gardner said:

I think we differ in our assumptions about the economics. 

i agree. wholeheartedly.

My thinking is that having a huge back-catalogue of titles is mostly a marketing gimmick. 

um, i'm not sure how that "gimmick" is supposed to work. having more titles in your catalog isn't going to make me any more (or less) likely to order a specific book from you. if i want a particular book, i'll order it. if i don't, i won't. if i don't want it, i'm not sure why your having a million -- or two million, or twenty million -- other titles would make me decide to change my mind and suddenly want it.

I anticipate that perhaps 1% of the available titles would ever get ordered by anyone. 

i anticipate that 50% of the titles will be ordered in any year. but the vast majority of those might only be ordered _once_. and i'd assume that, over the course of 5 years, every title will end up being ordered -- a couple of times, at the very least...

That means that 99% would not have to have any investment put into them at all in order to fill a slot in the back-catalogue. 

i agree, that -- until a specific book is ordered -- it won't be fixed. or maybe i don't. it would be nice to think that umichigan itself is thumbing through these scansets to identify any problem children. (and we could certainly hope that students will be using the scans, and bringing glitches to the attention of people who can fix them.)

In my vision, time spent on doing *anything* non-mechanical to a book is, on average, 99% wasted.

i guess it depends on what you would consider "non-mechanical". most of the scripts to handle the scans -- deskewing, centering, noise-removal -- probably don't need any human attention at all. but every single one of these scansets has garbage scans in them, such as excess blank-pages in the beginning and color-correction scans at the end, not to mention library-style check-in pages, etc. so it will absolutely be necessary to have a human thumb through all of the pages, to eliminate these unnecessary pages and also to ensure that odd-numbered pages always occur on the recto and even-numbered pages always occur on the verso, and that there are no missing pages or page-spreads, and no duplicated pages. this quality-control pass only needs to be done _once_ on a book, but it would be stupid to the point of ridiculous to go without it... (and as it's only a minute or two of work, it's economically feasible.) there might be some problems which would evade quality-control, such as a page with a few characters cut off at the end of each line, but for the most part, i'd think a simple pass would catch most stuff. if you look at any particular scanset, you should be able to see immediately why such a quality-control pass would be necessary.

Better to let the market help identify the 1% of valuable titles and concentrate your scarce fixup attentions on those.

any book for which an order has been received is "a valuable title".

Assuming that even without manual review 99% of your titles are actually fine as far as scan quality and completeness goes

none -- 0% -- of the scansets are "actually fine" right out of the box.

you are actually quite likely to get away with a 1 part in 10,000 return rate.  And customers might not actually *read* the whole book either. If 50% wind up on shelves unread, or have problems that are annoying, but that don't result in returns, the return rate drops to 1 in 20,000.

i don't think amazon thinks of quality-control issues the way you do.

And if, in the end, they have to ship two copies to a customer, they only have to do *that* once.  After a return they can go and fix up that exact book, and never have to double-ship that one again.

i think you underestimate their commitment to a quality product... -bowerbird