Working paper — not peer reviewed
The Verification Horizon
Trustworthy Autonomous Software Delivery as a Closed-Loop Control Problem
Autonomous software agents can produce observable behavior, but behavioral evidence does not establish that their internal interpretation matches an operator's intent. This conceptual paper models autonomous delivery as a closed-loop supervisory control problem and defines a verification horizon: the boundary beyond which available evidence cannot justify stronger claims. It proposes bounded interventions, immutable authority constraints, test-driven local stabilization, provenance-bearing ledgers, and reversible progression as a practical trust architecture.