Conceptions of scientific theories are usually distributed into two distinct subsets. The first one is normative and teleological. According to it, scientific theories have or should have "epistemic value"; they have or should have credentials for approaching isomorphism with a putative "external reality" construed as a final target and a criterion of truth; and therefore the ultimate structure of theories is necessary. The second one has an evolutionist tinge; it restricts its normative aspect to viability. Here, no epistemic value is required, but only adaptative value; no truth, but empirical adequacy; no pre-defined final target, but a proteiform quest for ecological niches; no necessity of correspondence, but historical contingency. A third conception, a "middle way", can however be identified in the history of ideas. This alternative conception (called transcendental epistemology) was first formulated in a highly fixist version by Kant, and later made more flexible and historically sensitive by the neo-kantian lineage. In this third conception, epistemic value is retained, yet only as a regulative ideal. The claim of truth is no longer discarded but it is thoroughly redefined. Truth is not restricted to logical coherence, nor does it imply mere correspondence with "things-in-themselves". Rather, "objective truth" means "connection according to laws of experience"¹ provided in advance by our understanding; namely connection of phenomena according to those very "constitutive laws' whose application are a condition of possibility of any experience of objects. Moreover, in Cassirer's version of neo-kantianism, the constitutive laws are historically drifting, by way of a progressive conquest of accurate 'symbolic forms"; but the content of scientific theories is still ascribed a certain amount of internal necessity in so far as it must incorporate the group-structure whose invariants define its own objects (Cassirer, 2004).