I want to reiterate here that my goal is never to discredit or point fingers. We all make mistakes (I regularly do myself, and back in my student days I made many that were laughable in hindsight). And I totally agree that "correctness" of a model has many shades of grey. (1/5) https://twitter.com/ditlevbrodersen/status/1381481290843762688

But on that spectrum lies "very dark grey" and "very light grey" - and we need to be pushing our models into the latter regime as much as possible.
I struggle every time with how much (if anything) to say publicly about issues with any given model, (2/5)

... and I only speak up when I think it's important. I'm not aiming to ruin anybody's career - but silence in the face of a systemic problem just leads to perpetuation. Ultimately I need *some* illustrative example. (3/5)

It's easy (and, in my opinion, wrong) to point fingers at an individual modeller and say, "you should have been more careful". But what we have here is a pattern of many structures from many different authors with clear errors yet excellent validation statistics... (4/5)

That says two things to me:
(a) we need better validation metrics (or wider adoption of lesser-used existing metrics); and
(b) we need our building and refinement tools to be more pro-active about communicating problems highlighted by these metrics in easily understood ways.