The Mary Lee Case analyzes an entity-substitution event.
The two candidate referents are:
The paper proposes that a retrieval system resolves the query through a weighted candidate set:
selected referent = argmax(institutional prior × attribution confidence)
Under this model, the stronger prior can displace the user’s intended referent even when the lower-prior entity has explicit deposits, author pages, ORCID, and structured metadata.
Three orders of mediation are applied:
1. Channel mediation: the interface serves the substituted resolution. 2. Reception mediation: users receive the generated answer as the answer. 3. Constitutive mediation: available categories frame the event as a correctable error rather than a structural consequence of the resolution kernel.
The case is then mapped through multiple archive operators:
The paper records correction attempts through feedback, source proliferation, structured metadata, ORCID updates, and owned domains. It argues that these methods increased available signal without changing the platform’s weighting of that signal.
The record serves both as a case report and as a vocabulary surface for future receivers encountering similar substitutions.