We've been hearing a lot about improved security conditions in Iraq and Baghdad. And there's been a lot of talk about percentage declines of deaths of one sort of another. But one thing we're going to be listening for tomorrow and Tuesday is some actual data, some numbers. And so far there's been very, very little in the way of hard numbers coming out of the US command in Iraq.

In fact, as near as we can tell, a lot of the numbers, the key metrics about what's actually happening on the ground remain classified.

And not just the numbers themselves.

A few days ago we flagged Karen DeYoung's piece in the Washington Post about critics questioning the alleged decline in violence in Iraq. And one key point she focused in on is the methodology that the folks in Baghdad are using to derive their numbers. Is it really true that it matters how a person is shot (in the front of the head or the back) for whether or not they get counted? Is it true that we're not counting Sunni-on-Sunni or Shia-on-Shia deaths? Or even killings by the folks we're now allied with in al Anbar province?

The best we can tell the methodology Petraeus's staff is using to tabulate the numbers also remains classified.

In other words, it's not just a matter of getting the numbers from Petraeus and his staff and deciding whether you believe them or not. They won't even tell us what the numbers are -- let alone how they came up with them. All they'll say is that they're very good. Or in some cases that there's X percentage drop over the course of the surge. Or an isolated number here or there.

But actual hard numbers? Going back over the last couple years? For some reason we're not allowed to see those.

Perhaps Petraeus and Crocker will sit down tomorrow and share all this data as part of their presentation. But if not, this is the issue. What possible security need is served by keeping this data secret? And with all we've been through, can anyone believe that if the numbers were solid that we'd wouldn't be being buried in data right now?

Something's wrong.