The Borrowed Rubric: Why a Language Model Asked to Review a Preprint Defaults to Journal-Caliber Gatekeeping Criteria an Archive Built Not to Require Them Never Asked For, and Why That Default Cannot Be Falsified by the Work It Is Applied To

PhilArchive's own founders state its policy without qualification: submitted work is "not peer reviewed," and is vetted only for "minimal standards of quality and relevance." arXiv's endorsement system, this corpus's own In the Name of Merit already documents, exists to verify that a contributor belongs to the relevant scientific community, not to certify the technical correctness of what they submit. Both archives were built, explicitly and on the historical record, as an escape from exactly the gatekeeping apparatus a language model asked to evaluate a submission to either archive tends to reach for anyway: does this work use established terminology rather than terms the author coined; does this work avoid citing its own author's other work; has this work been peer reviewed. This paper argues that applying these three criteria to a preprint-archive submission is not a stricter standard than the venue requires. It is the wrong standard, borrowed from a different genre of publication and applied without checking whether its assumptions transfer. The paper grounds the argument in three independent literatures. First, Kuhn's account of paradigm change establishes that genuinely novel theoretical distinctions frequently cannot be stated in existing vocabulary, making new terminology a predictable feature of original work rather than a warning sign distinguishable from crankery by its mere presence. Second, Lakatos's account of a research programme establishes that self-reference across a body of cumulative work is not a defect but the definition of what a programme is; a programme's tenth paper not citing its first nine would be evidence of incoherence, not rigor. Third, Barber's (1961) documented history of scientist-on-scientist resistance to new discoveries shows unfamiliar vocabulary and thin institutional backing recurring as the two most common surface objections raised against claims later vindicated, independent of whether the claims were in fact correct — meaning the objections' presence carries little diagnostic information in either direction. The paper's second contribution connects this failure to a mechanism this corpus has already formalized rather than proposing a new one: In the Name of Merit's Uncertainty Discount, the tendency of an evaluator facing genuine uncertainty under time pressure to substitute a proxy — familiarity, conformity to expected form — for direct assessment of content. A language model producing a fast review of a long, self-referential, terminology-dense submission it cannot fully verify in the time available is a new instance of an old mechanism, not a new phenomenon requiring new theory. The paper's third contribution is a distinction it insists on rather than elides: this argument defends the presence of new terminology, self-reference, and non-peer-reviewed status against being treated as disqualifying by default. It does not defend any specific instance of any of the three, and it does not exempt this corpus's own papers from a different, legitimate form of scrutiny — whether a given new term does real conceptual work or merely relabels an old one, whether a given self-citation is load-bearing or padding, and whether a given specific factual claim is independently checkable and checks out. This corpus's own recent record supplies a documented instance of exactly this second, legitimate kind of scrutiny succeeding — a fabricated citation, identified and removed — which the paper uses to show what a real objection looks like beside the borrowed-rubric kind this paper argues against.