simplification: but as a rule internal evidence is decisive to the contrary. If now we note the groups of documents which support each of the three variants; and then, repeating the process with other conflate readings, find substantially the same groups of documents occupying analogous places in all cases, we gain first a verification of the presumption of mixture by the mutual corroboration of instances, and next a determination of one set of documents in which mixture certainly exists, and of two other sets of documents which still preserve some portion at least of two more ancient texts which were eventually mixed together. Sometimes the three groups are found nearly constant throughout, sometimes they have only a nucleus, so to speak, approximately constant, with a somewhat variable margin of other documents. This relative variability however, due to irregularity of mixture, does not weaken the force of the inferences to be drawn from each single instance. If a reading is conflate, every document supporting it is thereby shown to have a more or less mixed text among its ancestry; so that, in considering any other doubtful variation, we have empirical evidence that the contingency of mixture in each such document is not a priori unlikely. About those documents which habitually support the conflate readings we learn more, namely that mixture must have had a large share in producing their text. Similarly we learn to set an especial value on those documents which rarely or never support the conflate readings; not necessarily as witnesses to a true text, for in all these eases each true reading is paired with a simple wrong reading, but as witnesses to texts antecedent to mixture.
