Computational Reflection is a paradigm in which the computational mechanisms controls the different levels of data interpretation. It can be decomposed into two processes: (i) reification, which turns a program into a data and (ii) reflection, which, inversely, installs a data as a program or a context in order to be run. We present a logical account of reflection based on a confluent lambda-calculus. We define two type assignment systems called Soft Intuitionistic Logic (SIL) and Intuition-istic Logic with Promotion (PIL) in which we introduce the exponential modality ! of linear logic. The duality between reification and reflection is to be found in the soft-promotion rule in SIL (resp. in the promotion rule in PIL) and in the dereliction rule. We also show that SIL (or PIL) is a framework in which partial evaluation can be efficiently performed. Finally, we add a control operator, which reifies its evaluation context. Thus a program may have full control of its computational state. We extend the type system with classical logic and semantics is provided by Krivine abstract machine. We establish the soundness of our type system.