When the calculated columns are at a group level, I will usually calculate them as expressions in the objects. So if I have a chart when I want to display a monthly total of sales, I would add a dimension of Month and an expression of sum(Sales). Then if I decide to add Customer as another dimension, I'll get my totals by month and customer without any script changes. I could also use a pivot table to have my total for each customer within a month, and then a total for each month, or a total for each month within a customer, and then a total for each customer. All of that could be done by just moving dimensions around rather than by changing the script.
Something like Profit, though, I might well calculate in the script, as it could be calculated at the lowest level of detail. So my script might include "Revenue - Cost as Profit," so that I could make my expressions just say sum(Profit) instead of sum(Revenue - Cost).
In regards to the tags, it sounds like an example would be for a user to go in and tag the customers and regions of the country that they most commonly analyze. They could use multiple tag collections to create different sets of data to analyze, and then select which set to use for analysis. Is that sort of what you're talking about? If so, that particular example sounds like bookmarks.
Or would an example be that we're looking at a bunch of orders, and the user is going in and typing notes on each order, like "customer over credit limit" or "verify that material is being produced quickly enough". And then they could just select a value in this column to, say, find and filter down to all of the cutomers that they had tagged as being over their credit limit? You could sort of do that with bookmarks, but that might lend itself better to... oh, what are they called. There's a way to add fields to a table that are enterable, which might work here. I haven't used it in any real applications.
