I redid the equation I use to decide how many of an item are likely to be sold per day.
The first version was okay for high-volume items, but was bad at identifying items which sell less than one unit per day, which I don't really want to bother with. When I ran the trade finder many of the results were low-volume items that it thought I could trade and make a huge profit on, while actually doing so would have involved days of 0.01ing to sell a single unit.
The second version had the opposite problem -- it successfully ignored low-volume items, but thought it would be a super good idea for me to haul four trillion units of tritanium from jita to amarr.
So I put away the automatic curve fitter and did the variables by hand. Version three successfully ignored low-volume items, and thought jita moved about 15 billion units of tritanium per day. Huzzah.
Version four is version three with a flatter curve to make mid-high-range numbers more accurate. When I ran it for the first time the list of most profitable trades contained largely mid-range production stuff like ice products and PI or reaction materials, so I sighed and went to the game to get real numbers to compare with eve-central values... and it turned out the equations's numbers were correct. So I'm calling it good.
Also, I hadn't included taxes before, which resulted in the "profit" numbers it returned being inaccurate. If you move 4 billion isk worth of stuff for a before-tax profit of 150 million, the tax eats half of it. If you can move 1 billion for a profit of 149 million, that's a much better trade.
I guess this is the point where I do a trial run and see what happens.