Well it is a rank - a simple bayesian avg of number of mentions together with karma of the user. But as someone below mentioned it kind of boils down to the karma of the user who mentioned the book. A good post about that is - http://fulmicoton.com/posts/bayesian_rating

Why not use upvotes of the post instead of user karma? Having the community upvote the recommendation in context is more important to me than how much karma the user has. I find that some of the most insightful book recommendations come from people that are new to a community or lurk until they see an opportunity to contribute.

Tried that but there were few who had upvotes, probably a combination of the karma and upvotes would work well.

Sounds like the problem is that your sample size is too small. What happens if you expand the timeframe out to a month? Or a year? That should give you books that are cited repeatedly, not just the books which have recently been mentioned by high-karma HN users.

Agreed. Monthly seems like a much more natural timeframe for book recommendations than weekly. Trailing 30 days would probably be the best for the homepage. Then allow browsing by calendar month.

You can rank by upvotes and use karma as the tiebreaker.

Having/not the numbers seems like the sort of thing that might be A/B tested. Though there might never be enough results to be statistically meaningful, aggregated anecdotes are probably more informative than unmeasured hypotheses.

You just boosted the ranking of "Palm OS Programming Second Edition" again.

Exactly what I thought (including the book you picked :D)