This paper addresses reflected backward stochastic differential equations (RBSDE hereafter) that take the form of Here τ is an arbitrary random time that might not be a stopping time for the filtration F generated by the Brownian motion W . We consider the filtration G resulting from the progressive enlargement of F with τ where this becomes a stopping time, and study the RBSDE under G. Precisely, we focus on answering the following problems: a) What are the sufficient minimal conditions on the data (f, ξ, S, τ ) that guarantee the existence of the solution of the G-RBSDE in L p (p > 1)? b) How can we estimate the solution in norm using the triplet-data (f, ξ, S)? c) Is there an RBSDE under F that is intimately related to the current one and how their solutions are related to each other? This paper answers all these questions deeply and beyond. Importantly, we prove that for any random time, having a positive Azéma supermartingale, there exists a positive discount factor E -a positive and non-increasing F-adapted and RCLL process-that is vital in answering our questions without assuming any further assumption on τ , and determining the space for the triplet-data (f, ξ, S) and the space for the solution of the RBSDE as well. Furthermore, we found that the conditions for the G-RBSDE are weaker that the conditions for its F-RBSDE counterpart when the horizon is unbounded. Our approach sounds novel and very robust, as it relies on sharp martingale inequalities that hold no matter what is the filtration, and it treats both the linear and general case of RBSDEs for bounded and unbounded horizon. * Tahir Choulli is grateful to Polytechnique for the hospitality, where this work started in March 2018, and he is grateful to Nizar Touzi for introducing him to the RBSDEs and proposing him this project and its main ideas.