Of all these different kinds of results one only is definitive--_the statement of an author who can have had no information on the fact he states is null and void_; it is to be rejected as we reject an apocryphal document.[169] But criticism here merely destroys illusory sources of information; it supplies nothing certain to take their place. The only sure results of criticism are _negative_. All the positive results are subject to doubt; they reduce to propositions of the form: "There are chances for or against the truth of such and such a statement." Chances only. A statement open to suspicion may turn out to be true; a statement whose truth is probable may, after all, be false. Instances occur continually, and we are never sufficiently well acquainted with the conditions under which the observation was made to _know_ whether it was made ill or well. In order to obtain a definitive result we require a final operation. After passing through the ordeal of criticism, statements present themselves as probable or improbable. But even the most probable of them, taken by themselves, remain mere probabilities: to pass from them to categorical propositions in scientific form is a step we have no right to take; a proposition in a science is an assertion not open to debate, and that is what the statements we have before us are not. It is a principle common to all sciences of observation not to base a scientific conclusion on a single observation; the fact must have been corroborated by several independent observations before it is affirmed categorically. History, with its imperfect modes of acquiring information, has less right than any other science to claim exemption from this principle. An historical statement is, in the most favourable case, but an indifferently made observation, and needs other observations to corroborate it. It is by combining observations that every science is built up: a scientific fact is a centre on which several different observations converge.[170] Each observation is subject to chances of error which cannot be entirely eliminated; but if several observations agree, this can hardly be in virtue of a common error: the more probable explanation of the agreement is that the observers have all seen the same reality and have all described it correctly. Errors are personal and tend to diverge; it is the correct observations that agree. Applied to history, this principle leads to a last series of operations, intermediate between purely analytical criticism and the synthetic operations--the comparison of statements.