Well, except in the case of variable shadowing.

Signs we may be reaching the Plateau of Productivity

Fig. 8, where the model becomes poorly calibrated in terms of text prediction (Answers are "flattened" so that many answers appear equally probable, but below the best answer)

In many subfields, the submitter isn't even attempted to be hidden from the reviewers. Usually, even the reviewers can be guessed with high accuracy by the submitters

It's MathJax, do you have javascript enabled?

Not working in Firefox on Android either

Note that Mathematica units impose a very large runtime penalty, making them unsuitable for a lot of applications

The applications that are WL-units suitable are much more than the WL-units unsuitable ones you refer to.

It is true that I (and others) avoid using Wolfram Language (WL) units. Or try to get the computation expressions units-free as quickly as possible. But that is also a normal procedure when crafting computational workflows.

It's also quite similar to Thread and MapThread in Mathematica

As an alternative to Pydantic, check out the wonderful mashumaro: https://github.com/Fatal1ty/mashumaro

I've also played around with writing my own dataclass/data conversion library: https://github.com/hexane360/pane

Also worth noting that test pilots of experimental aircraft generally wear parachutes, at least for higher risk tests. This includes tests of commercial aircraft.

This plus make a lot of sketches describing what each variable is physically, and use geometry to derive dependent dimensions.

Libraries like BOSL2 help with placement/attachment of features: https://github.com/BelfrySCAD/BOSL2