The gatekeeper loop is the severe failure mode at the center of the gatekeeper scenario. It requires no malice and no dictator — only millions of individually reasonable predictive decisions. Low-probability people stop getting opportunities; the model retrains on the world that produces; and it discovers, accurately, that those people rarely succeed. Of course they don’t — the ecology stopped letting them through. The prediction has begun creating the reality that confirms it.
Its danger is that it looks like accuracy improving over time. This is where the framework’s rule about efficiency bites: the more a gatekeeper optimizes average outcomes, the more deliberately the ecology must preserve routes for statistical exceptions — otherwise efficiency becomes ontological selection, quietly hostile to emergence and to any geometry that has not yet proven its value. Exception budgets exist specifically to break the loop, by keeping a route open for what the model cannot yet see and generating the signal that tests whether it is shaving away real value.