In our preceding Reply, we indicated that Mulargia et al. [1996> made (beyond their obvious error that they checked their predictive ''rule'' only for its ''learning period'') a number of mistakes; we also showed that their ''rule'' does not correspond to a meaningful algorithm. Mulargia et al.'s Re-Rebuttal admits that Mulargia et al. [1996> actually made a number of mistakes due to a ''bug in the [Mulargia et al.'s 1996> code'', which not only omitted from their list two (non-''predicted'' by their ''rule'') ''large'' earthquakes (EQs), but also scored two missed (''large'') EQs as successfully ''predicted''. Furthermore, they now admit that Mulargia et al.'s [1996> rule ''is certainly not an efficient predictor'', in contrast to their earlier claims. The main issue of our present Reply is to point out that Mulargia et al., in their Re-Rebuttal, now make a very serious error, when constructing the errors diagram: they confuse predictions of main shocks with those of the aftershocks, and hence incorrectly conclude that one can ''build very simple, zero-cost predictive tools superior to VAN''. We show that their erroneous procedure leads to the following paradox: when a ''rule'' (which fails to predict all main shocks) correctly ''predicts'' a number of aftershocks, one can (incorrectly) claim that he found a predictive ''rule'' superior to the ideal prediction method; the latter (i.e., the ideal one), in spite of the fact that it predicts all main shocks, is (incorrectly) obtained to correspond to ''random predictions''. ¿ American Geophysical Union 1996