Reproducibility Package for: Human-Study Dataset and Reproducibility Package for Staged Multi-Criteria Disclosure in ML Model Selection

This record provides the human-study dataset and reproducibility materials for a staged multi-criteria disclosure study in machine-learning model selection for higher-education quality assurance. The archived materials were prepared by Adil Joldić and Nina Bijedić and provide the shared empirical basis for multiple publications derived from the same study. This version corresponds to the locked analytical snapshot 20260314_003853Z and includes: 38 participants and 228 scenarios in the confirmatory layer; 90 participants and 539 scenarios in the sensitivity layer; anonymized scenario-level CSV datasets; frozen criterion weights and stimulus specifications; exported statistical reports and tabular results; a 155-column data dictionary; Python scripts for regenerating descriptive tables and H2 statistical reports; a package manifest, validation note, and SHA-256 checksum inventory. Artifact names follow the current dissertation-wide hypothesis convention: H1 denotes the technical-functional and offline verification stream; H2 denotes the empirical human study represented by this package. Accordingly, the analytical datasets, statistical reports, and reproduction scripts use H2-aligned names, including analysis_ready_h2.csv, analysis_ready_h2_sensitivity.csv, run_h2_stats.py, h2_stats_tests.csv, and h2_task_breakdown.csv. The update aligns artifact names and documentation with this convention, removes internal model-run identifiers from the public analytical CSV files, expands the data dictionary, and adds the complete H2 statistical reproduction script. The analytical values in the 155 public columns, sample sizes, statistical results, frozen policy, frozen stimuli, and analytical snapshot remain unchanged. The archive is intentionally scoped to processed study data, frozen scenario specifications, statistical outputs, and scripts required for analytical reproducibility. The live study application and the complete intelligent-agent codebase are maintained as separate software artifacts. Publications derived from the broader study include: Joldić, A., Bijedić, N., Gašpar, D., & Mabić, M. (2026). From Unaided Choice to Multi-Criteria Disclosure: A Human Study of ML Model Selection in Higher-Education Quality Assurance. ISDA 2026. Joldić, A., Bijedić, N., & Gašpar, D. (2026). Behavioral Evidence from Staged Multi-Criteria Disclosure: A Companion Human-Study Analysis of ML Model-Selection Deliberation in Educational QA. ICIICE 2026. Joldić, A., & Bijedić, N. (2026). Human-Centred Acceptance and Selective Reliance in AI-Assisted Decision Support for Higher-Education Quality Assurance. TILE-TEC 2026. Together, these materials support independent verification and reproduction of the reported tables, descriptive summaries, confirmatory and sensitivity analyses, and key inferential results associated with the frozen public snapshot.