Influence Is Not Economic Credit: Data Attribution Under Replacement

Data-attribution scores are increasingly proposed as inputs to creator compensation and data markets. Accurate attribution alone generally does not identify a buyer's willingness to pay; removing incumbent data is a different intervention from allowing the buyer to acquire substitutes. We formalize this distinction with replacement-aware pivotal value, a nonnegative buyer-side reservation price conditional on the other suppliers. When independently obtainable exact-copy options are not ruled out, replacement economies span the full interval from zero to the supplier's leave-out contribution while the learner and incumbent counterfactuals remain fixed. Shapley rankings can reverse. We derive sharp bounds from incomplete-replacement audits, a condition that certifies economic rankings, and a finite-error guarantee for estimated utilities. Generative artificial intelligence makes catalog timing consequential: a legally retained teacher can create an ex-post outside option unavailable before training. A worked certificate separates evidence of influence, provenance, and replacement. The results specify what attribution must additionally measure to support procurement decisions without treating economic replaceability as an entitlement rule.