Answers are grounded in this paper and cite their sources.
How can a closed feedback loop be assembled from stochastic components without assuming what it should prove? This paper builds one exactly from a controller, a noisy command channel, a plant with a hidden bit and a return channel. Part I connects the parts by an initialized staged exchange. Each stage receives the actual output of the stage before it, so every recipient's assumption is discharged by that same intermediate input, and no circular fixed point arises. The acknowledged success probability has an exact formula, with worst cases 1/2, 3/8 and 1/4 for clean, noisy and erasing channels under fresh tickets; correlated tickets with the same uniform marginals lower the noisy case to 1/4. Part II closes the loop through a delay register. Every run's order of events is acyclic although the wiring is a cycle, and at gap bound one a long enough admitted run commits or faults within fourteen steps. For each fixed local policy, the greatest set that stays safe under every scheduler move and every positive outcome is computed; for the complement policy it yields an initialized, mass-preserving controlled law on 944 states. A Lean 4 formalization accompanies both parts.
No takes yet. Share an insight, caveat, or question.
Ioannis Tsiokos (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: