HAAS Studio — Reproducibility Package: Automation Gains, Human Costs

This deposit contains all materials needed to reproduce the tables, figures, and robustness checks reported in the paper entitled: Pelechano, V., Mestre, A., Albert, M., & Gil, M. (under review). Automation Gains, Human Costs: Deskilling Pressure in Human–AI Task Allocation. Manuscript in preparation for submission to the International Journal of Human-Computer Studies. Results were generated with HAAS Studio, a simulation platform for Human–AI task allocation; the platform source code is documented in the companion preprint (arXiv:2605.02832) and is available from the corresponding author on request. Package contents: data/summary/ — aggregated benchmark outputs (means, SDs, bootstrap CIs) that directly feed every table and figure in the paper: main benchmark across 18 governance conditions and 3 domains (30 seeds), D-UCB robustness slice, pairwise contrasts with 95% bootstrap CI, longitudinal capability ledger (12 cycles), and company-profile sensitivity sweep. data/raw_per_seed/ — un-aggregated per-seed outputs for the main benchmark, healthcare detail, D-UCB robustness slice, and high-value-threshold sensitivity sweep (thresholds 0.50, 0.60, 0.70). Aggregating by condition reproduces the corresponding summary file. configs/ — exact parameter definitions used in the benchmark: governance policies, reward profiles, company profiles, affinity calibration, and domain scenario definitions (software, manufacturing, healthcare). seeds/seed_lists.json — baseline seed list (11, 17, 23) and benchmark seed list (first 30 primes ≥ 11). reproduce_figures.py — Python script (requires pandas, matplotlib, numpy ≥ standard versions) that reads data/summary/ and writes the six paper figures as PDFs. Run: python reproduce_figures.py fig_*.pdf — the six reference figures as they appear in the paper. CODEBOOK.md — variable definitions for every column in every CSV file. What this package does not include: the production source code of the HAAS Studio simulation engine. The HAAS Studio user manual is deposited separately (DOI: 10.5281/zenodo.20735034). Benchmark configuration (focal slice): G2 governance, UCB1 allocator, high_value_threshold = 0.60, 30 seeds. Contrast CIs computed by paired-sample bootstrap (10,000 resamples).