Author of "Leprechauns" here.
I'm not sure I follow you in the above sentence. What happened time and again in my investigations was that I would find a citation (in McConnell or in Boehm or elsewhere) that was metaphorically accompanied with the statement "this here is a solid piece of land in the swamp".
When I got there, however, what I found was, in fact, just more swamp.
It takes a higher standard of proof to demonstrate that something is solid ground than to demonstrate there's a flaw in it. That may be unfair, but it's why scientists need a lot of training.
To take just one example, Graham mentions the Hughes Aircraft study, cited (by McConnell and others) in support of the usual "exponential" curve for defect cost (as "Willis, Ron R., et al, 1998").
When you actually read that paper - which is both one of the easiest to obtain and one that has the more detailed data about what was studied and how - and look for the raw data, you find numbers that do not obey an exponential law OR a monotonic increase. For one column for instance the numbers in fact vary within a narrow range .36 to 2.00 with two maxima at the "Coding" and "Functional test" phases, dropping off before, between and after. In a different column the costs vary only by a factor of two between Unit Test and System Test, by a factor of less than four between Code and System Test. The exponential rise is generally not true for the pre-1991 period; some post-1991 measurements get closer, but many do not.
It's hard for me to see this evidence as doing anything else but undermining the original claim of the exponential rise as a generally reliable regularity in software development.
Given this, I don't think that citing this paper in support of the claim is appropriate.