Messy Research, Certification and the Monetization of Science

I study when science starts charging for what reputation used to provide. In the model, AI lowers the cost of producing a polished manuscript faster than it lowers the cost of judging the work behind it. Polish then stops telling good work from plausible work. Entry expands. The willingness to pay for credible certification rises, and once polish fails entirely as a signal of quality, it rises to the full reputational gap between certified and uncertified work. What happens next depends on what an overloaded certifier does with the submissions it cannot read. In the leading case, where high-quality work is plentiful relative to review capacity yet less than half of the population, the premium of a journal that certifies its unread pile peaks and then declines permanently, the premium of a journal that rejects that pile ends congestion above where it began exactly when high-quality work is scarce enough, and the two paths cross, so that for a time the tolerant journal's stamp is worth more. The congestion prices (submission tolls, diluted stamps, rents to scarce refereeing) exist only inside a window that narrows as verification capacity grows faster, and that never opens at or above a threshold growth rate. The full certification premium turns on the polish threshold, not on the window, and it persists after the window closes. Whether the new prices are transitional or permanent turns on the relative growth rate of verification.