Jens Egholm Pedersen works in neuromorphic computing and spends a lot of his time on software. Colleagues keep asking why — software is a time sink, something to rush past on the way to results and papers. His answer is that in computational science, the software is the result.

The argument starts from Wikipedia’s definition of science: systematic, testable, organized knowledge. Then it asks whether a reader can actually test a typical arXiv paper. Usually not directly — you have no access to the knowledge inside. Drawing on Craik’s idea of inner models, he argues a paper is science only insofar as it lets the reader improve their own predictive model of the world. Code is how a computational model gets encoded and shared, so hiding it breaks the method rather than just annoying readers.

From that follow two properties he treats as the checklist:

  • Reproducible — the artifact must be executable and modifiable, not merely re-runnable by its authors
  • Reliable — results must be consistently trustworthy, not flaky and partial

The thought experiment that makes it land: nobody accepts a physics paper that says “our equations predict X, but we’ll keep the math to ourselves.” Accepting computational science that hides its methods is the same move. And since results depend on the software, a bug is a scientific problem, not an engineering one — he cites retractions in climate science and immunology, and notes researchers rarely have time to audit their dependency trees. If the software is wrong, the science is wrong.

Open source is the closest existing embodiment of both properties, and it carries a second benefit: imperfections end up on public record where they can be fixed, in the same way scientific understanding gets corrected. He is honest about the limits — IP and security concerns, residual bugs, stability issues — and argues the difference is that these problems are visible.

His vision section is the concrete part. Every result instantly reproducible, with the analysis running in the browser inside the authors’ preserved environment. Climate models maintained like a Wikipedia page, so a turbulence bug found in Kenya propagates worldwide instead of sitting in one lab’s silo. Discovery compounding because work is runnable and modifiable, not just readable. Trust improving because models join the conversation rather than only their conclusions. None of it is speculative technology; he points out every piece already exists.

What he asks for in practice:

  • Share and document code from day one — a paper without code is less scientific
  • Write stable code on reproducible environments; he rates NixOS above Docker and Conda for guaranteeing the same execution years later
  • Build on existing tools instead of writing your own, especially in small fields like neuromorphics where a Nature paper argues open source is key to scaling
  • Value software contributions in academic promotion

The takeaway is that open source is not a licensing preference bolted onto science — it is the reproducibility mechanism. Software is executable mathematics, and treating it as second-class output is what the publish-or-perish framing gets wrong.