There are several articles about the IRS's IT revamping, done by outside contractors lead (?) by CSC, getting out of hand 2 years into a 5 year schedule. They're not very in depth, e.g.,:
"Both the IRS and CSC get blamed for failing to establish an environment of trust, confidence and teamwork between themselves. In fact, the report states, the opposite is true, leading to a tense and inefficient working environment and regular finger pointing. The report also criticizes the IRS and CSC for laying down an unrealistic program schedule."
I can't find The Report either at the IRS Oversight Board Website, though, there are some other interesting looking things there.
The IRS says it can still process returns and send out refunds on time, but its dependence on the 1960s-era Assembler and Cobol computer languages makes it difficult to investigate and resolve taxpayers' problems. Finding a record using the existing system can take a week; the new system is supposed to do the job in seconds.
. . .
Paul M. Cofoni, president of the Computer Sciences unit running the project..., said "in the early part of the program, we did a poor job of defining" what needed to be done. But that was in large measure because the IRS had no records of many changes to its old system, he said, and was reluctant to approve specifications for the new system until it could be sure that it would be able to find and display all the old information.
. . .
Five years into the project, some aspects are as much as 27 months behind schedule.
It looks like they've encountered Glass's top two reasons for software project failure: missing requirements (and the inability to adapt to "finding" them re: schedule and over-all project re-jiggering), and, bad estimates (and the lack of revising the almost usefullness initial estimates throughout the projects).