An evaluative axis organises mental representations of prospective experiences and predicts choice

People constantly make choices between prospective experiences that share almost nothing in common, yet it remains unclear how these experiences are mentally represented such that comparison, and choices between them, are possible. Here, we mapped the representational space of prospective experiences by fitting a computational model to 30,000 odd-one-out similarity judgements of 200 scenarios spanning the mundane, aversive and fantastical. The model recovered seven reproducible dimensions. The task never required evaluation, yet the principal axis of the space was evaluative, tracking independent ratings of valence and prestige. In 10,000 separate choices, valence was the strongest predictor of which experiences participants preferred. The model also predicted these choices, but valence exhausted its choice-relevant information, and the valence and choice directions were nearly identical. These results suggest that although the space is multidimensional, it also contains a unidimensional evaluative ordering that emerges in the absence of preference-based choices, but nevertheless predicts them.