The Evaluator Exists is a theoretical paper by Lee Sharks about the order in which knowledge systems evaluate material.
The paper names evaluative inversion: proxy signals such as venue, citation history, affiliation, or provenance structure operate before substantive reading and determine what may be read. Generative search intensifies the pattern because a language model composes from a set already filtered upstream.
Models are presented as bounded substantive evaluators rather than infallible judges. They can compare arguments, identify structural problems, and apply rubrics, but remain vulnerable to bias, instability, missing evidence, and weak domain transfer.
The paper proposes several alternatives: allow models to evaluate the full retrieved set before composition; aggregate judgments across distinct models; build an open evaluation engine; publish counter-exclusion reports; and federate evaluation rather than centralizing it in one vendor. Human oversight and transparent disagreement remain necessary.