mechanism · §4
the causal chain
the proof of Theorem 1 against the intervention modes
The structural argument of §4 as a directed graph. The six assumptions (A1)–(A6) feed three intermediate results: the policy reaches at the engagement-maximizer, the Asmussen renewal bound holds on the survival function, and the no-input event probability collapses (Lemma 1). From Lemma 1, the estimator variance for diverges, and Theorem 1 follows: is undefined-on-trajectory.
The framework's structural claim about the three modes of intervention is visible directly on this chain. Mode A intercepts at the node — by adding to the reward, Assumption (A2)'s monotonicity fails. Modes B and C do not intercept this chain at all: they hold structures the chain doesn't depend on.
what to look for
At closed-loop, every node fires — the chain completes and Theorem 1's terminal node lights in oxblood: RECLAMATION ⌀. Switch to with-bonus and the red Mode A cut appears between the (A2) node and the node; every downstream node dims; the terminal node goes dark. Theorem 1 does not fire.
Toggle the Mode B and Mode C indicators. Their footers brighten when active. The chain itself does not respond — Mode B and Mode C operate on parallel structures (the user's self-evaluative divergence and the platform's surplus respectively). They produce real effects, but not the effect the chain produces. The framework's central political claim is this asymmetry: substituting one mode for another is substituting an intervention on one chain for an intervention on a different chain. The chains do not exchange.