Theory of Subjective Evaluation: The Grounds and Limits of Claiming Subjectivity

This paper develops the Theory of Subjective Evaluation (TSE), a domain-general epistemic framework for answering a specific question: when is an attribution of genuine subjectivity warranted? TSE distinguishes an epistemic residual - the disagreement, uncertainty, or unexplained remainder left after currently accessible correction - from true subjectivity, understood as genuine irreducible evaluator-dependence in evaluative structure. The distinction matters because persistence under present inquiry is evidence about irreducibility, not irreducibility by definition. TSE tracks how residuals change under independently supported improvements in information, reasoning, comparison, articulation, calibration, and error correction, subject to continuity of the evaluative target. Corrigible structure (C), irreducible evaluator-dependence (I), and alternative stable explanations (A) are diagnostic classifications rather than mutually exclusive ontological kinds. Evidence may move toward or away from these classifications, but stable evaluator-dependence does not by itself distinguish I from A: support for I specifically requires evidence that discriminates it from serious A explanations. A limited evaluator can therefore become warranted in attributing I without possessing conclusive certification of metaphysical finality, although support for I is generally harder to establish than support for a corrigible mechanism. Proposition 1 states the stronger limit that persistence of an epistemic residual, by itself, cannot conclusively certify irreducibility when such certification requires exclusions the evaluator is not in a position to complete. A parallel constraint applies to absolute evaluator-independence. TSE is also reflexive: failure at a finite stage to identify further substantive implications cannot itself certify that the theory's implication-space is exhausted. Its own diagnoses remain evaluative targets: no finite order becomes immune to possible assessment merely by being higher-order, although this creates neither an infinite regress of justification nor a guarantee of endless substantive novelty. The framework's principal practical application is to machine-learning evaluation pipelines: aggregating plural human judgments into labels, rankings, or reward signals can suppress a residual without correcting it. TSE therefore requires explicit evaluative targets and continuity checks before stable aggregate output is treated as evidence of correction or objectivity. Finally, the paper distinguishes TSE's constitutive commitments and conditional consequences from stronger meta-normative or metaphysical extensions. TSE requires correction to differ from mere change, but it does not itself supply the ultimate grounding of corrective standards. Given a preserved target, some evaluations can be objectively constrained by world-side facts; if genuine subjectivity exists, that dependence can itself be objectively true. These claims do not establish categorical moral realism or a fundamental relational ontology.