found a problem while it is still cheap to fix.

jupyter nbconvert --to notebook --execute analysis.ipynb

The command above does exactly that from a terminal: it runs every cell in order, in a fresh session, and saves the result as a new notebook.

Key idea

If the notebook does not run from top to bottom in a fresh session, its outputs are a record of what you did, not evidence of what is true.

3What it costs, and what it saves

[A paragraph weighing the minutes spent rerunning against the hours saved when a reviewer, adviser or future you asks how a number was obtained.]

[A closing paragraph with one concrete recommendation for the reader.]

Further reading

  1. [Author, A. Title of the reference. Journal Volume, pages (Year).]

  2. [Author, B. Title of the reference. Journal Volume, pages (Year).]

Cite this note
Solomon, M. B. (2026). Rerun Your Own Analysis Before Anyone Else Does. Chemolytica, INS 001.

Found an error, or disagree? Let me know.

REVISIONS12 November 2026 · First published.

FILED UNDERreproducibility · Jupyter notebooks · research practice

More from the catalogue

  1. RES 001[Research title]Research
  2. TL 001[Tool name]Tools
  3. RES 002[Research title]Research