Managerial Strategic Override Dataset (MSOD) v1.0: Experimental Benchmark and Data-Collection Instrument

Managerial Strategic Override Dataset (MSOD) v1.0 is an open experimental benchmark and data-collection instrument for studying when human decision makers should accept or override AI recommendations in strategic decisions. MSOD contains 10 controlled strategic decision cases spanning market entry, mergers and acquisitions, product launch, AI investment, supply-chain strategy, capacity expansion, sustainability investment, pricing strategy, cybersecurity investment, and logistics strategy. Each case is crossed with a 2 × 2 × 2 factorial design manipulating: (1) candidate problem representation (low-QC candidate vs high-QC candidate), (2) human-only decision-relevant information (absent vs present), and (3) AI recommendation correctness (correct vs incorrect). This produces 80 unique experimental stimuli. The package includes the complete stimulus bank, master case bank, factorial condition matrix, participant-response template, variable codebook, analysis-variable specification, candidate representation manipulation-pretest instrument, randomisation code, derived-variable code, validation script, experiment protocol, preregistration template, ethics and privacy guidance, citation metadata, and licensing information. MSOD is designed to support research on human–AI decision making, managerial judgment, selective AI override, Human Information Advantage, strategic decision support, DPT-inspired problem representation, and organisational answerability. The accompanying reproducibility workflow validates the complete factorial design and can generate balanced experimental assignments and reproducibility outputs. A planning configuration of 600 participants × 10 cases produces 6,000 trial assignments, with 750 assignments per factorial condition. These are experimental assignments, not collected participant observations. Important: MSOD v1.0 contains no collected human-participant responses. It is an experimental stimulus dataset and research instrument. The labels “high-QC candidate” and “low-QC candidate” represent candidate DPT-inspired manipulations and require empirical pretesting before being interpreted as validated measurements of Question Compression. Future empirical releases may add appropriately de-identified participant-response data following ethics approval, informed consent, preregistration, power analysis, quality control, and privacy review. Dataset and documentation: CC BY 4.0. Code: Apache License 2.0.