CAREER: Task-Oriented Graph Validation and Repair for Programmable Cyber-Physical Systems

Buildings, water-treatment facilities, and other cyber-physical systems generate growing volumes of operational data that could support energy efficiency, fault detection, grid-responsive control, and resilient operation. Yet applications that use these data remain costly to deploy and difficult to transfer because each facility describes its equipment, sensors, topology, and controls differently—and often incompletely. Knowledge graphs can organize this information using shared semantics, but they do not by themselves tell an application whether the graph contains the specific information it needs, why a data binding succeeded or failed, or how a missing requirement could be repaired. This CAREER project will establish a foundation for task-oriented graph validation and repair that makes cyber-physical applications portable, explainable, and robust to incomplete metadata. The project introduces a *semantic reference*: a deployment-independent specification of an application's information need, the evidence and derivations that constitute an admissible resolution, and the named bindings returned to the application. Three research thrusts develop this idea. First, Semantic Reference Algebra will formalize shape programs that select deployment entities, express alternatives, constraints, inference, and provenance requirements, and return application-facing bindings while supporting semantics-preserving optimization. Second, Semantic Reference Validation and Evidence Traces will shift validation from global schema conformance to *resolution sufficiency*: whether a particular deployment graph provides enough evidence for an application need to resolve. For each selected entity, validation will return bindings, a satisfaction judgment, support for successful resolutions, and structured witnesses for incomplete, ambiguous, or failed ones. Third, Witness-Guided Abductive Repair will transform failure witnesses into bounded, factorized families of possible graph augmentations, substitutions, and virtual constructs. Candidates drawn from graph context, documentation, telemetry, human expertise, and model-generated evidence will be ranked by physical admissibility and provenance, then revalidated before acceptance. Evaluation across smart-building and water-treatment applications will combine real deployments, controlled graph degradations, scalable generated graphs, developer studies, and expert review to measure binding and trace correctness, repair fidelity and safety, portability, and scalability. The project's intellectual merit is a unified formal account of application-directed binding, validation, explanation, and repair over heterogeneous knowledge graphs. It will connect knowledge representation, database optimization, graph repair, and cyber-physical systems programming while producing open-source software, benchmarks, and datasets. Its broader impacts include lowering barriers to applications that reduce energy waste and operating costs, support decarbonization, and improve infrastructure reliability. Research and education will be integrated through annual real-world data science modules for Career Technical Education students, work-based learning with building, energy, and water professionals, and practitioner tutorials and self-paced materials that broaden access to computing-enabled infrastructure careers and emerging graph-based standards.