In light of the controversy surrounding the Pentagon's Iraq numbers, hilzoy explained the larger dynamic nicely:

Here are some of the things we know about these statistics: they don't include Sunni-on-Sunni violence, or Shi'a-on-Shi'a violence. They don't include car bombings. There are unexplained changes in the figures from one report to the next. They don't seem to take seasonality into account. [...]

[T]here is an easy way to resolve these issues. If the government were to release its figures, and explain the methodology behind them, then it would be clear whether they had been cherry-picked or not.... Explaining the methodology behind their numbers is obviously the best way for the administration either to put these sorts of doubts to rest once and for all, or else to prove that they are well-founded.

It's an idea that's making the rounds.

Members of Congress urged the Pentagon yesterday to declassify its data on sectarian killings, just days before General David Petraeus, the top US commander in Iraq, is expected to report a dramatic decrease in the level of violence between the Sunni and Shi'ite sects.

Does anyone seriously believe that if the internal data pointed to legitimate progress in Iraq, the administration wouldn't release it?