Extension to continuous action spaces

Determine whether the reported effects—base models outperforming aligned models in predicting human behavior in multi-round strategic games and the reverse in one-shot or non-strategic tasks—extend from binary or ternary decision settings to continuous action spaces.

Background

All evaluated tasks involve discrete choices (binary or ternary), leaving open whether the same base-versus-aligned performance patterns hold when actions are continuous (e.g., price offers on a real-valued scale).

Establishing this extension is necessary to understand the generality of the normative–descriptive trade-off across a broader spectrum of strategic and decision-making problems.