I think the usual back-and-forth does converge on better policy, but feedback systems from an engineering perspective are notorious for having stable and unstab

I think the usual back-and-forth does converge on better policy, but feedback systems from an engineering perspective are notorious for having stable and unstable regimes. If you crank the gain or introduce too much delay, your stabilizing control system turns into an oscillator, slamming against the system's extremes rather than settling down at a compromise. It's not an encouraging analogy.