I have always been skeptical of the official numbers because those numbers make no sense given historical experience and motivations for fighting. Additionally, they also assume near infallibility in the US ability to correctly categorize individuals from supporter, to neutral, to inactively sympathetic for the insurgents, mildly supportive of some faction of the insurgency, active supporter of some armed non-governmental faction, to active shooter/IED emplacer.
Andrew Sullivan brings in an estimate as to how large of a salt shaker is needed to hold the proper skepticism:
Colin Powell's former Chief of Staff Lawrence Wilkerson estimates that the US administration has arrested between 30,000 and 50,000 suspects during the past year. Eighty-five percent of them were innocent, according to Wilkerson. "We really have created a mess here. A terrible mess," Wilkerson says. "This has been incredibly damaging."
Counterinsurgency,if there is a no-actively seeking genocide constraint is a battle of intelligence, information and effective OODA loops. Our OODA loops suck if there is an 85% false positive rate on initial captures and detentions. The metric of processing individuals is what is being measured as a sign of success, so it will be emphasized despite illustrating further downstream that the US intelligence system is not working effectively in either cultivating informers, using their information, or verifying their information against other sources to make sure that we are not being sucked into a business feud. This failure of the US OODA loop due to GIGO exposes our soldiers to more risks as they are embarking on raids that have no effective value, create further distrust between the Iraqi population and US forces, devalues the intelligence value of the cadre of useful and accurate informers as their claims probably have a discount value due to the overall veracity of intelligence, and finally creates more insurgents from the released individuals.