LLMs basically can't do arithmetic directly, trying to get them to do so is a skill issue. Most models can and will happily write and execute code to do that work instead.
What I described is a setup that DOES have pandas & python access and uses it heavily to figure out the excel files. Which is neat, but the output is still wrong
I'm sorry, what are you saying exactly? That a model like GPT 5 can't read an excel file and correctly sum rows based on filters? I find that incredibly hard to believe. GPT 4 from 2 years ago could do this with no issue.
