In point of fact the data of any inquiry are always a _selection_. They are such of the recognized facts as are thought to be _relevant_, i.e. to be truly ‘facts’ for the purpose in hand. But being a selection they involve us in the risk that we may have selected wrongly, and omitted what is important while admitting what is not. _From this risk there is no escape._ For we cannot effect a compromise by including merely so much of the facts as we can lay hold of. Not only does this yield no guarantee that everything that is needed has been included, but it may be a positive hindrance to try to include too much. For if our data grow into an unwieldy mass, they will not seem susceptible of any order or principle, and even the most penetrating inquirer will lose his way. It is better, therefore, to give up altogether the idea of securing formal validity by postulating an all-inclusive exhaustiveness. The obvious alternative is to operate simultaneously with a plurality of theories, each of which means a certain ordering of the ‘facts’ relatively to what seems a relevant and promising point of view. Each will involve a selection and induce a bias; but with any luck they will neutralize each other’s bias, and so will increase the probability that no really relevant fact has escaped notice. This will not satisfy the logical ‘ideal’, but in practice it means a good deal, and is enough for scientific progress. Of course it must be understood that the hypotheses employed are in a general way relevant to the problems and the condition of the sciences, and not random guesses. This proviso will cut down their exuberance even more than the limitations of the human imagination, which seems to be psychologically incapable of really departing very far from the suggestions of experience. § 26. When logic has recognized the use and value of ‘working’ as the test of truth, it must, however, make it clear to itself and to others both what precisely this test is, and what it can, and cannot, accomplish. In the first place, it must be made clear that it is _not_ a logical implication of the test that ‘whatever works is true’, and the reasons for disputing this dictum must be set forth. The fact is that we all have a strong psychological tendency to believe in the truth of what is found to work, without much criticism of the sort and extent of the ‘working’. But the logician should carefully investigate the various sorts of working that occur, and take special note of those which either do not themselves lay claim to full truth, or do not (ordinarily) have their claim conceded. For example, ‘_fictions_’ are not supposed to be strictly true; but they may ‘work’ and be ‘as good as true’, or ‘pragmatically true’, or ‘sufficiently true for the purpose in hand’. They work, in fact, within limits; but these limits are _known_, and so they are not confused with full-fledged truths, to the applicability of which there are no known limits.