Regulators currently govern the AI data economy based on intuition rather than evidence, struggling to choose between inconsistent regimes of informed consent, immunity, and liability. To fill this policy vacuum, this paper develops a novel computational policy laboratory: a spatially explicit Agent-Based Model (ABM) of the data market. To solve the problem of missing data, we introduce a two-stage methodological pipeline. First, we translate decision rules from multi-year fieldwork (2022-2025) into agent constraints. This ensures the model reflects actual bargaining frictions rather than theoretical abstractions. Second, we deploy Large Language Models (LLMs) as "subjects" in a Discrete Choice Experiment (DCE). This novel approach recovers precise preference primitives, such as willingness-to-pay elasticities, which are empirically unobservable in the wild. Calibrated by these inputs, our model places rival legal institutions side-by-side to simulate their welfare effects. The results challenge the dominant regulatory paradigm. We find that property-rule mechanisms, such as informed consent, fail to maximize welfare. Counterintuitively, social welfare peaks when liability for substantive harm is shifted to the downstream buyer. This aligns with the "least cost avoider" principle, because downstream users control post-acquisition safeguards, they are best positioned to mitigate risk efficiently. By "de-romanticizing" seller-centric frameworks, this paper provides an economic justification for emerging doctrines of downstream reachability.

翻译：目前，监管者基于直觉而非证据来管理人工智能数据经济，在知情同意、豁免与责任等互不一致的体制之间艰难抉择。为填补这一政策真空，本文构建了一个新颖的计算政策实验室：一个具有空间显式特征的数据市场基于主体模型（ABM）。为解决数据缺失问题，我们引入两阶段方法论流程。首先，我们将多年实地调查（2022-2025年）中的决策规则转化为主体约束条件，确保模型反映实际议价摩擦而非理论抽象。其次，我们部署大语言模型（LLMs）作为离散选择实验（DCE）中的"受试者"。这种新颖方法可恢复精确的偏好原始数据，例如现实中无法经验观测的支付意愿弹性。经过这些输入校准后，我们的模型将相互竞争的法律制度并置，模拟其福利效应。研究结果挑战了主流监管范式。我们发现，知情同意等财产规则机制无法实现福利最大化。反直觉的是，当实质性损害责任转移至下游买方时，社会福利达到峰值。这符合"最低成本避免者"原则——由于下游用户控制购置后的保障措施，他们最具备高效降低风险的能力。通过"去浪漫化"以卖方为中心的框架，本文为新兴的下游可触及性理论提供了经济学依据。