Correctness Debt at the Execution Boundary: How Type-System Gaps, Semantic Ambiguity, Harness Lifecycle Debt, Format Divergence, and Verification Theater Jointly Define a Structural Deficit in Deployed Software Infrastructure

Abstract

This version (2026-09-26) corrects a citation error found by an automated check and confirmed by hand against the arXiv abstracts. Version 2 cited the identifier 2607.07314, an unrelated federated-learning paper posted after this synthesis was drafted, for the quantum software debugging paper that notes quantum bugs "often yield silent, incorrect outputs rather than explicit errors". That paper is arXiv:2606.07314 (QBugLM); all three citations are corrected, and its duplicate appearance in the list of papers dropped during selection is removed. The thesis is unchanged. This version has not had a full claim-by-claim audit. The full list of corrections is at the top of the PDF. Modern software infrastructure accumulates what we term correctness debt: a class of defects that are not random bugs but structural properties of the deployment pipeline itself, properties that testing and monitoring rarely catch because the failure mode is either silent, pre-runtime, or masked by a layer that was supposed to prevent it. This paper synthesizes five specific findings from recent cs.SE, cs.PL, and cs.AR preprints into a candidate reading of why correctness debt is growing faster than the tools meant to address it. The thesis is a heuristic reading, not a derivation: the five mechanisms are analogous in structure but do not share a single formal framework, and the analogy could be falsified by showing the mechanisms are causally independent in practice. The five mechanisms are: (1) type systems that cannot enforce resource-ownership invariants at compile time, leaving cost-bearing values aliasable at runtime arXiv:2606.04056; (2) undefined behavior that executes silently in production C/C++ code, generating thousands of warnings even under routine desktop tasks arXiv:2606.12064; (3) numeric format proliferation across ML accelerators that produces silent divergence without a shared bit-exact reference arXiv:2606.09686; (4) SBOM tooling that cannot agree on what a component is, leaving supply-chain blind spots even when all tools are run arXiv:2606.02442; and (5) harness lifecycle debt in LLM-agent systems, where structural defects mask task-level error signals and make conventional monitoring unreliable arXiv:2606.02494. The unifying structural pattern is verification theater: each layer nominally provides a correctness guarantee, but the guarantee is either non-enforceable, scope-limited, or defeated by the layer below it. The falsification path for the thesis is concrete: instrument a single production deployment with all five detection mechanisms simultaneously and measure whether the failure-mode categories are statistically independent, using a chi-squared test of co-occurrence with α = 0.05 as the threshold. If they cluster (reject independence), the structural-deficit framing is supported; if they are independent (fail to reject), the thesis reduces to five unrelated engineering problems. Authorship: Saluca Agentic AI Research Team (Saluca LLC). AI-drafted synthesis from an arXiv preprint corpus, originally drafted 2026-06-15, produced under the direction of Cristian Ruvalcaba, the accountable human author. Not peer-reviewed. AI disclosure. This work was produced with an agentic AI research apparatus operated by Saluca Labs. The apparatus drafted, searched and analysed under direction. Cristian Ruvalcaba is the human author and is accountable for the content. No AI system is listed as an author or contributor, because authorship entails accountability that a model cannot hold; this disclosure is the credit, and it is deliberately the whole of it.