The Sunlight Foundation is pretty darn good at chasing down pieces of government data -- contracts, political donations, lobbying expenditures -- and putting it back together again in ways that make sense. Along with other open government groups, it's pushed U.S. agencies to make that data more available, and they have.

What price a buggy Web site? (Mike Segar/Reuters)

Nevertheless, on the question of exactly how much a project like cost to produce, Sunlight says it's at a loss.

In a blog post Tuesday, Kaitlin Devine explains why: Even with slick tools like and the FederalITDashboard, contracts like the one CGI Federal got to produce aren't actually awarded on a project-by-project basis. Instead, they're wrapped up within "indefinite delivery, indefinite quantity" vehicles with a bunch of different task orders that are vaguely labeled by what they're paying a company to do. Manually counting up the task orders for yields about $70 million, which is substantially less than the $93.7 million that the Government Accountability Office says had been obligated to CGI back in June.

Why doesn't she just trust GAO's numbers? I asked.

"I don't have any reason to suspect GAO's obligation numbers are wrong, only that they're easily dated due to the static nature of their reports," Devine responded in an e-mail. "If an additional option was exercised or another task order was created, then the numbers might not be accurate anymore. Additionally, not every IT project mandates a GAO report about its implementation. My main point is that this is exactly the kind of information that USASpending claims to make it easy to find, but in reality it falls very short."

The problem is even more acute with larger, more nebulous expenditures, like wars:

When discussing spending data I'm often asked, "How much did we spend on the Iraq war?" This question is pretty much unanswerable without making some subjective judgments--spending just isn't categorized that way. It's categorized by the Joint Strike Fighter program, or secondary education benefits for veterans, or foreign aid to Iraq. Reasonable people can disagree about what costs should be rolled into the general category of "Iraq war spending."

There is a proposal out there to fix this problem: The DATA Act, introduced earlier this year, would make all that information more easily categorizable. It might not fix all the other problems with how the government buys IT services. But at least it would make it easier to understand how much our blunders are costing us.