ONLINE RESOURCE 1 - REPRODUCIBILITY PACKAGE

Article:
Completion-Shock Queues: Departure-Induced Invalidation and Endogenous Service Correlation

Journal:
Queueing Systems

Author:
Igor Kleiner
Department of Data Science, Holon Institute of Technology, Holon, Israel
Corresponding e-mail: igkleiner@gmail.com

Contents
--------
code/verify_qbd_and_simulation.py
    Exact d=2 matrix-geometric QBD calculations, an independent customer-level
    discrete-event simulation, the IID M/G/1 benchmark, and verification routines.

code/verify_heavy_traffic_symbolics.py
    SymPy checks for the direct-QBD heavy-traffic kernel identities and the
    exact symmetric-rate maximizer of the pure-dependence penalty.

code/verify_d3_saturation.py
    Checks the d=3 permanent-backlog formulas stated in the manuscript remark:
    invalidation probability, lag-1/lag-2 covariance, and vanishing lag-3
    covariance, with an independent Bernoulli-shock Monte Carlo check.

code/generate_figures.py
    Regenerates the manuscript figures as vector PDF/EPS and 600-dpi PNG previews.

Data
----
data/simulation_verification.csv
    Long-run independent simulation checks against exact QBD calculations.

data/heavy_traffic_qbd_verification.csv
    Direct-QBD near-critical checks of the heavy-traffic coefficient and
    geometric-tail scaling.

data/heavy_traffic_symbolic_output.txt
    Saved output of the symbolic identity checks.

data/d3_saturation_verification.txt
    Saved output of the d=3 saturated covariance check.

Software
--------
Python 3.11+ is recommended.
Required packages: numpy, pandas, matplotlib, sympy.

Quick checks
------------
1. python code/verify_heavy_traffic_symbolics.py
2. python code/verify_qbd_and_simulation.py
3. python code/verify_d3_saturation.py

The second command includes Monte Carlo simulation and therefore takes longer.
Random seeds are set in the verification routines where applicable. Small Monte
Carlo differences from the saved CSV values are expected.

Notes
-----
The computational checks supplement, but do not replace, the analytic proofs in
this manuscript. No external research dataset is used; the CSV files contain
numerical values generated from the model and independent verification runs.
