In general, it would be helpful if you could provide full code snippets like your first post that allow me to load your data and reproduce your issue. If you could provide a script that loads the data and generates the sampling issues then I would be happy to run it and see what comes out.
That said, I think you want to make sure you use the log of the CANTIDAD variable like before since it spans several orders of magnitude. I also think interpreting the day of the month as as a continuous variable isn’t very sensible. For example, treating the 0-31 values as usual numerical covariates suggests that March 31st is farther from April 1st than it is from March 10th. Does that make sense for the process you are modeling? Could you perhaps create a new categorical variable that indicates whether a day is in the first week, second week and so on?