Artificial intelligence has made generating research artifacts nearly costless; verifying them remains expensive. This paper studies delegation when output is fully observable but the principal's comprehension is stochastic, capacity-constrained, and bounded by a knowledge map. Beyond the map, verification buys nothing: the principal retains formal authority yet must trust the agent or screen by origin. Volume shocks push markets across this abdication frontier; origin bans are second-best only when capacity binds or artifacts lie beyond local maps; diversifying generators without diversifying verification leaves the common blind core of error untouched; and shared verification capacity is overgrazed.
No takes yet. Share an insight, caveat, or question.
mzpl (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: