Beyond Line of Sight: A Cross-Layer Survey of Vehicle-to-Everything Communication for Intelligent Connected Vehicles

Vehicle-to-everything (V2X) communication has become an end-to-end cyber–physical evidence problem rather than a radio-interface problem alone: sensed events, transmitted messages, edge computation, trust decisions, and control or service outcomes must be evaluated as a connected chain. The nearest recent surveys clarify V2X architectures, datasets, protocols, cybersecurity, cooperative perception, and energy services, but they do not provide a study-level synthesis that systematically separates architectural claims from evaluated cross-layer effects while preserving validation context and metric compatibility. This paper addresses that gap through a systematic evidence-to-inference map, reported in accordance with PRISMA 2020, built on an analytical corpus of 235 full-text V2X/ICV research publications whose first public versions appeared from 1 January 2025 to 4 August 2026. The reference list contains 249 entries: 235 coded corpus publications and 14 supporting methodological, standards, or contextual sources outside the analytical denominator. Each corpus publication is assigned to one of ten mutually exclusive primary domains and coded across six analytical layers: sensing/perception, representation/fusion, communication, computation, trust/security, and decision/control/outcome. Claimed and evaluated layer integration are coded separately, validation maturity is assessed on an E0–E4 scale, and communication-to-outcome evidence is synthesized only when metric families, denominators, and experimental contexts are compatible. The synthesis identifies recurring fragmentation between broad cross-layer architectural claims and narrower evaluated evidence, limited high-maturity validation, incomplete communication-to-downstream-outcome linkage, and reporting heterogeneity that limits direct quantitative comparison across studies. The resulting framework produces evidence-gap maps, layer co-evaluation patterns, metric-bundle interpretations, domain-level engineering implications, and testable priorities for benchmark refinement. It provides a practical route for moving V2X evaluation from isolated layer performance toward operationally credible and auditable cross-layer evidence.