> Think about a restaurant: the food has been bought and will rot if not sold This is a good point to raise, though I think we can refine the meaning of "margi

> Think about a restaurant: the food has been bought and will rot if not sold This is a good point to raise, though I think we can refine the meaning of "marginal" to preserve a meaningful distinction between, say, restaurant meals and software: Although the ingredients for making the meals have already been purchased, so there is no direct causative link between an additional customer arriving and additional ingredient cost, there remains a statistical link: Ingredient purchases are done based on a statistical model of what number of customers, with what preferences, are likely to arrive in the near future. I would very roughly characterise this as: There is some duration x such that if n more customers than expected show up during the next x days, n additional portions of ingredients will be purchased sometime later (in anticipation that this increased demand is now "the new normal"). The above could do with improving, but what's important is that software (and media) do not have any such statistical link: If 100 people buy your program this month, 200 the next month, 300 the next month, there is no additional cost outlay for you to make in order to meet the requirements of selling an (estimated) 400 copies the following month.