Policy convergence
The state in which the policy engine's adaptive weights have stabilized and further reinforcement produces diminishing changes. One of the trigger conditions for open-ended evolution: if policy convergence occurs while the system still experiences persistent failure, it indicates the current architecture's capability ceiling has been reached.
Policy convergence is not inherently a problem - it can mean the system has found a stable, well-performing behavioral policy. It becomes significant when convergence coincides with persistent failure: the system cannot improve further by adjusting its weights, yet outcomes remain poor. This combination is the signal that the problem is architectural, not parametric - the current capability set lacks the tools to solve the problem, and open-ended capability search is warranted. The policy engine detects convergence by monitoring drift magnitude over recent sessions: when per-step drift consistently falls near zero despite negative reinforcement signals, the policy has converged. The distinction between healthy convergence (stable high performance) and failure-coincident convergence (stable low performance) is what the developmental engine and meta-cognition layer evaluate before authorizing open-ended evolution mode.