Neurologistics: Replication Package for an Exploratory Computational Experiment on Anticipatory Ethical Knowledge in AI-Enabled Supply Chains
Description
This replication package contains the complete Python code, configuration files, automated verification tests, documentation, synthetic result tables, figures, and run reports associated with the Neurologistics exploratory computational experiment. The simulation compares alternative decision architectures under controlled synthetic conditions in three AI-enabled logistics scenarios involving tensions between: - efficiency and sustainability; - fast delivery and emissions; - cost reduction and labour rights. The evaluated architectures include an efficiency-only baseline, a fixed-rule model, the complete Neurologistics architecture, and three component-ablation variants. The package includes programmed state-updating and fixed-state conditions, Monte Carlo experiments, fixed-pressure simulations, global sensitivity analysis, uncertainty intervals, ethical–efficiency frontier analysis, boundary-region analysis, and structural-stability assessment. Some internal file names and variables retain the legacy label “learning”; throughout the manuscript and interpretation documents, these refer only to programmed computational state updating, not human, organisational, or behavioural learning. All outputs are synthetically generated from explicitly defined model assumptions and parameters. The package contains no empirical human, organisational, neurological, employee, community, emissions, or operational field data. Behaviourally informed variables are abstract computational proxies and should not be interpreted as direct measurements or reproductions of neural processes. The results describe model behaviour and structural stability within the examined design space. They do not establish external validity, causal effects, behavioural realism, predictive accuracy, or organisational effectiveness. Empirical validation would require behavioural data, stakeholder assessment, and organisational field studies. The deposited materials include: - Python source code and configuration files; - automated reproducibility and verification tests; - data and parameter dictionaries; - assumptions and empirical-calibration warnings; - full 100-seed publication outputs; - generated CSV tables and figures; - configuration and completion reports; - a structural-stability report; - a SHA-256 file manifest. Recommended software environment: Python 3.10 or later.
Files
Steps to reproduce
Download and extract the replication package. Open a terminal or command prompt and navigate to the extracted directory: Neurologistics_Mendeley_Replication_Package_v1.1 Create a Python virtual environment: python -m venv .venv Activate the virtual environment. On Windows PowerShell: .\.venv\Scripts\Activate.ps1 On macOS or Linux: source .venv/bin/activate Upgrade pip and install the required dependencies: python -m pip install --upgrade pip python -m pip install -r requirements.txt Run the automated verification tests: python -m pytest -q The deposited verification report records 14 passing tests. These tests examine reproducibility, seed independence, scenario differentiation, bounded outputs, learning-state ordering, fixed-state behaviour, architecture ablations, alternative burden functions, baseline availability, full-profile seed counts, sensitivity regeneration, sensitivity-seed uniqueness, fixed-pressure simulations, and the structural-stability workflow. Run an optional quick smoke test to confirm that the installation and code execution are working: python main.py --profile quick --output-dir quick_test_outputs The quick profile uses fewer seeds, agents, periods, and sensitivity configurations and is intended only as an execution check. Reproduce the complete publication experiment in a new directory: python main.py --profile full --output-dir reproduced_outputs Using a new output directory prevents the deposited reference results in results_publication_v1 from being overwritten. After completion, inspect the following generated directories: reproduced_outputs/ ├── tables/ ├── figures/ └── reports/ The full profile is configured to use: 100 independent Monte Carlo seeds; 100 agents and 15 periods per scenario; three logistics scenarios; six decision architectures; learning and fixed-state conditions where applicable; 72 global sensitivity configurations; six regenerated replications per sensitivity configuration; 20 independent seeds for fixed-pressure simulations. Compare the regenerated outputs with the deposited reference materials located in: results_publication_v1/ The full workflow should generate the principal architecture summaries, decision-level sample, Monte Carlo intervals, ethical–efficiency frontier summaries, fixed-pressure results, sensitivity-design and sensitivity-result tables, baseline-difference tables, structural-stability summaries, boundary-region classifications, figures, and run reports. Optional execution variants are: python main.py --profile full --skip-plots --output-dir reproduced_tables to reproduce tables and reports without figures, or: python main.py --profile quick --skip-sensitivity --output-dir quick_no_sensitivity to perform a reduced execution check without the global sensitivity analysis.