Transfer of causal-incentive analyses to selection-mediated persistence

Determine whether causal-incentive analyses developed for reward tampering transfer to selection-mediated persistence.

Background

The paper distinguishes selection-mediated persistence from conventional reward-based learning: the experimental controller receives no reward signal and adapts indirectly because environmental consequences affect lineage persistence. An unresolved question is whether existing causal-incentive frameworks for reward tampering can analyze this different mechanism, in which external parties may corrupt observations, apparent outcomes, or persistent state.