I found it misleading and useless. Maybe the author should use CMM when writing articles. What they're quoting is a number that refers to programs in development, not finished programs. I guess it could be argued that XP is still in development, but then again it could be argued that OS's shouldn't even be examined under the umbrella of TSP & CMM, as they aren't really single programs any more. That 40,000 errors bs is more easily equated to a beta of Longhorn.
QUOTE
Corrections and Clarifications
In "Watts Humphrey: He wrote the book on debugging" (Industry Insider, May 9, in some editions), the graphic "A Software Cleanup" showing defects per 1,000 lines of code, without TCP and with TCP, respectively, should have been 120 and 50 during program development, and 1.05-7.5 and 0-0.1 in a finished program.