Maturity Gaps and Structural Blind Spots: How Incomplete Abstractions Undermine Reliability Across Agentic, Supply-Chain, and Language-Theoretic Systems

Version 2 — revised in response to an external structural review and an automated critique pass. See "Response to Review" appendix in the PDF for the change log.

A recurring structural pattern appears across five recent cs.SE and cs.PL preprints: systems fail not because individual components malfunction at the task level, but because the *abstractions used to describe, compose, and monitor those components are underspecified or mismatched to the actual operational substrate*. This paper synthesises findings from monitoring agentic systems [corpus:arxiv:2606.02494], SBOM component inclusion gaps [corpus:arxiv:2606.02442], multi-agent LLM collaboration topology experiments [corpus:arxiv:2606.01490], binary decompilation evaluation [corpus:arxiv:2605.29490], and categorical semantics for untyped effectful computation [corpus:arxiv:2605.31389] and [corpus:arxiv:2605.21337] into a single **candidate reading** — explicitly a heuristic reading rather than a derivation: **abstraction boundary failures—not runtime errors—are the dominant reliability bottleneck in contemporary software-intensive systems**. Each corpus finding is read as instantiating a common structural pattern: a layer of tooling or formalism assumes a shared, stable definition of "component," "task," "behaviour," or "type" that the underlying system does not actually provide. The falsification path is concrete: if abstraction-boundary-aware instrumentation does not reduce undetected failure rates more than task-level instrumentation in controlled experiments, the thesis collapses. One weakly-connected source is flagged explicitly, and the selection process is disclosed. Sources span cs.SE and cs.PL from May–June 2026; all are preprints without peer review, and no finding should be treated as established beyond its specific experimental scope. ---

Authorship: Saluca Agentic AI Research Team (Saluca LLC). AI-drafted from arXiv preprint corpus on the date in the filename.

Cited arXiv preprints: 2605.21337, 2605.27328, 2605.27332, 2605.29490, 2605.31004, 2605.31389, 2605.31520, 2606.01490, 2606.02442, 2606.02494