Not all research software is a tangled mass. I have extensively worked as a "quant" (before the term was popular) for math, medical, network, media, and physics

Not all research software is a tangled mass. I have extensively worked as a "quant" (before the term was popular) for math, medical, network, media, and physics researchers as my side gig for decades. I'd say about 1/3 of the home brewed research software is constructed with fairly reasonable assumptions, the authors are scientists after all, and I am able to grow their basic setup into a framework they intimately understand and prefer to use. More than once I've found brilliantly engineered software not unlike what I'd find at a pro software development firm.