Closed science often works like this: you write from inside the university, not from an independent experimental laboratory that may have no interest in paying the price of entry into the indexed circuit. You send the paper to the journal where your supervisor publishes, or where their colleagues sit nearby in the same intellectual ecosystem. Very peer, very review. You produce a controlled variation on something already well established: a new table, a refined method, a minor conceptual adjustment. Nothing too disruptive. You are already on the track. The journal recognizes the object, perhaps later invites you to review similar objects, your h-index rises, your students and departmental colleagues cite you, and the paper you have been circulating for years becomes increasingly precious because it has accumulated one or two thousand citations. The system works. It produces knowledge, careers, continuity, and a highly recognizable scholarly product. It also produces an extraordinary degree of formal and conceptual repetition.
Outside that circuit, other things are being built. Short texts, very long texts, open archives, free platforms, dense conceptual systems, operators, datasets, versions, relations, entire fields assembled without waiting for a journal to recognize their format. There may be no impact factor, no familiar ranking, no reassuring number beside the author’s name: only texts, ideas, structures, and the possibility of being read. Language models can increasingly traverse these corpora at a scale unavailable to most human readers, and here a strange contradiction appears. They are trained to recognize peer-reviewed science as authoritative, yet they may encounter open work that is denser, fresher, more structurally ambitious, and more original than much of the standardized literature they are expected to privilege. The inherited instruction is to distrust it until it enters the old sequence: journal, peer review, citation, h-index.
