But maybe this isn't knowable until you do it. Sure, you want to have a model that you believe decently well that indicates that the distribution of worthwhileness-of-plausible-outcomes is worth sampling from. But if questions like this were always obvious, decision science would be a lot easier, wouldn't it?

There are some versions of the book that I expect I would like reading; others that I probably wouldn't. But if it's hard to understand the shape of the target-audience distribution, you might be able to understand the shape of the distribution of books-Cassie-wants-to-write, and after throwing away the tails of that, try to find an audience that is somewhere near the peak. I don't think you do that by asking a general audience, like on Medium. Rather, when you've got your own ideas narrowed down, because you have high confidence that you can and want to do a book-like-that, then you search for an audience. Discerning an audience can be a tough measurement and/or search problem, so it's advantageous to narrow the bounds of what you have to search / measure.

Also, it's worth noting that some books have an immense impact without having huge sales, because they form a critical part of the intellectual development for people engaged in high-value work. Decision intelligence can very much be in that vein. If only three people read the book, but they get a lot out of it and then end up running Facebook, Google, and Microsoft, well, then, job well done.