This paper examines the consequences of various attitudes towards climate damages through a family of stochastic optimal control models (RESPONSE): cost-efficiency for a given temperature ceiling; cost-benefit analysis with a "pure preference for current climate regime" and full cost-benefit analysis. The choice of a given proxy of climate change risks is actually more than a technical option. It is essentially motivated by the degree of distrust regarding the legitimacy of an assessment of climate damages and the possibility of providing in due time reliable and non controversial estimates. Our results demonstrate that a) for early decades abatement, the difference between various decision-making frameworks appears to matter less than the difference between stochastic and non stochastic approach given the cascade of uncertainty from emissions to damages; b) in a stochastic approach, the possibility of non-catastrophic singularities in the damage function is sufficient to significantly increase earlier optimal abatements; c) a window of opportunity for action exists up to 2040: abatements further delayed may induce significant regret in case of bad news about climate response or singularities in damages.