"How can the inflow be negative?" is a commonly asked question. Here's the answer:

This is from a document I found regarding a different lake:

A similar local inflow term was calculated for Lake Tapps reservoir by applying a mass balance to the lake. The local inflow to Lake Tapps term represents measurement error, evaporation, precipitation, leakage to groundwater, and surface water inflows to the lake or the diversion canal (Ramey and Yoder 2004). The Lake Tapps local inflow series is sensitive to measurement error in the water surface elevation of Lake Tapps. A small measurement error represents a large volume of water and can create large positive or negative inflow to the lake. The average local inflow to Lake Tapps is 46 cfs and is higher in winter than in summer. During some summer months, the average inflow series is negative indicating that losses to evaporation and recharge outweigh any inflows. The local series ranges from -3,075 cfs to 2,268 cfs, but 80 percent of the values are between -150 cfs and 250 cfs. The large range in the Lake Tapps local inflow series is a limitation of the model and causes some counterintuitive results such as a few days of large changes in flow in the Puyallup River with the WSP.

It's caused by any factor that "pulls" water from the lake, be it leakage to groundwater, water processing plants, evaporation, etc. Outflow only covers water that actually flows out of the dam. Although I have not confirmed this with the ACoE, I suspect the calculation is something along these lines:

The lake level is a known quantity (well, ignoring the fact they were 2 feet low for who knows how long...). The outflow will cause the lake level to drop by some amount. The inflow is measured from the stream gauges on the streams that feed Allatoona. The lake should rise a given amount for a given inflow. The expected rise is subtracted from the actual lake level, and the difference is converted back into a flow rate, which is then added to the measured inflow rate. If the measured inflow is less than the calculated inflow, the inflow is then negative. More water was lost to leakage to groundwater, evaporation or water processing plants than came in from the streams.

Here's a very crude example:

If the lake is at 1000', they release 1000CFS which drops the lake 1 foot, but the measured drop is 1.1 foot, you've got negative 100CFS inflow. If the lake is at 1000', they release 1000CFS, and the lake only drops 0.9 foot, the 0.1 difference is because of a 100CFS inflow. Of course, I'm just making up numbers to show the idea.

If this answer is wrong, please feel free to email me (address below) with how it actually works. This is based on inference from various reading. If the ACoE would ever actually answer their phone, this is on my list of questions to ask them.