ProvSeat: a provenance-annotated lifecycle knowledge graph benchmark for automotive seating engineering

ProvSeat is a synthetic knowledge graph and benchmark for automotive seating engineering. It links requirements, functions, parts, simulation, manufacturing and quality records through relations that depend on configuration and effectivity, and records provenance for each assertion. It also connects revision-controlled parts to an embodiment–function layer of actors, working surfaces and working surface pairs, so a change to one part can be traced through physical contact to other parts. Release 1.1.0 contains: an OWL ontology and 12 SHACL shapes; a knowledge graph of 127,714 triples in N-Triples and Turtle, with the same content as 20 CSV tables; 15 SPARQL competency questions; two benchmark task sets with complete ground truth: change-impact analysis (600 tasks) and configuration-resolved evidence retrieval (2,386 tasks); a register of 1,497 deliberately injected link and provenance defects; reference baseline results; the generator, validation and verification code. Running the generator with seed 20260921 and scale 4 reproduces the released graph byte for byte. All data are synthetic. The release contains no proprietary, personal or commercially confidential information, and its statistics do not describe any real engineering program. The files in verification/ are for checking the ground truth only and must not be used as input to a benchmarked method. The README explains this. Data and ontology: CC BY 4.0. Code: MIT. Keywords for the separate field: knowledge graph; digital thread; product lifecycle management; provenance; PROV-O; SHACL; benchmark; change-impact analysis; configuration management; automotive engineering; synthetic data.