Lede: A preprint by Yang Lu proposes a two-layer certification framework separating knowledge quality from human contribution to address AI research pipelines in academic publishing.

The framework classifies contributions as Category A (pipeline-reachable), B (human direction at key stages), and C (beyond current automated reach), with contemporaneous grading and benchmark slots for fully automated work (Lu, arXiv:2604.22026). It draws on the Sakana AI Scientist system that produced full papers via automated hypothesis-to-publication pipelines (arXiv:2408.06292) and aligns with Nature's 2024 analysis of AI transforming scientific discovery while exposing reproducibility gaps (https://www.nature.com/articles/s41586-024-07781-0).

Original source correctly identifies that traditional publication certifies both validity and human authorship but understates how 2023-2025 surges in papermill AI papers exposed similar attribution failures documented in Science (Hutson, 2022). Dry-run validation on two attribution cases demonstrates tolerance for irreducible uncertainty yet omits scalability testing against high-volume arXiv submissions.

Synthesizing these sources reveals the framework's benchmark slots could calibrate reviewer judgment across journals without new institutions, grounding frontier credit in epistemic achievement at the formulation stage rather than unverifiable human-origin claims.