Front-end cartographer. Research Engineer in Geographic Information Science at CNRS & UAR RIATE. Author of Mad Maps (2019), Practical Handbook of Thematic Cartography (2016, 2020, 2025) and Cartographia (2025).
Sharing the sentiment here to some extent.
What helps for me, is requiring search for prior art and attributions to existing and related concepts. Then only describe what is not sufficently covered there.
Biggest problem to me with AI is constant threat of scope creep not confining the problem space to a very specific thing. To me the distortion in the edge regions is too high to accept this as recreation of the projection would set projected screen space pixel error target budget and absolute threshold as some acceptence criteria. In the Arctic Ocean the recreation is simply not holding up and moves stuff hundreds of km.
Also we don't know if the wikipedia version is a faithful recreation or digitization to begin with without further validation.
for this one sample map, the reference parameters seem to come from a single 1980 publication:
https://github.com/jkunimune/Map-Projections/blob/028222e91a533ef14f4ba96431117d4c28914a7f/src/maps/Lenticular.java#L440