An expert-informed multi-scale synthetic benchmark dataset for calendar-aware resource-constrained project scheduling...

We provide a synthetic, expertly designed benchmark dataset for Calendar-Aware Resource-Constrained Project Scheduling Problems (CA-RCPSP). The dataset consists of 12 problem instances across four scale levels: tiny (5-8 tasks), small (25-40 tasks), medium (120-180 tasks), and large (500-1000 tasks). It includes (i) a directed acyclic task graph (DAG) with six different task types, (ii) a heterogeneous resource pool consisting of CNC machines and skilled workers, (iii) a structured work calendar with definitions of shifts, lunch breaks and statutory holidays, and (iv) two to four stochastic disruption scenarios taken from a catalogue of fourteen disruption scenarios. The parameters in the scenarios were based on industrial conditions and were validated with two structured interviews performed with a senior production planning engineer, without violating confidentiality requirements of production records. All instances are created with fixed random seeds (base seed = 42 and scenario seed = 42+i) to ensure that the instances are byte-for-byte reproducible. Baseline results for three solvers that were used with the dataset are provided, but only as descriptive baselines to ensure that the dataset format can be used with different solver paradigms: a Max-Plus algebraic scheduler, Google OR-Tools CP-SAT, and Gurobi MIP. To our knowledge, the dataset is the first openly licensed, DOI-archived benchmark to distribute the work calendar as a structured first-class data object within each instance, and it addresses a structural gap in the widely used PSPLIB library [1], which omits calendar information. The dataset and the code to generate the data are released on Zenodo [2] under CC BY 4.0 at https://doi.org/10.5281/zenodo.20553817. The open source code repository can be found here: https://github.com/aftab-fr/carm-plus-dataset.