rehashing deep learning papers making spurious claims doesn’t make it a fact, informer and other BS papers were shown to be outperformed by naive forecast.

Transformers are routinely outperformed by simple models including linear models.

https://medium.com/@valeman/-86655805a676

Amazon team did not see any performance carried out to settings outside of Amazon and Zalando uses boosted trees as well. Amazon has billions of clean data points and a large team of PhDs who were building it for over decade. And yet in their paper Forecasting with Trees (that you conveniently omitted) they admit that for most companies boosted trees outperform deep learning.

If you are “certain” they can show “remarkable improvements” what stops you proving it in either forecasting competitions or by publishing peer reviewed papers?