The classic finding from short-term relative JOR tasks is that correct response time (RT) depends on the lag to the more recent item but not to the less recent item (Hacker, 1980). For decades, researchers have argued that this finding is consistent with a self-terminating backward scanning model (Muter, 1979;Hacker, 1980;Hockley, 1984;McElree & Dosher, 1993). This finding has taken on new importance in light of recent proposal that many forms of memory depend on a compressed representation of the past (Howard, Shankar, Aue, & Criss, 2015). This paper replicates and extends the results of the classic papers. A Bayesian t-test showed substantial evidence for the null e↵ect of lag to the less recent item on correct RT. In addition, this paper reports that correct RT is a sub-linear function of lag to the more recent probe and replicates the classic finding that error RT depends on lag to the less recent probe. These findings place new constraints on models of short-term memory scanning.In a relative judgment of recency (JOR) task participants choose which of two probes from a list was presented more recently. Recency in this task is traditionally measured in units of lag, which is the number of time steps in the past at which the probe was presented. That is, if the last item in the list was presented as a probe, it would be associated with a lag of one. The classic finding from short-term relative JOR tasks is that correct response time (RT) depends on the lag to the more recent item but not to the less recent item (see Fig. 1, Muter, 1979;Hacker, 1980;Hockley, 1984). For decades, researchers have argued that this finding is consistent with a self-terminating backward scan (Muter, 1979;Hacker, 1980;Hockley, 1984;McElree & Dosher, 1993). Because the scan starts at the present and proceeds backwards in time, RTs show a recency e↵ect. Because the scan is self-terminating this naturally accounts for the finding that correct RTs do not depend on the lag of the less-