Artifact of the paper: Constella: A Novel Framework for Cost-Efficient Distributed AI Inference in LEO Space Data Centers

This artifact accompanies the paper Constella. The paper studies how Low-Earth Orbit (LEO) satellite constellations can be configured for distributed AI inference under cost, energy, and communication constraints. The artifact provides the source code, scenario configurations, model-layer profiles, plotting scripts, and validation logic used to reproduce the empirical evaluation. It reproduces the evaluation scenarios, Table 1, Figures 2–5, and the main reported claims on inference success rate, deployment cost, latency, energy consumption, and execution overhead. The workflow is self-contained and does not require external datasets, pretrained model weights, GPUs, proprietary software, or administrator privileges. To reproduce the results: Create the Python environment with: ./scripts/create_env.sh Run: ./reproduce_paper_artifacts.sh And validate the generated outputs with ./scripts/validate_results.sh