Compounding Rigor: Continuity-Independent Assurance and Dual-Regime Discovery for Autonomous Software Development

Autonomous software-development agents increasingly work across fresh sessions, delegated lanes, model changes, and context compaction. This conceptual preprint introduces the Compounding Rigor framework for continuity-independent assurance: verified lessons become durable only when they catch known invalid classes, permit faithful work, execute at required boundaries, survive discontinuous principals, and remain governed throughout their lifecycle. The paper defines assurance-discounted coverage of recurring decision classes; separates correctness, discovery, and invocation/intent recursions; and specifies Dual-Regime Meta-Fractal Auditing, which combines compiled flat auditing with verification-gated recursive discovery. The v1.1 framework extends this model with budget-constrained, multi-resource allocation; governed queues for generative recursion; control-interaction accounting; sentinel and allocation governance; platform-agnostic adapter contracts; and two-sided auditing of both required and prohibited system properties. It presents an integrated architecture in which novel findings initiate bounded audit processes, validated patterns progressively compile into cheaper enforced checks, and irreducible frontier and sentinel functions remain subject to human authority. The paper proposes falsifiable relationships among accumulated assurance, resource expenditure, residual discovery, lifecycle burden, verification cost, and retained human control. Its contribution is a bounded formal synthesis and empirical research program. It reports no empirical results and does not claim a validated theory, general law, exhaustive prior-art result, complete determinization, or elimination of the open-world residual. The record also includes a living plain-language companion that explains the framework through accessible technical and everyday examples; it is explanatory and non-authoritative.