"TAKTKRONE-occ-corpus"

"TAKTKRONE-occ-corpus is a supervised instruction-tuning dataset with 5,247 dialogue samples for training metro operations control center (OCC) decision support systems. Samples cover situation summarization, disruption diagnosis, recovery planning, and after-action review across 8 severity levels. Each dialogue includes multi-turn conversations between dispatchers and AI systems, grounded in real transit networks using GTFS data. The corpus spans five major metro operators: MTA, MBTA, WMATA, BART, and TfL. Samples are annotated with task type, difficulty, network context, expected actions, and confidence scores. Training\/test split: 4,197\/1,050 samples. This dataset enables development of specialized language models for metro disruption management, offering a benchmarking resource for operational reasoning tasks in public transit. All data is synthetically generated using operationally-realistic procedures and network constraints, enabling reproducible evaluation of decision support systems."