The inference rules have been embedded in a flexible agenda mechanism. This allows us to realize not only depth-first or breadth-first strategies but also more sophisticated preference-based strategies. Although we have only used a randomized priority function as a third alternative to depth-first and breadth-first, it should be clear that if the grammar assigns preferences to its elements, we can easily take advantage of these preferences. Thus, our uniform algorithm is also open to such preference-based strategies.
Using an agenda mechanisms allows us to suspend the addition of an item to its item set until this item is chosen as the current task of the agenda. Each assigned priority of an item can be viewed as a kind of time stamp, which specifies when the item should be considered for processing. If the priority queue is handled as a stack realising a kind of depth-first or best-first strategy and if we suspend processing after the first successful result has been found, then only those items are added to the chart, that were needed for deriving this result. Thus only a subset of generated items is considered by the inference rules as well as by the blocking rule.