(1) We begin by inquiring whether the statements are independent, or are reproductions of one and the same observation. This inquiry is partly the work of that part of external criticism which deals with the investigation of sources;[173] but that investigation only touches the relations between written documents, and stops short when it has determined which passages of an author are borrowed from other authors. Borrowed passages are to be rejected without discussion. But the same work remains to be done in reference to statements which were not committed to writing. We have to compare the statements which relate to the same fact, in order to find out whether they proceeded originally from different observers, or at least from different observations. The principle is analogous to that employed in the investigation of sources. The details of a social fact are so manifold, and there are so many different ways of looking at the same fact, that two independent observers cannot possibly give completely coincident accounts; if two statements present the same details in the same order, they must be derived from a common observation; different observations are bound to diverge somewhere. We may often apply an _a priori_ principle: if the fact was of such a nature that it could only be observed or reported by a single observer, then all the accounts of it must be derived from a single observation. These principles[174] enable us to recognise many cases of different observations, and still more numerous cases of observations being reproduced. There remains a great number of doubtful cases. The natural tendency is to treat them as if they were cases of independent observation. But the scientific procedure would be the exact reverse of this: as long as the statements are not proved to be independent we have no right to assume that their agreement is conclusive. It is only after we have determined the relations between the different statements that we can begin to count them and examine into their agreement. Here again we have to distrust the first impulse; the kind of agreement which is really conclusive is not, as one would naturally imagine, a perfect similarity between two narratives, but an occasional coincidence between two narratives which only partially resemble each other. The natural tendency is to think that the closer the agreement is, the greater is its demonstrative power; we ought, on the contrary, to adopt as a rule the paradox that an agreement proves more when it is confined to a small number of circumstances. It is at such points of coincidence between diverging statements that we are to look for scientifically established historical facts.