I took into account an Ammianus passage, namely 20,8,3–4, where the text edited by all modern scholars has been marked with two crosses to show that the text is not sure. As a matter of fact in every modern edition we read: [ Iulianus ] quamquam non † repugnanter, tamen nec adrogantibus uerbis, quicquam scripsit, ne uideretur subito † repugnasse . Here the adverb tamen causes problems, because the reader expects to find quamquam non repugnanter, nec adrogantibus uerbis, quicquam scripsit without tamen . A solution is to place tamen after uerbis as Pighi and Seyfarth did to substitute quamquam non repugnanter, tamen nec adrogantibus uerbis, quicquam scripsit with quamquam non repugnanter, nec adrogantibus uerbis, tamen quicquam scripsit . But such a solution is too arbitrary. Another solution was proposed by Petschening, i.e. to replace non repugnanter with non repurganter . But the difference between non repugnanter and non repurganter was too reduced and a similar difficulty occurred also with repurganter . However, I considered more attentive the reading of the manuscript Fuldensis, the father of all manuscripts which transmitted Ammianus’ text, and found that the word tamen , which caused the problem, did not exist and was produced by modern editors from the following reading: repugnant certant V m1 repugnanter tamen m2 ( V = Fuldensis). Here certant seems to have been introduced in order to explain the infrequent word repugnanter : repugnant certant of the first hand of V . The word certant was a gloss, and the reading of the second hand of V was produced by dividing and eliminating certant , the gloss, in this way: repugnan terr̭ tanῑ , where cer ( of cer tant ) was read ter , considering the great similarity of c and t in Carolingian script, and added to repugnant /. This produced repugnanter and what remained of certant ( cer tant ) was read tam which for its part was considered an abbreviation of tamen . On the other hand, I found in Ammianus some other examples of this behaviour employed by the librari, e.g. 22,8,29 Sauromatae, per quos amnes fluunt perpetui // Clark Rolfe Selem Seyfarth Fontaine Viansino per quosdam esfluit V m1 per quosđ/ amnes fluunt V m3. In this case quosđ/ am and the letter n added over and es from es fluunt has produced amnes , with the letter d deleted and another n placed over, thus arriving finally at the expression: per quos (i.e. Sarmatas ) amnes fluunt perpetui: per quosđ am n es fluu n t per petui V m3. A clear example how the librari of V worked.