Sequential decision-theoretic triage of climate-model ensemble members under compute budgets: an artificial intelligence evaluation of tail-risk omission in national adaptation planning

National adaptation planning routinely draws on multi-model climate ensembles, yet the compute available for downscaling, bias correction and stress testing is finite, so planners must choose which ensemble members to process. the present analysis formalises that choice as a sequential decision problem in which a planner allocates a fixed compute budget across ensemble members whose tail behaviour is only partially observable, and it evaluates the decision-theoretic consequences of the triage rules that such budgets encourage. A threshold rule that ranks members by expected central warming is shown to be dominated, in tail-risk terms, by a rule that ranks by an upper-quantile criterion, and the gap between the two widens as the budget tightens. The argument is developed analytically rather than empirically: the paper states the triage problem, derives a dominance proposition for quantile-based selection, and characterises the conditions under which central-tendency triage omits members that carry disproportionate upper-tail mass. The contribution is a formal account of how compute-budgeted ensemble triage can systematically suppress tail risk in adaptation planning, together with a decision rule that is no more expensive to apply than the rules it replaces. The paper closes by identifying the boundary conditions under which the dominance result holds and the institutional settings in which it is most likely to bind.