I'm quite sure that this error is related to some update as, the same application works fine within another machine with the same R version (4.0) and the same Shiny version (booth are MACos).
Do some of you know how to trouble-shooting this?
After some additional troubleshooting seems that the problem is related to predict() function with an xgboost model trained with tidymodels. I don't know if it's an error of tidymodels,xgboost or predict but now it's more clear where to investigate.
The lates version of hardhat did contain a change around this composition argument, controlling the datatype that gets passed around internally.
This seems to brake the backward compatibility with the models trained with older hardhat version.