GRAD-RL MaCySTe Maritime Dataset: Live Telemetry and Attack Scenarios from a Containerized FSRU Digital Twin

This dataset contains live telemetry collected from MaCySTe, a containerized Floating Storage and Regasification Unit (FSRU) maritime digital twin, for validation of the GRAD-RL autonomous cyber-physical defense framework. MaCySTe represents the fourth domain validation of GRAD-RL following SWaT (water treatment), BATADAL, and GRFICSv3 (chemical reactor). Telemetry was collected at 1Hz via OpenSearch (Modbus TCP registers) and NATS message broker (heading, speed). Attack scenarios implement MITRE ATT&CK for ICS techniques via UDP NMEA injection (T0856), ASTERIX radar DoS (T0814), NATS KV manipulation (T0831, T0836, T0827), and Modbus TCP flood (T0813). Key results on macyste_dataset_v2.csv:- Accuracy: 98.28%, FPR: 0.88%- T0856 Heading Injection: F1=0.6424, MTTD=1s- T0814 ASTERIX DoS: F1=0.6607, MTTD=0s (immediate)- LSTM-AE val_loss=0.000277 Architectural finding: Sensor-layer attacks (T0856, T0814) are detectable via cross-variable physics inconsistency. Control-layer attacks via NATS KV (T0831, T0836) do not propagate to sensor telemetry due to BridgeCommand physics engine architecture — documented as a testbed limitation.