We study the behavioral differences between two isomorphic social dilemmas. In the first dilemma individuals need to refrain from appropriation to maintain a socially efficient public good (Maintenance), while in the second they need to contribute to create an inexistent one (Provision). Recent literature has shown that conditional cooperation is higher in Provision compared to Maintenance but conflicting explanations for this finding have been provided. On the one hand, it has been argued that reciprocal preferences are different across the two dilemmas because the status quo, that is, where resources are initially allocated, differs. On the other hand, it has been proposed that differences are due to frame-dependent mistakes in the recognition of the game form. In this paper, we first provide an assessment of the robustness and stability of the framing effect, showing that it is (i) replicable across subject pools, (ii) stable over time within subjects and (iii) predictive of subsequent simultaneous game play. Then, we provide a detailed test of the two conflicting explanations uncovering elements of perception of the decision problems. Our results reveal that misperceptions of the game form can account only for part of the observed framing effect, but that revealed reciprocity is stronger in Provision compared to Maintenance, even after controlling for game form misperceptions. We finally show that, in contrast to what hypothesized in current theoretical models, differences in reciprocity across the two problems are due to different perceptions of kindness of others' actions.