It has been more or less steadily increasing since May 2003.
The last six months are all at or above the linear fit, suggesting
that the death rate might even be increasing faster than linear.
This trend is robust: there are fluctuations above and below the best linear fit, presumably related to events like the capturing of Saddam Hussein, the alleged transition of power to an Iraqi puppet government, and so on, but the average trend shows no sign of slowing down.
Moreover, the numbers of deaths in the last six months are all very close to the linear fit or well above the linear fit: this suggests that the death rate might be caught in a positive feedback loop and have entered an exponential growth phase. For example, the more US soldiers are killed, the more desperate the leadership like Cheney and Rumsfeld become, and the more nervous become the soldiers, so they make more mistakes, and therefore more and more soldiers are killed, which in turn causes more nervousness and errors still, and the positive feedback loop takes off.
Linear and exponential growth have very different implications.
The present linear trend suggests that the total number of US military dead in Iraq will only be about 5000 by the end of 2007 (prediction: 5500 by end of 2007 - this assumes that the Cheney-Rumsfeld regime does not collapse earlier than this).
Exponential growth would predict many more, 10,000, or 100,000, but saturation effects would most probably occur before such large results could be attained. For example, unless the draft is reinstated, the total troop numbers cannot sustainably go much above 150,000.
dying like flies