While there is extensive accumulated experience regarding the execution of SUO plans, selecting which types of alerts to detect involved trading off several factors, such as whether the alert can be detected from available data, the utility of the alert to the user, the cost of implementation, and the ability to maintain reactivity given the computational expense of detecting the alert. We earlier gave the example of balancing the usefulness of fine-grained terrain reasoning for movement projection with its computational impact on reactivity. Thus, modeling the value of information and types of alerts to be detected involved interaction between the domain experts and system developers. Here we describe the types of alerts we decided to detect. More details of how we implement these and model the value of information and alerts is given in Sections 6.6 and 6.7.
Figure 4 describes the 13 types of alerts that are detected by the SUO EA. Most of these are time and location checking. Comparing these alerts to our categories in Section 4, the proximity alerts are all instances of adversarial activity detected. The adversarial alerts also fit this category and the last three adversarial alerts are also of the type plan constraint violated because expectations and requirements specified in the plan (such as locations and routes to monitor) are violated. The contingency alert, which can be triggered by either friendly or hostile actions, is of the type contingency plan suggested. The out-of-position, coordination, and schedule alerts are of type plan constraint violated, but would be of type constraint violation projected when the violation is projected. The fratricide alert is of type policy constraint violated, and the unknown-position alert is of type reporting requirement.