Digital Twins (DTs) are everywhere, but they often have been developed before consensual definitions, reference models or adapted tools were proposed. Thus, these ad hoc DTs cannot be easily manipulated as actual entities, hampering both their sustainability and their life-cycle management (e.g., their redeployment). In order to deal with this issue, this paper chooses to learn from practice by re-engineering existing DTs. Starting from a generic a priori metamodel that tries to clearly separate concerns (so that DTs can be seen as first class software entities), this paper implements experiments in three steps: (i) reverse engineering an existing Digital Twin, seen as a prototypical instance, (ii) generalization of the DT architecture descriptor into a DT model, (iii) alignment between this reverse engineered DT model and the a priori DT metamodel. First experiments, run on three use cases, show that our proposal can relevantly analyze and document DTs to support their re-engineering as first class entities.