Recent research on attentional control processes in the Eriksen flanker task has focused on the so-called congruency sequence effect a.k.a. the Gratton effect, which is the observation of a smaller flanker interference effect after incongruent than after congruent trials. There is growing support for the view that in this paradigm, the congruency sequence effect is due to repetition of the target or response across trials. Here, results from two experiments are presented that separate the contributions of target, flanker, and response repetition. The results suggest that neither response repetition alone nor conflict is necessary to produce the effect. Instead, the data reveal that only flanker repetition is sufficient to produce congruency sequence effects. In other words, information that is associated with a response irrespective whether it is relevant for the current trial is bound to response representations. An account is presented in which the fleeting event files are the activated part of the task set in which flankers, targets, and response representations are associatively linked and updated through conflict-modulated reinforcement learning.