MPA-CPS: labelled cyber-physical attack data for a shift- and resource-aware manufacturing line (v1.0 and v1.1)

Synthetic, fully labelled 1 Hz time series of a six-stage discrete manufacturing line (stamping, welding, assembly, testing, painting, packaging) under three classes of cyber-physical attack: network-induced delay (DDoS), sensor spoofing (freeze and replay, hiding a real slowdown) and ransomware-induced stage freeze. Processing times follow a calendar- and resource-aware Max-Plus Algebra model with three shifts. v1.0 is the exact data used in the article A Physics-Based Agentic RAG Framework for Anomaly Detection in Cyber-Physical Manufacturing Systems (Journal of Intelligent Manufacturing, 2026): 7 days, 432,000 rows. v1.1 is an extended benchmark: 14 days with weekends, a public holiday, meal breaks and maintenance; load-coupled resource occupancy; 11 attack events per scenario; sequential and fork-join topologies; three seeds (6 x 1,209,600 rows). The record contains the data, the generation and validation code (MPA-CPS_code.zip) and the reference benchmark results (MPA-CPS_benchmark_results.zip). Both releases are reproducible from the code: v1.1 byte for byte, v1.0 to floating-point round-off. See README.md and DATA_DICTIONARY.md. Code repository: https://github.com/aftab-fr/mpa-cps-dataset