The keystone. Intelligence operates on a single agent's marginal distribution; institutions operate on the joint distribution of many agents — and joint properties aren't expressible by marginals. So no amount of intelligence-is-satisficing-under-scarcity can buy what an institution provides.
Four goods acceleration can never buy:
- convergence compatibility — each agent accelerating toward its own G can reach Pareto-inferior fixed points (prisoner's dilemma, arms race, tragedy of the commons; "the smarter, the faster they sink"). Intelligence optimizes within a game; institutions design the game (North).
- common knowledge of convergence — promise ≠ proof's final form; "I know I'm fine" ≠ "everyone knows everyone knows." The coin/ledger (round-tree-flight-recorder) transports private convergence into circulable certainty (Chwe's rational ritual).
- cross-time commitment — dynamic inconsistency; Ulysses' mast must be outside his own will ("a lock is what Ulysses bought").
- tail pricing — intelligence satisfices at its own threshold, which doesn't price others' disasters (externalized cost returns).
"The smarter the agents, the larger the share of value held by 'toward what, whose, how believed, how bound' → institutional value rises monotonically with intelligence." So vibe-linter is an external copy of grids 2/5/8 — its existing name is institution. ("Alignment solved as a byproduct of capability" is a category error: a marginal-distribution operator can't do joint-distribution work.)
Up: vibe-linter