On explainability in scientific reasoning systems

It is tempting, in an era of capable models, to accept useful answers without asking how they were reached. In science, that temptation is dangerous.

A recommendation a scientist cannot interrogate is a recommendation a scientist cannot responsibly act upon. This is why we hold that every conclusion a reasoning system reaches should be traceable to the evidence that supports it — not as a feature, but as a principle.

Explainability is slower to build and harder to maintain than opacity. We consider it non-negotiable. The credibility of computational science depends on it.