SlopFilter: A Portable Epistemic-Hygiene Protocol for AI-Generated Content

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

SlopFilter asks whether a model-generated statement has earned the status its language gives it. A usable answer keeps five things visible: what is claimed, what supports it, where the support came from, what inferential step connects support to claim, and what may happen to the output next. If one of those elements cannot be recovered, the answer is revised, labeled as partial, parked, or rejected. Fluency is never promotion evidence. The public method is sufficiently specified for manual use. A current software implementation exists, but its operator vocabulary, scoring architecture, flag registry, schemas, calibrated weights, thresholds, and misuse controls are private technical material. This paper does not disclose or validate those internals.