Contextual Dataset Valuation and the Limits of a Universal Price

Research note with a reproducible synthetic experiment. Dataset utility can be measured conditional on the learner, baseline inventory, target population, and valuation estimand; predictive gains alone do not identify a universal market price. The note distinguishes acquisition value, removal value, and Shapley attribution, and gives a conservative simultaneous held-out acquisition bound. Across 50 generated instances and three target populations, the two specialist datasets reverse their acquisition ranking in 42 instances. Exact duplicate suppliers have zero removal value but equal Shapley attribution. This is a transparent worked example using established methods, not a new pricing algorithm or evidence about real-market prices. Not peer reviewed. Substantial OpenAI Codex assistance is disclosed in the manuscript. Code, complete results, and internal review: https://github.com/WINSTON672/winston-lin-research/tree/main/papers/data-valuation