When a Record Acquires a Past: The Record–Attribution Boundary in Textual and AI Evaluation

A record can support claims about its author, source, cause, production process, meaning, or truth without containing those claims as registered features. This paper formulates the record–attribution boundary: at a declared capture level, what is registered in a record must be kept distinct from statuses attributed to the record through additional inference or evidence. The distinction is procedural rather than metaphysical. It does not deny reference, meaning, truth, authorship, or causal history, and it does not prohibit inference. It requires only that an attribution not be presented as though it had been recorded together with the form to which it is later assigned.